Datadog, Inc. Aktie

Good day, and thank you for standing by. Welcome to the Third Quarter 2026 Datadog Earnings Conference Call. [Operator Instructions] Please be advised that today's call is being recorded.

I will now hand it over to our first speaker, A.J. Ljubich, Senior Vice President, Strategic Finance and Investor Relations. Please go ahead.

Thank you, Victor. Good morning, and thank you for joining us to review Datadog's Third Quarter 2026 Financial Results, which we announced in our press release issued this morning. Joining me on the call today are Olivier Pomel, Datadog's Co-Founder and CEO; and David Obstler, Datadog's CFO.

First, I'd like to thank Yuka Broderick for her excellent work leading our IR function over the last 5.5 years as she takes on a new role within Datadog. I'm excited to step back into the IR function after having worked with many of you through the company's 2019 IPO.

During this call, we will make forward-looking statements, including statements related to our future financial performance, our outlook for the fourth quarter and for the fiscal year of 2026 and related notes and assumptions, our product capabilities and our ability to capitalize on market opportunities. The words anticipate, believe, continue, estimate, expect, intend, will and similar expressions are intended to identify forward-looking statements or similar indications of future expectations.

These statements reflect our views as of today and are subject to a variety of risks and uncertainties that could cause actual results to differ materially. For a discussion of the material risks and other important factors that could affect our actual results, please refer to our Form 10-Q for the quarter ended June 30, 2026. Additional information will be made available in our upcoming Form 10-Q for the fiscal quarter ended September 30, 2026, and other filings with the SEC. This information is also available on the Investor Relations section of our website, along with a replay of this call.

We will discuss non-GAAP financial measures, which are reconciled to the most directly comparable GAAP financial measures in our earnings release, which are available at investors.datadoghq.com.

With that, I'd like to turn the call over to Olivier.

Thanks, A.J., and thank you all for joining us today to go over our Q3 results. Let me begin with this quarter's business drivers. I am pleased to say that we executed a strong Q3. We saw strength across our business, including across our product lines and across our diverse customer base. Revenue growth from our non-AI customers accelerated for the fifth straight quarter and was in the high 20s percent year-over-year, up about 1 percentage point from last quarter and up from about 20% in the year ago quarter.

Additionally, our AI customer cohort remains a strong growth driver. This cohort continue to grow and diversify, both in the number of customers we serve and in the scale of those customers. Overall, we continue to see healthy demand as customers from the most tech forward startups to the largest enterprises in the world are building with AI and with modern technologies, which we believe is fueling expanded adoption of the cloud and of the Datadog platform.

Regarding our Q3 financial performance and key metrics. Revenue was $1.2 billion, an increase of 35% year-over-year and above the high end of our guidance range. We ended Q3 with about 33,500 customers, up from about 32,000 a year ago. We also ended with about 4,860 customers with an ARR of $100,000 or more, up from about 4,060 a year ago. These customers generated about 91% of our ARR. And we generated free cash flow of $251 million with a free cash flow margin of 21%.

Turning to product adoption. Our platform strategy continues to resonate in the market. 61% of our customers now use 4 or more products, up from 54% a year ago. 41% of our customers use 6 or more products, up from 31% a year ago; and 16% of our customers use 10 or more products, up from 8% a year ago. So we're landing more customers and delivering value across more products through our unified platform.

One milestone we are very proud of, our infrastructure monitoring suite of products now exceeds $2 billion in ARR, growing in the mid-30s percent year-over-year. Infrastructure monitoring was the first product we brought to market more than 10 years ago. Today, our customers rely on Datadog for visibility across their environments, including cloud infrastructure, on-prem servers, containers, serverless applications and GPU fleets. And while we're thrilled to reach this milestone, we still feel that we're just getting started. According to industry analysts, we are the market share leader in global observability, but with a percentage market share still in the teens.

Moving on to R&D. As AI changes how our customers build and run software, we've been investing and executing on 2 fronts: AI for Datadog and Datadog for AI. First, AI for Datadog is our AI products and capabilities that make the Datadog platform more useful and increasingly automated for our customers. First off, our flagship agent, Bits AI, is off to a very strong start. Since we launched Bits, it has become the fastest adopted product in our history. And in October, Bits testing became generally available. With a single click, it discovers a critical user flows and applications and generates tests to cover them.

Meanwhile, our AI research team is hard at work pushing the boundaries of what Bits and our platform can do. Last month, we introduced Scout, our own reearning model specifically optimized for SRA tasks. Scout was built using our in-house distributed reinforcement learning infrastructure and now powers our investigations. This is exciting for 2 reasons. First, it performs on par or even better than Frontier models on alert investigations at a fraction of the inference cost. This is valuable as it allows Bits to investigate many more alerts, start earlier and go deeper on incidents as an always-on first responder for our customers. Second, this is also exciting as a demonstration of what we can achieve when we train models with our own production data, critical user workflows and internal tooling.

Scout is only the beginning. With the Adaptive ML team, which joined us in June, we are broadly accelerating our work on reinforcement learning and model post training. Over time, we expect our research efforts to drive a step change in capabilities across our platform and move us closer to our goal of autonomous operations.

Moving on next to Datadog for AI. These are the products that deliver end-to-end observability and security across the new AI stack. As AI moves from experimentation into production, customers today need to answer 3 key questions. How is it performing? Is it secure? But mostly ROI or is it worth of spend? And because AI doesn't run in isolation, they want these answers on the same platform as the rest of their tech stack.

On performance, we cover the whole AI stack from compute all the way to agents. GPU monitoring now goes beyond hardware reliability with new capabilities that help customers maximize fleet utilization, increase training throughput and save inference at a lower cost per request. The number of GPU devices we monitor is up by a factor of 8 so far this year. And some of our Frontier lab customers are working with us as event partners on these capabilities.

Going up the stack, Agent Observability helps teams monitor and evaluate their agents in production, including the quality of their responses, latency and token usage. The number of spans sent to Agent Observability is now up by a factor of 9 this year. We've also made Agent Observability available in the Anthropic marketplace, and we are a launch partner to OpenAI's new enterprise marketplace as well. So now customers can buy our products directly through the Frontier labs they already work with.

On the security side, our AI Guard product discovers the custom agents running across an organization and protect them at one time against prompt injection and data exfiltration, bring on the same real-time data we collect to assess them.

And on the cost side, companies everywhere are grappling with a very fast growth in token and AI spend. Beyond the 3 major clouds, AI providers are now the largest category of spend our customers monitor in Datadog Cloud Cost Management or CCM. Now with AI cost in CCM, customers can see which teams, models and agents drive their AI spend, can govern that spend with budgets, forecasts and anomaly alerts. More than 1/3 of our CCM customers now track AI costs. The number of customers doing so is up more than 11x.

Moving up the stack again. Our agent console adds to our CCM data and shows exactly which agents are actually used by whom, for what tasks at what cost and most importantly, with what outcomes.

As we continue to innovate across AI for Datadog and Datadog for AI, independent research is taking note. Forrester named Datadog a leader in its 2026 Forrester Wave for AIOps platform. And Gartner's August 2026 AI Vendor Race report called Datadog the Company to Beat for observability platform. We believe this validates our leadership in observability as well as our ability to help customers manage complexity in the edge of AI.

Let's move on to sales and marketing and look at a few of the deals our go-to-market team have closed in what has been a very strong quarter. First, we signed a 7-figure annualized extension with a leading software company, more than doubling their annual commitment. This customer is moving AI agent into production and will replace a stand-alone commercial tool with Datadog Agent Observability to monitor their costs and quality. And by using Flex Logs, they can gather and correlate larger log volumes at a lower cost. In total, they will consolidate 3 tools onto Datadog and go to 20 Datadog products, including Oncore and Bits AI.

Next, we landed a 5-year 7-figure annualized deal with the U.S. Federal Defense Agency. This customer runs thousands of cloud accounts across hundreds of mission owners. And as they've modernized, their manual tooling left them with pragmatic visibility and security blind spots. They chose Datadog as their enterprise observability and security platform and we'll start with 9 products, including infrastructure monitoring and cloud security.

I want to congratulate our public sector team on this milestone win, which was a direct result of our investment in government and the FedRAMP high certification we secured earlier in the year. And we are actively expanding our government platform to support workloads up to Impact Level 5 or IL5, the highest level of accreditation for unclassified data, which is part of our long-term investment in the U.S. public sector.

Next, we added a 7-figure annualized deal with a leading home health care software company. As this customer has grown, fragmented tools have slowed incident response and hurt customer experience. They will consolidate 4 tools into Datadog, replacing their legacy log and vendor with Flex Logs and Cloud SIEM and relying on our sensitive data scanner to help them stay HIPAA compliant. This customer will start with 9 Datadog products, including incident management and Bits AI for faster resolution.

Next, we signed a 7-figure annualized extension with a leading online entertainment company. This customer is seeing a surge in traffic with the season and is scaling effectively with Datadog. Their own AI agents already use [indiscernible] server. And today, they are also adding Bits AI. This customer will grow to 23 Datadog products, including Agent Observability and Cloud Cost Management.

Next, we landed a 7-figure analyze deals with 2 of Japan's most important banks. As they modernize their core financial systems, both banks are standardizing on Datadog for unified visibility across cloud and on-prem environments. Both customers are adopting the full Datadog platform, including all 3 pillars: digital experience, security, software delivery, service management and Bits AI as they move towards AI-driven operations.

Next, we landed a 6-figure annualized deal with a top European AI lab. These customers developers were losing valuable time to slow build and fled tests with deed releases in a highly competitive market. They chose Datadog AI Observability and test optimization, replacing a commercial tool and freeing their own engineers to focus on shipping products. Their engineers also use Bits AI to build monitors and custom analytics.

Next, we signed a 7-figure analyze expansion for an 8-figure analyze deal with a leading AI coding company. This customer is growing rapidly and renewed early to keep pace with their expanding footprint. With this expansion, they will adopt Flex Logs and our new [indiscernible] metrics to scale cost effectively with volume. And they are also one of the largest users of our MCP server. This customer will grow to 13 Datadog products.

Finally, we signed an 8-figure annualized expansion with the AI division of one of the world's largest technology companies, which nearly tripled their annual commitment. Just 2 quarters after we landed them, they are scaling Datadog across a large number of AI training workloads. They are also adopting APM and Bits AI to further reduce engineering friction.

Alongside these wins on the other side of the customer spectrum, we are leaning into the large number of new AI native companies we see create every day by investing in our Datadog for start-ups program. We are broadening its reach to a larger number of start-ups that are earlier in their life cycle and offering deeper engineering collaboration with the ones that are seeing explosive growth. This is a truly exciting time to partner with start-ups.

Now before I turn it over to David for a financial review, I'd like to say a few words on the longer-term outlook. There is no change to our overall view that digital transformation and cloud migration are long-term growth drivers for our business. And AI is another large growth driver for us today as AI workloads move into production to drive more cloud consumption and more usage of the Datadog platform, and we see that today across our business.

All 15 of the top 15 AI leaders are our customers. And more broadly, we now serve over 850 AI customers, up from about 750 last quarter. And across our broader customer base, we also see AI activity growing substantially. About 1/3 of our customers now use one or more of our AI products, including 3/4 of our $100,000-plus customers with many of these customers still just getting started with these products, using a large opportunity for future expansion. And we continue to see rapid growth in Agent activity with a number of ACP tool calls up by a factor of 50 so far this year.

We are tremendously excited about where this is going. As AI writes and runs more of our customers' software, they're shipping more code faster across a growing mix of models, agents and infrastructure. Of course, it generates complexity that is very quickly outpacing what even the best engineering teams can track on their own. And that's exactly the problem we're investing to solve.

AI for Datadog takes on more of that work for them. Datadog for AI help them run their AI with confidence and our AI research turns those data into models that will deliver more autonomy over time. Because of that, now more than ever, we feel ideally positioned to have customers of every size and every industry as well as every type of user, whether humans or AI agents, so they can transform, innovate and drive value through AI and cloud adoption.

And with that, I will turn it over to our CFO. David?

Thanks, Olivier, and good morning to everybody. Our Q3 revenue was $1.197 billion, up 35% year-over-year. We continue to see robust usage growth from existing customers as well as a strong ramp in our new customers. Revenue growth accelerated within our broad base of customers, excluding AI customers, to the high 20s percent year-over-year, up about 1 percentage point from last quarter and up from about 20% in the year ago quarter. And we saw robust growth with broad-based strength across customer segments, spending bands and industries.

And excluding our largest customer, revenue growth accelerated again. As we communicated in August, our largest customer reduced their usage in the third quarter, while their usage remained above their minimum commitment. Meanwhile, our AI customers growth continues to significantly outpace the rest of the business. This over 850 strong group continues to diversify and grow, including 39 customers spending more than $1 million annually with us.

And we continue to land larger and grow faster with new logos, contributing more to revenue growth. The portion of our year-over-year revenue growth that relates to new customers was about 30% in Q3, up from 25% in the year ago quarter.

And now regarding our retention metrics. Our trailing 12-month net revenue retention percentage was in the low 120s%, similar to last quarter, and churn remains low, with gross retention stable -- gross revenue retention stable in the mid- to high 90s. We believe this metric highlights the mission-critical nature of our platform for our customers.

And now moving on to our financial results. Billings in the quarter were $1.16 billion, up 30% year-over-year. Remaining performance obligations, or RPO, was $3.85 billion, up 38% year-over-year, and current RPO grew in the mid-30s percent year-over-year. RPO duration increased slightly year-over-year. As a reminder, we continue to believe revenue is a better indication of our business trends than billings or RPO.

Let's review some of the key income statement results. Unless otherwise noted, all metrics are non-GAAP. We have provided a reconciliation of GAAP to non-GAAP financials in our earnings release.

For Q3, our gross profit was $972 million, and the gross margin was 81.2%. This compares to a gross margin of 79.6% last quarter and 81.2% in the year ago quarter. Q3 OpEx grew 25% year-over-year versus 26% last quarter and 32% in the year ago quarter. And Q3 operating income was $334 million or 28% operating margin compared to 23% last quarter and 23% in the year ago quarter.

And now turning to the balance sheet and cash flow statements. We ended the quarter with $5.2 billion in cash, cash equivalents and marketable securities. Cash flow from operations was $290 million in the quarter. After taking into consideration capital expenditures and capitalized software, free cash flow was $251 million with a free cash flow margin of 21%. Quarterly, the timing of collections and payments vary and on a 12 -- trailing 12-month basis, free cash flow margin was 26%.

And now for our outlook for the fourth quarter and for fiscal year 2026. First, our guidance philosophy overall remains unchanged. As a reminder, we base our guidance on trends observed in recent months and apply conservatism on these growth trends. Regarding our largest customer, our Q4 guidance contemplates their revenue at their minimum commitment.

Now for the fourth quarter, we expect revenue to be in the range of $1.19 billion to $1.2 billion, which represents a 25% to 26% year-over-year growth. Excluding our largest customer, Q4 revenue growth implied in our guidance is in the low to mid-30s percent range. Non-GAAP operating income is expected to be in the range of $280 million to $290 million, which implies an operating margin of 24%. The fourth quarter includes large marketing investments, including AWS Reinvent. Our non-GAAP net income per share is expected to be in the $0.69 to $0.71 per share range based on approximately 379 million weighted average diluted shares outstanding.

For the fiscal year 2026, we expect revenue to be in the range of $4.515 billion to $4.525 billion, which represents approximately 32% year-over-year growth. And non-GAAP operating income is expected to be in the range of $1.095 billion to $1.105 billion, which implies an operating margin of 24%. Finally, non-GAAP net income per share is expected to be in the range of $2.73 to $2.75 per share based on approximately 375 million weighted average diluted shares outstanding.

Now some additional notes on our guidance. We expect net interest and other income for the fiscal year 2026 to be approximately $200 million. We expect cash taxes to be about $30 million to $40 million, and we continue to apply a 21% non-GAAP tax rate for 2026 and going forward. And finally, we expect capital expenditures and capitalized software together to be in the 3% to 4% of revenue range in fiscal year 2026.

To summarize, we are pleased with our execution in Q3, which combines strong growth with strong profitability. We continue to invest to capture our long-term opportunity in the cloud and in AI. And I want to thank Datadogs worldwide for their efforts. Now with that, we will open the call for questions.

Operator, let's begin the Q&A.

[Operator Instructions] Our first question will come from the line of Sanjit Singh from Morgan Stanley.

Congrats on the strong results on Q3. David, I want to start with the Q4 guide. If we wind back 90 days ago, has anything changed with respect to your underlying assumptions between the different cohorts, so between the core business and the sort of largest customers versus 90 days ago? Or is this outlook just sort of a reflection of that sort of minimum commitment methodology that you're applying in Q4? Just want to get a sense of has anything changed with respect to the outlook.

There are a couple of things to talk about. First of all, I think we said versus our largest customer that they renewed and they were spending above the minimum commit, and that happened in Q3. We said that in our guidance, we were going to only include the minimum commit despite the fact that in Q3, they were spending above the minimum commit. The -- as we've talked about, we've accelerated in Q3, excluding the largest customer.

So I would say the only thing that has changed was in sort of Q3, we did some renewals extensions with some large and large AI customers where as normal, we essentially extend and increase the commit, but do better pricing for that. And that happened in a number of large customers. It's a normal behavior, and that's really the only thing that changed since our last communication.

Oli, do you want to add anything?

Yes. There's a few -- we discussed on the call, I could bring up a few of those large customers where we saw renewals way earlier than we thought we would. There's another one that happened in October that we're not discussing. So these are all good news, but they do create like headwind into the short term into the revenue, and that's one change from last quarter. Other than that, like everything has played out pretty much the way we were expecting with some happy surprises increasing, in particular, the continued acceleration of the non-AI customer, which has been very good at all.

Understood. And if we think about the ex AI native customer growth, that's been on an accelerating trend for several quarters now. As we think about this core non-AI native business, you obviously have a lot of tailwinds on the AI portfolio. It seems like the $100,000 customer quarter is more of the AI product. Do you feel that, that can continue to drive sort of sustained trajectory? Or are we -- after multiple quarters of acceleration, do you sort of see a leveling off in that ex AI native cohort?

I mean, look, in the -- if you zoom out, all of these customers are adopting AI and they continuing more cloud and they're building more software and they are changing the way they work. So we expect a long period of growth for these customers. Now with the overall growth rate is, does it keep going up a lot, does it stay stable? It's a very high growth rate already, and we feel pretty good about it. It's hard to tell where it's going to go. But we're very confident in the broader trend that is underpinning that.

Next question will come from the line of Gabriela Borges from Goldman Sachs.

David, I want to stay on this point on the AI native cohort. Because I think you've been very transparent in how you talk about your largest customers versus the rest of the cohort. And my question for you is you have these funky sequentials because you have had this incredible growth rate in this cohort over the last 12 months. And now you've got this renewal dynamic with your largest customer, you're also talking about the renewal dynamics on the early renewals.

So my question for you is, what can you tell us about how to think about modeling the sequentials in this business? Is it that we get through the next, call it, 2 quarters and then we're back to steady state? Is it that maybe we see another step function down every year in the sequential because you have this early renewal dynamic, which again, is a really good place to be because you're getting really large deals that the only way you get better pricing is because the dollar volume is going up.

So maybe just the puts and takes to how we think about the sequentials for the AI native cohort and when we will be back to something that looks like normal, which I get is a tricky definition of the word normal anyway.

Yes. So good question. So in terms of the largest customer, I think as we said, the customer renewed the customer, we decided to put only the commitment, the minimum commitment. So the customer in Q3 spent substantially above the minimum commitment. So one thing that is very important in the sequentials is what happens there. We don't control that. The customer can't go below the minimum commitment, but can be various degrees above the minimum commitment. So that is one factor that we -- so essentially, one thing that will create less of a sequential guidance or actual would be that, that number stabilizes and isn't running above the minimum commitment. So that's one.

I would say -- the point here is there's one part that is very usual, which is that we have all these customers that are expanding and they're growing and they're renewing. And then when they renew, they get better deals and they also adjust their traffic over time with an underpinning of growth, which is what we've seen since the beginning of the company. There's something that is new that we have some of these companies that are scaling incredibly fast to a level that we've never seen in the industry or any industry really before, which creates new dynamics. This is why we can be a bit surprised when we see customers renew early just a few quarters into there starting to add up the product.

In general, though, we're getting a larger number of these customers like the AI native cohort is broadening. We have many more customers, many more large customers into that cohort. We called out a few wins that we made another AI lab, for example, this quarter that's in the note. So we -- the long term, this is going to broaden. This is going to represent a much larger fraction of our customer base. And as a result, there will be more stability over time.

I still don't think it's a very unstable business. I mean we're talking about [indiscernible] of a few percentage quarter-to-quarter in terms of where we end up versus where we think we're going to be. But we think at the end of the day, this is a usage-based model, and we get more visibility into the mid- to long term than we do into the very short term.

I do want to remind everybody that all things being equal, the sequential between Q3 and Q4 tends to be less because of the late December effect when our usage-based products and vacations and some code freezes. So that can be transparent based on the other factors or it might be underneath of that, but we do have that effect in Q4, which we don't have in the other quarters.

Our next question will come from the line of Raimo Lenschow from Barclays.

Can I stay on one last time -- one last question on this one. So if I think about the customers renewed early. So they get better pricing. But obviously, the volume is going kind of going crazy there, but it kind of creates a short-term headwind because the same volume is now on a lower price, but then the volume grows and from then on in a couple of quarters, we should see like the normal progression. Is that kind of the right thing to think about? I think we have that in the past, but I just wanted to clarify that. And then I have one follow-up.

Yes, that's totally what happened. There's a few other things that typically happen in those situations, which is they try to understand what is that going to mean. So they look at what they're using and they think, okay, this part, maybe we don't need as much of this part, we're probably more of or maybe in some situations, we tell them, hey, maybe you should use this different tier of retention for this type of data because of the way you're using it. So there's a bit of account optimization there when we work with them. But the -- that this is on the background of growth. These customers are growing and they're deploying much larger workloads, and they're looking at doubling, tripling, decoupling their footprint with us over time, and this is basically we're working on with them.

Okay. Perfect. And then the main question for me was on the ex AI part, like historically, we would think about that growth coming from cloud migration, people modernize and doing more stuff. You mentioned another aspect, which is kind of because of AI, et cetera, we can code more, there's more application modernization, all these stuff happening. So is this kind of changing the way it's kind of evolving for you from classically with people modernize, you have to go to Datadog. Thank you very much. Well, now actually the companies are working differently, and that's also starting to drive the ex AI business. Can you speak to that, please?

Yes. I mean -- well, the one thing I didn't mention on the call, that's always true, we also do more things for our customers. We have consolidated a lot of what they were using. We have more products to cover more of the surface area. As it comes to customers building more with AI, like it also brings some of our other products more into the front and center. So an example is customers are building a lot more software. So they're putting a lot more pressure on their CI and deployment and testing. And so they're trying also to adopt more of our CI products. I think we called out a few wins around that too in the call. So that opens the door for more products beyond core observability for us.

Our next question will come from the line of Samik Chatterjee from JPMorgan.

I had a couple. Maybe for the first one, I know you mentioned your largest customer did spend above their commit in Q3. Is there any way to tell us how much of the beat relative to your guidance in Q3 was because of the large customer spending of the commit relative to the rest of the business depending on how you forecasted them in Q3? And I have a quick follow-up.

Yes. We're not answering -- we're not being specific, but it is true. It just -- it is true that if you're basically guiding like in below there -- we said we'd be conservative. In Q3, we said conservative relative to what they were spending. And in Q4, we've been explicit and said minimum commitment. It is true that if they spend more than that, that would be implicit beat mathematically. Yes.

Again, we didn't change the assumption there over time. So we get everything unchanged. I think brought up earlier, the one difference since the last quarter is some positive trends on the overall customer growth, but also some early renewals of larger customers, which includes some other AI customers. That's the one thing you should remember as being the difference.

Yes, sure. And for my follow-up, I mean, the contribution of revenue from new customers continues to ramp up steadily. Can you just talk about sort of what is the average land size that you're seeing now with new customers and how that's changed? And what sort of behavior you're seeing on that front? How you're thinking about it as you sort of move into next year in terms of how much new customers contribute to your overall growth?

We haven't provided guidance on we said that over time, we've seen an increase of average land size due to the increase of the platform and as we give the metrics on the adoption of products, the adoption of more parts of the platform.

Oli, do you want to...

Yes. I think -- look, this is something we see in the customer base that we see that one is with the new and the AI native companies, some of them start with more scale earlier, like we've discussed it before, but we see companies scaling faster than used to. So they have -- they represent more of the revenue faster. But we also are developing new segments and landing large customers from day 1. So we called out a few on the script today. We mentioned, I think, a defense agency in the U.S. government. We mentioned Japanese banks, like 2 of the 3 largest Japanese banks in the same quarters. Those are all sizable low land deals. And if you compare the business to what it was, say, 5 years ago, like this is definitely a new side of business that we didn't necessarily have before. And by the way, those wins I mentioned, they don't even register into the revenue yet because these are land, but these are not customers that are ramp users yet.

Our next question comes from the line of Matt Hedberg from RBC.

I wanted to start with training. You noted in your script, you had an 8-figure, I think, annualized expansion with an AI division of a large software company. It sounds like they're using you guys for AI training workloads. I guess this is -- to us, it seems like a really interesting growth opportunity. Can you step back and kind of talk about what that training opportunity is? And what could it mean for your growth?

I mean it's been evolving positively. If you look a year or 2 years ago, basically nobody was doing training. It was only a few Frontier labs who were doing their own training and nobody else was. Now we see an order of magnitude more large-ish labs or labs that are doing training at a large scale. And then there's another whole category of companies below that, that are starting to do their own, including us. We started doing training at a pretty decent scale, and we showed some of the results, some of the results on the call as well. So I think there's some interesting potential development there where training becomes a part of everybody's business and everybody's customizing model in building their own. And so we are investing.

I would say it's still early to call it as a mass market. Right now, we see some interesting indications of it. We also see that we can be extremely useful to some very large customers. But I think we still have to invest. We have to build. We have to see how that market develops.

Got it. And actually, David, if I could just have one more follow-up. On your largest customer, you used the word substantially above their minimums, which I find interesting. And I know you're assuming only the minimums for 4Q, but philosophically, looking at this customer and their recent consumption patterns, does this perhaps change your opinion about how they might use Datadog longer term, maybe new use cases above and beyond what they're using today that could make that growth opportunity really interesting for this customer?

I think that's true for any customer that there are many different use cases. For that, in particular, we -- as we've said all along for many quarters, we don't control. We have the contract, the renewal, and we're not in the business of projecting that. But it is true from -- across our customer base that there are many use cases in a company, whether it be division, end user, et cetera.

Oli, do you want to...

Our next question will come from the line of Karl Keirstead from UBS.

Okay. Great. Maybe, Olivier, for you. You mentioned in your comments that a lot of customers are grappling with higher token costs, higher AI spend and are trying to get it under control. I'm just wondering, based on all your customer interactions, how enterprise spending behavior is changing in light of the fact that they're trying to get their token costs under control. And to the extent that any of the customers that early renewed to try to get price discounts were not AI natives. They were traditional enterprises. Do you think that might be because they're trying to grapple with budget pressure?

Yes. So I think, look, this is a very widespread problem. I think what we've seen is there's been an explosion of adoption of AI, in particular, coding agents. And that was not planned largely. A year ago, nobody had the amount they spend in their budget on AI. So they really had to try and figure out what that works and what part of that they actually need and what they get from that. So there's a strong need for tooling, which is something we provide. It's a great opportunity for us, and we see great growth and adoption of our Cloud Cost Management products on that or agent console that is our products that -- now in terms of the early renewals that we see, they're not directly related to that. I think they come from large customers that are either AI labs or that are focused on that are tech companies in general, I would say. And that doesn't really have anything to do with their own something has more to do with their own growth as a business.

Our next question will come from the line of Fatima Boolani from Citi.

This is Joel on for Fatima. I'll maybe just ask 2 real quick. So firstly, just on margins and operating and gross income leverage, maybe for you, David. Strong results this quarter, at least relative to our expectations, but we do know there's also certain areas that you are investing fairly heavily in. So just curious what some of the drivers of the performance were specifically this quarter? And then what your updated or what your latest investment posture is? And then just the quick second question, I'll just throw it in now is any data points on Bits AI penetration specifically in the customer base?

Do you want to go on margins first?

I'll go margins first, yes. It's very similar to what we said. We are investing aggressively in the go-to-market, we continue to have white space uncovered, both from a geographical or an end market, for instance, Fed. And we are continuing to be successful in scaling our quota capacity and you see that in the growth. In addition, we're continuing to invest in R&D. In this case, it's a combination of AI, of investments in a combination of AI and people. We see lots of opportunities. I think Oli will now go talk about the products, but we continue to see a lot of opportunity for us to get market share, get platform penetration and increase our feet on the street or on a global basis.

Yes. And on this, I think we've mentioned on the call, we see amazing adoption of it, like this is -- when you look at the number of individual customers or users that are using it, this has been the strongest launch we've ever had as a product. So we feel very good about it. We've also revamped the way we package it over the past quarter. So we used to charge specifically for investigations. Now we have a model of AI credits that cover a whole different set of use cases that involve Bits [indiscernible] and we've seen great results thanks to that. So it's super encouraging. I would say it's still early because there's still so much that we were building into this. So many things our customers want us to do for them, so many opportunities to automate their workflows. But we are -- we feel very, very good about the current status.

Let me just circle back on one thing. I think I neglected to go over gross margin, as you mentioned. I think it's very similar to what we said before that through a combination of the penetration of our platform as well as our engineering, we have been plus or minus in this 80% range. I think you see in this quarter, -- we ended up around 81%. And so this is common practice for us. We invest, optimize, do engineering projects and what was communicating gross margin indicates stability there, which we've evidenced in previous quarters.

Our next question will come from the line of Kirk Materne from Evercore ISI.

I guess, David, on -- or Oli, on this native AI cohort that's renewing a little bit early. Obviously, they're coming back because they're spending so much with you all and obviously want to get better economics. I'm just curious, what's the sort of attach rate on some of those newer renewals? When they come back and renew early, do you still have time to sort of cross-sell them some of the other products? I guess what are you seeing in that dynamic? I realize it creates some near-term headwinds from a revenue perspective, but it would also, I think, open up the opportunity to have a discussion about you expanding their usage across multiple products.

Yes. I mean, look, if you look at these cohorts, and if you compare them in particular to the native cohort we might have had like 5 or 6 years ago, like these are adopting many more products. So we see much broader product adoption and platform adoption with those customers than we saw 5, 6 years ago. So that's definitely something that is happening. Some of that is happening at renewal time, sometimes just happening organically as new needs emerge and all we spread to more teams inside those companies.

But I think it's important to clarify, these are renewals, but they are at higher levels than their previous commit, which is how they're getting the volume discounts. But we are renewing these customers at higher commits.

Okay. And David, I guess just for you, I mean, this would seem this cohort coming up for renewal earlier is all good over the long term. I'd imagine just visibility is a little bit harder from a sequential perspective. Is that sort of -- to Oli's comments on the long term, I'd imagine it's all good from a longer-term perspective, but it just creates a little more volatility for you on a sort of quarter out guidance perspective. Is that fair?

I would say it has a small amount of variation in the timing of the revenue path of these customers. The sequential here was really 3 things. It was really, one, that factor. Two, I think we mentioned the largest customer in the guide, which we're not predicting what they're going to do. And three is we know about the small amount of seasonality. So it's one of the factors. We try to get a good sense of that. But of course, as our RPO and as our contract, as we extend, it does create some volatility around the edges in the sequential nature.

But again, it's all good news. Look, we use model and when we deal with growth customers, these customers themselves don't know exactly how they're going to grow over the next few months and what part of their business and how their consumption is going to change. So we happily work with them and make sure that we support their growth and we do that track. Beyond this time we think about it in difficult terms, when we prefer for earnings call, we have to figure out, okay, so what part can we communicate, what part do we know, part do we not know. But overall, these are all great news, great customers, great growth, great adoption, amazing trends that are underpinning the industry as a whole. So we feel very good about it.

Yes. And this is similar to what we -- in the company's existence that you talked about. I mean our motion is to increase the commitment of customers, broaden the platform and extend the contract. So that's what we do all the time. It has some variability, but it's very similar to what we've been doing in the company throughout its history, and it's a good thing.

And our next question will come from the line of Peter Weed from Bernstein Research.

You talked about it a little bit, but it's really exciting to hear about the release of the Bits AI Scout and the strong early adoption. Could you help numerate that a little bit? And maybe remind us how the pricing works there? Is it kind of directly monetized? Or does it really drive the adoption or use of Bits AI overall? It would be really exciting to kind of appreciate how broad Bits AI adoption is now and how it might move the needle on overall growth and perhaps become another major pillar for the business.

Yes. So it's actually a bit of both. So look, it largely sits on the -- all of the data we acquire and the way we can process it and all the various products we have. So that's the core. But we also have a model where we charge for AI creative specifically for Bits on top of that. And the model we've chosen is a model where we stay fairly close to inference costs there so that we can pass on the savings as the models become more efficient to our customers, and so we can run many more investigations.

And the reason we've done that is the field is changing so fast, and we're improving the efficiency of those models so fast that it's not very productive to have a fixed price on you do as you expect the volume of things you do to increase by orders of magnitude while the price goes down by orders of magnitude as well. And so we on that model, which works really well. So long term, we expect both to monetize it directly as we do today and to drive broader adoption of the platform and deliver more value to our customers. In terms of what becomes the dominant monetization, it's hard to tell, and we don't really care. Again, we're fully user based. So as long as we get more usage, deliver more value, we'll be fine.

Our next question will come from the line of Koji Ikeda from Bank of America.

I wanted to ask a question on infrastructure monitoring. So you guys disclosed it's now exceeding $2 billion in ARR and growing in the mid-30s. Could you quantify how different that growth rate is from when you last disclosed the scale of this business at $1.5 billion earlier this year? How much of that growth is coming from AI infrastructure? And how sustainable is that growth rate as GPU efficiency improves?

So I mean, look, we've seen an acceleration of the business in general. And not just thanks to AI native, but also thanks to the rest of the customer base. So we mentioned that, and that applies to infrastructure monitoring also. Today, that growth is not actually related directly to GPUs. GPUs are still largely on -- we see them for our customers on training. We have products there, but we don't monetize those products widely yet. And so for us, that's an opportunity we have in front of us. That's not something that's baked into the current growth and the current numbers.

Our next question will come from the line of Patrick Colville from Scotiabank.

To me, one of the most interesting points in the prepared remarks is about Datadog training, your own Frontier models. And then Olivier, your comments around behavior is not mass market, but seeing how the market develops. I guess I want to ask you because you're at the forefront of how things are changing. How does this affect Datadog's use of proprietary tokens internally or for customers versus open weight tokens? And then I guess, David, if this shift that Olivier is describing picks up steam, what might that mean for Datadog profitability that it nicely increased this quarter, but is there a potential that this might help the bottom line as well?

So look, what we see internally and what we see with customers is -- there's a mix of models that are being used. So I mean I can say we include both customers and us into that collective we because the behaviors are fairly similar. We are big consumers of the top labs, top models, but we also consume open web models, a growing amount of open web models, and we train our own [indiscernible]. And I think it's probably representative of where the industry as a whole is going.

We do that because there are certain things you can only get from the top-of-the-line models that are being generally trained by others. And there's a number of things that are good enough and very cost efficient with open web models, you can just pick up and run. And there's finally a number of other things that you can make way more efficient with your own data, your own workflow and your own post training, which is what we've done with some of the models we announced. So I expect that mix to remain about the same.

In terms of the impact on margin, look, at the end of the day, you get some control on what you spend your money on. If it's very proprietary to you and you can deliver superior economy, you should absolutely do so. If not, like if it's very undifferentiated and applies broadly in some parts of your business that are the same to millions of other businesses, then it's fine probably to use off-the-shelf models and keep consuming those.

Our next question will come from the line of Alex Zukin from Wolfe Research.

Oli, maybe a quick one for you and David, a follow-up. Just to what extent are you seeing the demand for agent observability from both AI natives and the core driving demand for offerings like your Traces product? And what is the -- like what has changed here longer term just given the kind of elevated adoption levels? And maybe also any comments on cybersecurity and the plans there?

Yes. So we see growing demand for agent observability. And what -- so there was always interesting demand from the AI, et cetera. But what's most interesting here is the non-AI natives are consuming more and more. And because they're getting deeper into AI, we see real demand, real volumes and really exciting -- really exciting pipeline for products actually. So we think that this is crossing the chasm, so to speak, and that's exciting for this product, even though we didn't mention any hard numbers today on that.

On cybersecurity, look, there's so many new opportunities with AI. We're very happy to see that we can land the full platform with large customers. We mentioned that for a defense agency. We mentioned that also a couple of other wins when we landed not just observability, but our security platform at the same time. So we feel good about the market fit there, and we feel good about the way the whole retooling and rebuild that is taking place with AI is playing in our favor.

Perfect. And then, David, maybe for you, just a clarification. If we think about -- first of all, it's great to hear about the renewal with your largest customer. And I think we totally understand the guiding for the minimum for Q4. Any pearls of wisdom on how to think about modeling that for '27? Because presumably, that's a 1-year contract. And should -- is that something we should think about kind of maintaining at some level? Is that something that you guys are going to take out of the model? And then if I think about how you're guiding with respect to some of those other early renewals for the AI natives that get renewed, are those also at the minimum commitment level? Are those slightly above? Just any help on those 2 items, I think, would be helpful for...

Yes. Reminding everybody -- yes, sorry. Yes, on the -- we'll give guidance for 2027 on our year-end call. I think we've said in the last couple of quarters that we're guiding at whatever the contractual minimums, and we've repeated that this quarter. We'll give some more flavor on that when we do '27. And in terms of the rest of the business, what we've been -- what we do is, and we've said this over the many years, is we take the growth rate of those customers and we discount that. So that would be how we apply it across the customer base in taking the most recent organic growth and then discount it. And we would certainly be doing that for both the early renewal customers and our broader customer base.

All right. And I think that's it for our questions today. So -- thank you all for listening. I also want to thank everybody at Datadog for a great Q3. I know we are right in the middle of an even greater Q4. All very, very busy. We're meeting -- I'm going to meet a lot of our customers at the largest US conference in a few weeks. And we hope to have many more exciting product announcement to share soon. So thank you very much all, and see you next quarter.

Thank you for your participation in today's conference. This does conclude the program. You may now disconnect. Everyone, have a great day.

Everyone, welcome to the Datadog session where we are debating strong suits versus weak suits in pop culture and trivia. It's a real pleasure to have David, CFO of Datadog on stage with me. David, I really appreciate you making the time to be with us this morning. We started having a couple of really interesting conversations over dinner last night that I'd love to talk through a little bit with a broader audience.

The first one is this idea of applying proprietary data to an SLM or an LLM.

Yes.

And Datadog has a really unique data set if I think about the types of observability data that you've been collecting since the founding of the company. Tell us a little bit about what Datadog can do with that time series forecast and how you could apply it to the next product cycles in your business?

Yes. Great question. So we used to call it ML or Correlation. So for a long time, Datadog's, one of their strengths has been to be able to aggregate enough information around the signals of the functioning of client-facing applications and to be able to be somewhat predictive. And AI and models have provided a very unique opportunity. We recently made an acquisition of a company called Adaptive ML, which is a specialist in reinforcement learning around areas in IT management and observability that we cover.

And that with the data we have on observability and the functionality of applications and the research lab that we've created, we put out a model a while ago called Toto, but that's just the beginning of what we think will be a very strong competitive advantage, which is using AI and models, some of which will be open source, some of which may be the foundation companies to own the data and produce models, which are going to be predictive of issues around the functionality of applications, observability.

And we're investing behind that. We talked about that last night, both in terms of people, the GPUs, the inferences, et cetera. And that data is not public. That data is Datadog has because of our position in size and observability. And we think that will deliver a lot of value to clients and also be a competitive advantage in the evolution of the platform.

Let's stay on the advantages that it can give your clients. If I think about the holy grail in observability, there's this idea of automated site reliability engineering. And you have a little bit of that with the Bits product. So maybe just paint the vision for us when you talk to the founding team, where could a technology like Toto go over time?

Yes. The vision here is to produce more accurate and quicker real-time signals, which can improve the functionality of the platform and go on the continuum to allow for auto remediation or close to it. And that will be the speed of analyzing problems all the way towards having the platform able to act independently. That will be -- when we get there, we're not there yet, will be a combination of the evolution of the platform.

Our Bits product line is what is going to use these models and data to be predictive. And then clients in our vision will be able to make the choice of how much to auto remediate to say, for this type of issue, the platform can auto remediate. For this type, there'll be a suggestion and then someone will have to press yes. And that has tremendous ramifications, both in terms of the speed and also the efficiency and human capital in this endeavor. And that's the vision of our founder, Oli, and where we're investing behind.

The other interesting thread to pull in here is this idea of an inference economy.

Yes.

And what I mean by that is we're so early when you look at AI adoption and inference in particular, at classic enterprises, classic Datadog customers, such that if you take a step back and say, well, we're at the very early stages of an inference cycle here that could have really interesting implications for Datadog's growth rate, not just this year, but over a 3-year time frame from a structural standpoint.

So talk us through that a little bit. What are you seeing as a typical customer? When a customer goes from 0 inference to 1%, 3% of the OpEx budget going to inference, what are the implications for Datadog's opportunity at that customer?

Definitely. So Datadog is in the business of monitoring whatever can affect the functionality of an application. And the more complex it goes, there's more in it, the more opportunity to get revenue. And I think we talked about last night that early on, the first wave was essentially calling out to the model companies through APIs. So the first way to monetize, this is Datadog for AI, meaning it's just like Datadog's monitoring code, Datadog's monitoring CPUs, Datadog's monitoring databases. This is something else that we've begun to monetize through our agent observability.

And although early on, we're seeing thousands of customers use this, and we're starting to get the revenue streams from it. And what we believe is happening is if we look at some of the other metrics of building out AI-enabled applications through a combination of the outside models, open source and inference is that creates a whole another set of things to monitor that we're starting to see in our Bits products, in our MCP Server calls and other indications, a lot of activity, which, again, will increase the complexity and we believe create more importance in observability, more to monitor and more velocity and application creation, which should all benefit Datadog.

There is a sovereign thread in here around -- we've had actually a number of companies. We think about Satya Nadella talking about Frontier ecosystems. CoreWeave talking about more customers wanting to run their entire training post-training inference stack in-house. Even at Goldman, we've been talking just this morning about proprietary data at Goldman that can be leveraged with the model.

Are you starting to see -- what is -- in your customer base, are you starting to see that shift occur between, hey, we're just going to use the Frontier Lab, we're going to outbound APIs versus actually, we want more of that stack to be in-house and we're going to run more proprietary processes internally.

We're starting to see that in the metric I mentioned on usage, but also at Datadog. So we talked earlier about our models. So we early on use the outside models and the foundational models, and we put that in the platform. And similar to what you're talking about at Goldman, we own the data. We're the experts in observability. We have the ability to create from a functionality and also a cost, models that we most performant. And we're starting to invest behind it, which I think is going to result in a reallocation of the cost of our GPU -- our token costs towards our proprietary models. Early on, it takes investment, it takes lift to do that. But just like you're talking about at Goldman, we're seeing that in the metrics we're seeing in our observability and also what Datadog is doing in its own research lab.

Agree with you. Yes.

Let me ask you about what some of your more sophisticated customers are doing, including some of the Frontier Lab wins that you talk about in your AI-native cohort. There is a tendency from the outside and we'll see reported ARR numbers from Frontier Labs. And we'll extrapolate that and we'll say, well, Datadog, depending on what the contract looks like, should have a really interesting correlation with what's reported publicly. Now in reality, it's a little bit more complicated than that.

So I know you can't speak on any specific customer. Talk about that cohort in general. What are you seeing in terms of usage patterns? And how does that map back to the Datadog wallet?

Definitely. It's an important thing. We basically have invented this AI-native and non-AI-native. Fundamentally, the AI natives are cloud-native companies. What they have in common is they're investing in modern applications. They're experiencing an accelerated demand environment, and they don't have legacy infrastructure. So essentially -- and their whole business is delivering this functionality in the models. So what we're seeing is, one, we're seeing strong growth in workloads; two, we're seeing a pattern of outsourcing to Datadog.

So a number of these companies started out doing -- there's always a trade-off between do-it-yourself or using a Datadog. Now these companies have a huge R&D pipeline. And what they're increasingly realizing is it's not efficient to spend that, building your own observability use Datadog. So we're seeing a movement from open source or initial efforts or cobbling it together towards Datadog. We're seeing a use of -- because they're newer customers, a use of many of our products, the platform. So we're seeing the average use of products to be at the higher end.

But the main thing we're seeing is accelerated growth. They're very similar to other modern software companies that are experiencing demand cycle. Now most of this is in production, inference in production. What we're also seeing, which we talked about on our earnings call, which really struck us, we originally said, well, Datadog is really the production company, inference and production. We're really not going to see a lot of demand on training. But then what happened to us is some of the larger hyperscalers and other companies came to us and said, we want to use you for training or post-training workloads, which we talked about.

And that was a surprise to us. But then we thought back and we said, you know what's happening. The speed of this is such that the preproduction or late training into production is starting to merge, which created this revenue stream. We don't believe we have the evidence to say that we're going to be the training company. But we're also seeing in some of these AI-native companies, the use of Datadog in training, which has been an additional source of revenues.

The question we get, and you address this every earnings call with your comments on large customers.

Yes.

How do you think about the risk that you will have more large customers over time that will say, look, we're willing to take on the pain of running a more complex stack or more DIY stack because the cost savings make it worth it?

Yes. I think when you look -- step back and you look at our gross retention, means, are you staying on Datadog or not, it tends to be, we said, in the upper 90s with larger enterprise being at the high end of that. So we don't retain every customer, but the vast majority of customers have been making the decision to use the Datadog platform. And in fact, over time, as I mentioned, the trend has been that it is economic priority-oriented to instead of building an observability platform yourself to put your scarce R&D dollars into your product and your business. So that's been the weight of it.

But it isn't the case all the time, and it also isn't 100% or 0%. Many customers are doing a combination. And what we said all along, which is from our largest customer is that we have commitment contracts, and that customer is growing rapidly and spending above the commitment. But we know that in certain customers, there'll be sort of a trade-off between. And what we said in our last earnings call was that, that customer renewed with us, staying with Datadog, extending the contract and the usage there has moved around relative to what they're doing.

Sometimes it makes sense to hook up a database that doesn't have to be or a metric store doesn't have to be real time with Datadog and put the other work that has to be real time. So we had some volatility. And what we decided to do, given the investor focus on this is to derisk our guidance by putting in our guidance only the commitment. It can't go below the commitment. But we've always said that's a fringe case. For the most part, our business is not about outsourcing the hyperscalers observability.

We're not a concentrated company. So in a weighted and you see that in our gross retention, the vast, vast majority of our customer base, given the gross retentions we're talking about, have stayed with Datadog and the fact that the net retention has been increasing means that they've been putting more of their workloads on Datadog rather than less, less.

Let's stay on the vast majority of the customer base in that case. There are some really interesting comparisons that you and I have talked about, about the 2021, 2022, 2023 optimization cycle and how your customers have actually learned from that and you as a company have learned from that, too. So when we have investors say, look, this category, by definition, has waves to it of optimization versus growth, could we be entering a phase of optimization? Or how do you know we're not about to enter a phase of optimization? Tell us a little bit about the forecasting tools that you use and how you think this time is different.

I think things are pretty different than they were when the bubble burst. So first of all, we were growing at 70%. We're growing well, but we're not growing at 70%. There was that 0 interest rate, growth at all costs. And we're not in that market. We also had a much higher percentage of these cloud-natives than we do now. So that's a bit of a different environment. We always will have a yin and yang of growth and optimization in cloud software. It's really the weighted average of that, that matters. And you can see that in the expanding net retentions that we're net-net in a growth environment.

I would say, to your point, we also, I think, learned ourselves. First of all, we're much bigger. We're much more diversified in all ways, geographically type of customers. Our highly growing AI natives is a very important factor because those are some of the progressive customers, and that's a great sign that they're adopting us, but it's much smaller than that cloud native. And I think our customer base has not forgotten what happened and has, if you look at the net retention, acted more responsibly.

We also have built out a number of functions to provide information transparency to our customers, work with them, account managers, SKU and non-SKU types of arrangements where we help our clients use it, use the product in the right way. So I think we've gotten a lot better in helping our clients. We also have things with contract extension, volume pricing, a number of things to ameliorate the risk. So I think you're right on. Is that the end market has learned. We are much more diversified. And we've learned, I think, how to be better in helping our clients to have long-term growth with us.

One of my favorite examples of this is actually the announcement on Infinite Cardinality.

Good point.

And this could be a much longer discussion, but maybe give us the shorter version of why Infinite Cardinality addresses one of the budget pushbacks that you would have gotten from customers?

Definitely. So I'll focus on Infinite Cardinality, but then I'll also mention it's one of our portfolio of technology innovations where we have met the customer where they are, Flex Logs, Frozen Logs, Metrics Without Limits.

I have another 5 questions on that.

Yes. And it's a really good question. So basically, what we've been, I think, much smarter at is understanding how the client is using the product. So Cardinality is the sampling and the use of the data. And what we came to understand is in some examples of how a customer is using it, it's much better to, I would say, curate or figure out how all of the metrics don't flow in and get processed but they get sorted before. I'm using very simple language.

But -- and then that is good for the customer because they get more value in their metrics. That's also good for us because it doesn't help us to have metrics or even logs that flow in and are costly to us. So there are many, many examples. I think it's a very good point of how we've been evolving the platform to meet the client at the value point. And what we've been doing is proactively converting those customers that can benefit from Infinite Cardinality to our SKU, which has essentially been retentive and also margin enhancing for us given the weight on our platform of all that -- of large Cardinality of our metrics. Yes.

The other pricing announcement, it's not really a pricing announcement, but I think we could frame it as driving value to customers. Federated Logs.

Federated Logs. Yes.

This is really interesting because it actually sounds like you can bring a more heterogeneous architecture into how customers organize their logs, for example, with ClickHouse. So talk about how Federated Logs works?

Yes. I talk about Federated Logs. It's in a group of innovations where we used to have all the data has to flow into Datadog to be used, but we know that's not efficient. So Federated Logs, data observability pipelines, Bring Your Own Cloud. What this means in the case of Federated Logs is that we know that not all logs have to flow into Datadog to enhance the platform. That we can link into and logs can be stored. In this case, it was ClickHouse and Databricks.

So what we're doing is we said we understand certain -- we want to be able to have more data available in the platform without having to port all of your data into Datadog. That we've proven and we've gotten feedback from clients enhances the value of the platform, eliminating one of the objections is I have to bring all my data into Datadog even though I'm storing it elsewhere. And this is a better way to use Datadog to have what flows in to be the most important, but to leave what isn't -- doesn't need to flow in there.

And essentially, what that's doing is that's exposing our platform to more data that is not in Datadog, enhancing the value of the platform. And that -- you could say that's also what we're doing with Bring Your Own Cloud. There are many examples where it might make sense for a client either because of the volume, regulatory reasons or otherwise, to leave the logs or the information on the client side, but use the Datadog analytics. So the Federated Logs are a good example of one of a number where we're opening up the Datadog platform, and we believe that will be enhancing value to clients, improving the cost structure and opening us up to business that we would not have had otherwise.

This word that you're using on opening, I think, is important because one could argue, okay, so if I structure my logs differently, less budget goes to Datadog because now part of my observability budget on the log side is going to ClickHouse and Databricks.

Right.

Talk about the point on, well, actually, that trade-off is worth it?

Yes. We've analyzed that. We actually thought, okay, we want that the logs that need to be accessed there to be in Datadog, but we want the platform to be able to use as a central point. The more we have our users centralized on Datadog, we've proven that we get higher average revenue per customer. So to close Datadog off and force them to go to another platform is not as optimal as staying in Datadog all day long and using the data efficiently.

We actually -- getting back to how we've evolved, we actually go to clients and we say, you know, you set this wrong. You're flowing too much data into Datadog that you don't need for this purpose. And it creates a much better long-term value of customer. And also, in many cases, is margin enhancing for us because to store all that data in Datadog that is never used also is a burden on our system.

Every 2 to 3 quarters, we all have to ask you about some new competitor in this space, whether it's an AI-native or tell us how you would respond to the premise that this time is different, meaning, well, because there's an architectural shift happening because AI involves different scalability of data, therefore, the observability companies founded 5, 10, 15 years ago are not the same ones that will benefit from a true AI-native observability stack 3, 5, 10 years from now?

Yes. A couple of different questions. So I'm 8 years into Datadog, we've been public for 6-plus years. So there are many, many companies outside of observability that have tried to do this, and there's one that's been speaking about it now. It's easier said than done. So very few have been able to -- or none, I would say, have been able to organically grow an integrated platform that where which you can see everything that's going on. A lot of the attempts have been through acquisition, platforms that aren't integrated together. So essentially, there's been a lot of companies that have tried to do that they have not succeeded.

Now this is really up to us. The question is, there are point solutions that are out there. Generally, there's been a preference for single pane of glass and everything knitted together. So that's a big lift. But it is important -- very important for Datadog to not rest on that, but to basically out-innovate all of these emerging companies by modernizing our own platform. This is AI for Datadog and Datadog for AI. But in this case, I think we're essentially investing behind making sure that we are the leading innovator here so that the balance of trade between modern point solutions and the platform of Datadog is weighted in our favor. And that's why we're investing significantly.

Now I think it is a very big competitive advantage, Datadog and if Oli was sitting here, he would say the thing that investors should think about is Datadog has consistently invested 30% of its revenues in R&D. That's over $1 billion. That is out investing everybody else in the market. And why that's important is exactly your point. We have to get there first and more scalably and more knitted together in the modern platform than the point solutions that are emerging. And I think we've been pretty successful in that. But we're not resting our laurels. We have to continue doing that.

I'd like to spend a couple of minutes on a secular trend that has been core to Datadog since day 1, which is the move to cloud. So there are 2 things happening that we've picked up in parallel. 1 is you're investing more on enterprise sales. And 2 is we have a number of enterprises telling us that migrating to cloud is more of a strategic imperative today than it was 2 years ago or even 5 years ago.

Talk to us about how those 2 trends intersect. Is there a bigger -- is the share of market and observability that's going towards cloud taking another step function up right at the same time as you're doubling down on the go-to-market?

Definitely. I think that's very true. So despite the fact that cloud has been around, there are -- one, there are so much legacy infrastructure. The research companies say that somewhere upper 20s, 30% of applications are in the cloud. And there are tons of enterprises that you'd be shocked are so immature, meaning they really haven't even started their first material projects.

But the fact that any time there's been change of technology, of which AI is a major one, the impetus or the urgency of both modernizing applications and putting them in the cloud is enhanced, and we're seeing that. That's some of the reasons why we're seeing the non-AI business accelerate. At the same time, it's not going to happen by itself. Our bottoms-up selling that's very effective, has to increasingly be complemented by top-down and between that. And I think it's been really important for us over the years to build out our enterprise sales team in order to catch that wave.

And some of the good examples of that are we have a key accounts group, okay? We used to, I would say, bank on the fact that our sales cycles were going to be discrete. Our commission plans were done that way. But I think what we've learned is some of these enterprises, this is a very long sales cycle. And you can't have an enterprise salesperson covering 10 accounts. They're covering 1 or 2 accounts where the IT budget and conversion to the cloud can be in the tens of millions. And we stick this person on that account, and they're going after all the projects there. And this is the evolution. We didn't use to be like that.

So I think in many, many ways, whether it be key accounts, named in major accounts, the partners, our investment in channels, our investment in data centers. All of this is aimed at exactly what you're saying is to follow that evolution of the enterprises that are still very early in their migration and be there at the time they do their migration projects.

And when you look -- if you go back to our earnings and you look at what we're announcing and you see the industry, banking, insurance, automotive, manufacturing, airlines, you see evidence of this happening, and it would not be possible if you weren't building out this enterprise sales team aggressively on a global basis at the same time this transformation is happening.

Innovation and R&D has been a consistent theme over the last 30 minutes.

Yes.

What is -- give us a couple of examples. When you reflect on how your team operates, how you see the R&D team operating within Datadog, give us a couple of examples on how things have changed over the last year?

Yes. Well, one thing I think is very important to think about is the value of the platform. The value of the platform and doing this has meant that it has been very efficient enough for us to build functionality on top. So one thing, I think, is we -- as the platforms got -- we talked about a lot of the innovations that have happened in the platform. That's accelerated, I think, the new product and has enabled us in a very time-oriented way to get innovation. What's happening now is a very important innovation.

We are working on the balance between AI and coding tools and human capital. And I think in our history, we've invested very rapidly in human capital. And we believe we'll continue to have to do that. But we are starting to see signs of real efficiency in terms of using coding agents to improve the velocity. And it's still early. We have A/B teams where teams -- we have a team that has a lot of people and less access to coding, and we have teams that are access to more of the coding tools. And we're experimenting with what the cross -- what the A/B testing is. So I think that's a very important evolution.

Another important evolution is the research lab. When you think about what we talked about Bits first, we are investing significantly behind our own models and the proprietary nature of that, that is starting to result in the weight of the R&D budget. But I think there's still a lot of consistency in that we still have our eyes firmly on the customer, firmly on what are the product innovation they demand. And we're still very commercial people in that we have the feedback loop always putting that with the revenues and the SKUs we create. So I think that's just some of the evolution of the R&D at Datadog.

I think that's an excellent place to leave things. Please join me in thanking David for his time. David, thank you.

Thank you very much, everybody. Thank you. Thanks. Thank you.

Good evening to those of us joining on the webcast. I'll give you all a couple of minutes to settle in here, but without further ado, just a few introductions. My name is Fatima Boolani. I jointly head up our enterprise software research coverage here at Citi. And I have the distinct pleasure of hosting our keynote session this afternoon with the CEO and Founder of Datadog, Olivier Pomel, Olivier, thank you so much for taking some time to sit down with me this afternoon.

Thank you for having me. We actually were 15-minute walk away from here. So I can get right back to work afterwards.

Good. You got your steps in. We want to give you a balanced schedule today. I wanted to set the stage with you with the words Sweet 16, Datadog is 16 years old as of June, and you're approaching right around half that time as a public company, went public in 2019. So I wanted to ask you, what is the latest and greatest on where Datadog is today, and how the company has evolved not only in its tenure as a public company over the last 16 years, but most notably in the last 12 to 18 months.

So well, first of all, it's been a lot of growth. So when we took the company public, we were right in the middle of what I would say, in the early innings of the cloud migration, and that's what we were known for. That's what made the company. Now today, we're in the middle of the AI transition, which I think is a lot of the same, quite a little bit bigger. We've come a long way since we took the company public. I think we are in order of magnitude larger in terms of revenue.

We're in order of magnitude. I think we also increased the share price in order of magnitude since we went public. So all that's great. And I think that's also -- I mean, people ask me sometimes, so why are we still here? And I'm here for the next 10 years. I'm here for the -- this other order of magnitude we think is in front of us that we can get. So when you look at the last quarter, we've seen an acceleration of growth. I think earnings are -- we had our earnings a month ago.

So I might get some of the numbers slightly wrong, but we accelerated to the, I think, 36% year-over-year growth after a number of quarters of sequential acceleration. What's especially exciting to us is that this acceleration happens throughout the customer base. So it's not just a matter of AI customers growing very fast and they are. But the rest of the business, the non-AI customers, all the companies that existed beforehand and that are not primarily in the business of AI have been accelerating over the past year or 2. I think we said on the call, we accelerated since -- from 1 year ago from, I think, 18% year-over-year growth for that part of the business to the high 20s, so massive acceleration there.

And what this really shows is that the AI transformation is happening. It's touching the whole customer base for us. And just as in the context of cloud migration, observability was a key part of the transformation. We also see that observability is a key part of the AI transformation. We see it in the way this happens with the customers, but we also see it in -- when we think of the end state, so where is everything going? Where do we end up 5 or 10 years from now?

Observability is probably the only category that remains. If AI does a lot more of the work, whatever the work is, keeping tabs on the AI, understanding what is doing, why, whether it's aligned, whether we're getting the right outcomes for it, whether we're doing it at the right cost, I think is the -- probably the last software category that remains at the end. So we feel pretty excited about it and very busy building it out.

Oli, you touched a lot of hot zones that I want to take a lot of time unpacking. But before I do that, -- you've had the benefit and even the privilege of watching and experiencing not one, but now on the cusp of 2 massive technological and computing paradigm shifts. And I know myself, certainly and a lot of investors in this room tend to look at historical precedents and analogs to form mental models about what could be the state of technology, the state of the enterprise software or infrastructure software stack 3 years from now, 5 years from now.

And so with the benefit of hindsight having gone through the cloud computing transformation, I'm wondering if you can opine on as we are on this cusp of a massive generational opportunity and shift with AI, what are some of the parallels that you can draw between this cycle of innovation and paradigm shift versus the cloud computing cycle?

So it's actually a lot of the same things. So when you think of what made us successful in the cloud edge, part of it was that the roles were changing. So you had -- you needed to ship faster, you need to iterate faster. As a result, roles that used to be completely separate between development and operations became so much together. And I think in the edge of AI, we'll see even more of that. I think we can talk a little bit more about it later, but we will see quite a bit of that between not just development and operations, but also security, maybe some of the business functions.

I mean you see the way people can build software today, like nontechnical product managers can build software. So all of those rules are being pushed together, which benefits us as a business because we're in the business of bringing those people into one platform and having that one representation of the world that cuts across all of those different concerns. We -- in terms of what we can observe, we do see a lot of the same underpinnings like we see companies need to be more digital.

They need to interact with their customers digitally. They need to build up their infrastructure. They need to have applications that perform well. So I would say 80% or 90% of the AI build-out actually looks like the cloud build-out. And then there's a few new layers that are appearing. There's a very deep stack we need to be able to observe and manage for the AI transition, but we see all that. And just to backtrack for a minute, like we talk a lot about the -- about AI and all of the net new and all of the new areas we can get into.

But even if you just restrict yourself to observability and the core of what we do, it's a market where we're the leader today. We have about 13%, 14% market share. That market is going 15%, 20%, 25%, 30% depending on who you ask year-over-year. And so you do not have to have a lot of imagination to understand how we can grow 5 or 10x from that. So just the core of what we do, just extending to a stack that is going to grow even faster now, I think, makes for a great business story.

It's never been a better time to be an observability company. And if I were to bifurcate your opportunity in maybe a simplified or simplistic form of there's observability for AI and then there's AI for observability. So, let's parse through those distinct opportunities for you and Datadog. And maybe how much of the business today is coming from just monitoring these novel AI systems, agentic AI systems versus helping customers have better platform efficacy and utility in using Datadog by embedding and productionizing your own AI capabilities. How would you size or bucket those opportunities and their impact on the business today?

Yes. So I mean, thank you. This is almost our branding. Our branding is -- we talk about Datadog for AI and AI for Datadog, and that's how we present the various parts of our business. The part that is like right now growing the fastest is Datadog for AI and how we observe basically the whole stack. So a lot of the stack is actually the same as the non-AI. So most of the AI companies are built on infrastructure that other companies also use.

Most of the AI agents are actually spending the majority of their time using tools. The tools that the AI agents are using are standard applications, and those applications also need to be built and monitored. And so, a lot of it is the same. I would say today, like 70%, 80% of the stack is the same. And then there's a number of new layers. At the bottom end of the stack, GPUs are a new type of infrastructure that has been around for a long time, but was never really used at large scale in production environments or in a way that was going to be repeatable across a large number of customers.

That's something that is happening now, and that's a part of the product we're covering. And at the higher levels, now you need to monitor not just the applications, but also the agents, the models the agents are using, the outcomes the agents are meeting, the way the agents are interacting with the real world, the security of the agents. So, there's this whole new set of problems that need to be handled by observability. So we see all that.

This -- today, the part of the business growing the fastest is the part that has to do with the buildup at the infrastructure and the application layer. We also see an explosion of traffic in all of the AI surfaces, like so the agents and the number of traces we get from that and things like that. But I think it's more of a -- showing us where the world is going, rather than already today being the driving part of the business.

Bits AI has been a fairly momentous set of product introductions. That portfolio of capabilities has expanded very swiftly. I know you've had new product leadership introduced into the organization. I'm hoping you can sort of opine here on the traction that you're seeing with Bits AI as, in my opinion, the manifestation of a lot of the AI for observability use cases that you're talking about. And where is that showing up in the geography of your financial results?

Yes. So this is now -- so AI for Datadog. So how do we make everything that Datadog does even better through AI automation. And Bits AI is our AI agent. And underneath, it's a collection of different agents that perform various different things, but all branded under Bits AI. And we started -- I think we released last year the first version of it, which was focused on investigations. So Bits AI would trigger when you get an alert, would investigate it, would tell you what happened, why and would help you fix it.

And we've now expanded it to a number of different surfaces. So Bits AI has a chatbot you can use. Bits AI can optimize your code. Bits AI can actually build agents and build applications from scratch inside of Datadog context. It can do a whole bunch of different things that it was not doing before. We have a large number of customers that are using it, starting with the investigations, and it's actually pretty transformational for them. So, I was actually reading this morning some notes from 2 customers in Japan.

So, I think maybe because they didn't have the weekend off, I got e-mails from the Japan team. And so it's actually, at the same time, 2 different customers, one large retailer, one bank that are using Bits AI and that liked it so much that they went out of their way to e-mail us say, this is great. This works super well. And in both cases, what happened is that they had some issue that happened in both cases was a recurring issue, something they've had for years. And they could never quite figure it out. They could always just apply band-aids and get things back and that's it, it would happen again.

And in both cases, they actually were able to figure out those issues, thanks to Bits AI and fix them in process so that they can't recur anymore. They were so impressed that they e-mailed us straight on that. So it's pretty transformative for them. In terms of where it is on the results, you won't really see it yet on the results. We're actually in the process right now of repackaging these products. We used to have one SKU we sold for -- specifically for investigations. And now we're packaging it to a set of AI credits instead that can work across a lot of different surfaces because as I said, we expanded the product to cover a lot of different surfaces.

And so we probably will comment on that in the future. But for now, there's -- the transformation is happening right now in terms of the way we're charging those. And by the way, if you zoom out, the world as a whole doesn't quite know how to charge for agents just yet. At the low level, you have the model companies that are charging by token. And then above that, it's a little bit unclear. As far as we're concerned, we don't really care. We are completely usage-based.

We can very easily add SKUs, change SKUs, attach it to different things that we -- that scale with the activity or the data volumes our customers send us. And so the question for us is more to understand what resonates best with the customers and also what is adopted industry-wide as the right way to charge for intelligence.

I think the notion of pricing and packaging is an important one and a conversation, I think, is very important to have, I don't think there's ever been a time in the technology landscape and from a procurement standpoint, where decision-makers have been so stretched from a budgetary standpoint, but at the same time, have to solve for, again, transformational capabilities that they have to embed in the IT stack.

So with this in mind and the magnitude of innovation that the average large organization now has to productionize by way of AI, how do you expect Datadog strategy to evolve from billing on units like hosts and log volumes? And where are you and where are customers in their conversations with you as it relates to their acceptance of decoupling a lot of the revenue attribution to a unit of data.

And I know you've made some forays here. So this is not sort of an unchartered territory for you. But what does that evolution look like when the Bits AI portfolio is eventually potentially not even going to touch any data and will have outcome-based resolution. So how does that paradigm shift? And again, the philosophy around buying change? And how are you influencing that for buyers?

Yes. So I mean, look, there's many possible solutions on the table there. And I think, again, everybody is trying to figure it out at the same time, including customers in terms of what they prefer, what they like or what they don't like. For the Cloud Edge, usage pace has been the best way for us to sell and customers to buy. I think the AI edge is going to be some version of that. But you're right, I mean, initially, we price everything according to the volumes of data we were receiving.

Nothing about what comes up, nothing about what happens inside the platform because the volumes of data were enough to model basically the usage and the value we provide to our customers. In the future, so we're shipping those AI agents. We have some AI agents that even work on data, we don't ingest. So we announced the decoupling of our security agent from our Cloud SIEM, for example, so you can actually use our AI smart on other data source today. So -- and we also can generate code. We can do things that don't necessarily manipulate a lot of data. So for now, we're packaging that as AI Credits that our customers can buy on top of the rest of the data consumption.

So far, it seems to be very well accepted. But again, we'll see where this goes. If we need to repackage that in different ways, attach that to different parts such as the data volumes, the host or anything else, that's something we can easily do. Now when you look at the recent customers consolidate on us, one of the reasons is that it's a lot easier for them to have one big pool of commitment and spending with one vendor as opposed to managing 12 different pools with different companies and have to make sure they hit the right number for everything.

So, when customers contract with us, especially large customers, what they use across different parts of Datadog is fungible. So they can say, you know what, we're going to actually -- we change our mind. We don't need as many hosts, but we need more smarts or we don't need -- we're going to use more APM and less logs because it's more efficient in these different ways. Customers don't have to care, like all of that is fungible, fully exchangeable, and that's why they -- one of the reasons they consolidate with us.

I wanted to continue on this thread around and maybe zoom out a little bit and talk to you about the addressable market opportunity. Earlier in your comments, you mentioned, hey, actually, the shift to AI and agentic AI and AI systems actually has 80% to 90% of the same underpinnings that was the impetus for organizations to move to the cloud. And you've had a remarkable ascendance on the back of that secular trend.

And so if I were to just distill the addressable market conversation to pricing, which we just talked about and some of your earlier comments around, hey, there's so much more of a proliferation and diffusion of AI surfaces that now you can touch. So, from a P x Q equation standpoint, where do you see the most torque? Because in prior models of monitoring, you're sort of constrained and limited by the rate and level of penetration, let's just say, you could have because at some point, an organization is just going to decide that only 80% of my environment is worth instrumenting, or only 40% of my environment is worth instrumenting. So, how does that calculus change entirely? And what are you doing to affect the most positive change that is disproportionately beneficial for you?

Yes. So I mean, first, I would say we have so much more penetration to be had in the market than we currently have. So I mentioned earlier, even though we're the leader, we have 13%, 14% market share. If you look at the largest customers, we are in about half of the Fortune 500. And I think the average contract -- annualized contract is $400,000, something like that.

It's a little bit higher than that.

So I think we can do so much more, like we can grow so much more with all of those customers. And so there's -- again, it doesn't take AI. It doesn't take imagination to see the huge amount of growth in the 5x or even 10x we can add on top of that. Now as customers expand and do more with AI, what drives them to buy us is that they have an escalation of complexity. They do so many more things with -- they're so much more productive in general.

And what I define by productivity is the ratio between the output and the time you spend on it or the human time you spend on it, that ratio has been exploding in software engineering over the past 40, 50 years, and it's going to keep exploding even more now with AI. So with that in mind, we end up with crazy amounts of complexity. Our job is to help customers deal with that complexity. And so that's why they're going to spend at the end of the day. In the cloud, when we're fully penetrated with the customer, they spend between, let's call it, around 10%, maybe between 10% and 20% of their cloud bill on their -- on observability and security and everything else with us.

And I think some version of that will remain true in the AI world, where the spend on us is going to be this percentage of their total bill that they actually justify that pays for sales basically not only in the savings you make from the usage of the underlying compute or applications you pay for, but also in your ability to deliver the outcomes you want and the trust in the quality of the outcomes you can deliver. So we think that this is going to remain true, and this is why this is such a great opportunity in the long term.

One other observation and theme that you've been very consistent on is, hey, you've shared with the investor community the sort of bifurcation in your revenue between, let's just call them the AI natives as you do and the non-AI natives, so classic enterprise, right? But you've been pretty steadfast in your opinion and observations that actually, the behavior is the same, the problem and the pain points are the same. But can you help shed light on why there is such a massive distinction between the way the AI natives are operating versus the non-AI natives?

Yes. And we talked about them separately because we saw different growth profiles. Obviously, the AI natives are more recent companies. They all -- I mean, you've seen all the -- they're mostly private companies, some of them are not, some of them are -- we classify as AI natives because they're substantially all about AI, even though they are not new companies. But they all are going through massive build-outs of their infrastructure and their applications and massive levels of investment.

So we separate them out for that reason. In the long run, we probably won't separate them out anymore just because as the emergence of AI goes further and further into the rear mirror, like it doesn't really make sense, like now pretty much every new company is an AI native and everybody else is not. So that -- and the AI is spreading to the rest of the world as well. So the distinction doesn't make any more -- any sense anymore. So in terms of the usage dynamics, what's striking is how similar they are, as you said.

Everybody starts with infrastructure and applications and logs. Everybody has to get things working into production. Everybody has to make sure the humans actually understand what's happening at the end of the day. Some of the day-to-day usage patterns are different. So especially when you go into the Frontier Labs, for example, like you see customers that are -- or users that are manipulating thousands or tens of thousands of agents individually.

Some of that is transferable to the rest of the market, and we expect to see it happen to other customers. Some of that is not. Obviously, when you have free inference and all the incentives in the world to push your models to the max, you are not in the same world as the enterprises that get the bill at the end of the month for every single token of inference they use. But we benefit a lot as a company from serving a very large part of the -- I said, the new wave of companies, whether that's the Frontier Lab on one end, but also the slightly smaller AI natives that are still scaling very, very fast because they give us a pretty good idea of where the world is going and what we need to build for the rest of the customer base in the end. But in terms of product footprint, a lot of it is the same.

You brought up the growth in the AI native cohort. I believe you were somewhere in less than 100 ZIP code in terms of size of AI natives, that size of installed base was less than 100 about 2 years ago. It's almost 8x. So that 750 customers in the AI native cohort. I know we spend a lot of time and energy talking about some of your very, very large customers, but we'll put a pin in that and come back to that. But this is a 750 large cohort of customers. What are you seeing in terms of their behavior, posture, disposition in how they're building their stacks?

And the flip side of that coin is there's a lot of concentration of capital in those companies, right? So the movie that all of us watched during the pandemic era was you had the consumer Internet discretionary companies really skyrocket in their end user activity, and there ended up being a little bit of a feast famine type cycle where there was an optimization cycle down the pipe, right? And so when you think about this AI native cohort of customers who are the most innovative companies in the world, and they have you architected into their day 0 stack, I mean, what are some of the challenges and considerations on the other side of, hey, these businesses are basically 3 years old, and there's a lot of concentration here.

Yes. So that's a great question. And obviously, that's something that we're very careful about because we -- so we've been through the highs of COVID. We've been through the lows of COVID after that. We've seen the -- a lot of those cloud-native companies at the same time, digital native companies expand like crazy and then contract quite a bit. We were very, very exposed to it at the time. I think it was about 40% of our business at peak. And then we suffered from the compression there.

So we are very, very careful about how we assess these accounts, how we handle the -- how we watch for unhealthy behavior, for example, from customers and things like that. And we're very, very good now at getting in front of all of that in a way that we've learned over the past couple of years. The thing I would say, though, is that the exposure we have to these AI is much smaller. So our business has grown quite a bit. It's very diversified.

And we are very, very far from the exposure we had to the cloud natives at the time in terms of the exposure to the AI natives today. So we think in the equation, we see a lot more upside there than potential downside in the future. And again, we're very, very disciplined as a business in terms of how we deal with customers, customer expansion, with healthy growth, what's unhealthy growth with healthy usage and how we think this will play out in the end.

The other thing I will say is that the -- when you see all of those build-outs from the AI natives, all of that is serving the rest of the customer base. I said earlier, I started the keynote with the fact that we've seen massive acceleration of the non-AI customers. And all of that is happening on the back of those infrastructure build-outs and those capabilities that are being shipped by the AI natives.

And so we see it is not an isolated benefit to one part of the -- an isolated runaway investment from one part of the ecosystem. We see it throughout the ecosystem with companies in the other 85% of our business that are very, very careful about the cost equation and making sure that they build sustainable businesses.

Along those lines, this notion and idea of price deflation, I think you've been generally very candid about the fact that you're going to get more volumes from your customers in providing the single source of truth and eyes on glass on the health of their environments. Well, the more volumes you're going to see, there's going to be more price deflationary impacts. That's a feature. It's not a bug. And so thinking about those dynamics with the role of -- the rising role of open source with open telemetry, potentially democratizing, even commoditizing your ability to price on data, how do you think about that influencing your market strategy? And also relatedly, how you think about embedding open source into your own processes from an R&D standpoint?

Yes. I think, look, we've -- the picture hasn't changed that much in the 16 years of the company. So there's always like a large number of companies that are doing observability. So -- and there's also -- the main thing people are using or start using from day 1 is open source. And there's a rotating cast of companies there and technologies and everything else. Some of them were very popular 16 years ago, don't exist anymore or not so popular anymore. Some of them were very popular 7, 8 years ago, are past their peak today. And then there's a few more that are popular today. So that's a constant.

And for us, that's a reality that we've always been dealing with and integrating with. So when our customers come to us, they use a bunch of open source. We need to make sure we connect to it very nicely. And we also need to make sure that if they want to consolidate out of it for some part of the ecosystem, we can do that for them. And we do that routinely. If you look back at the -- or if you go and look at the transcript of our latest earnings call, we -- I think we have -- we call it like 5 or 6 customer consolidations. And all of those typically involve a number of open source products that they consolidate on us.

In terms of open telemetry, like it's amazing. Like it reduces the friction to instrument. It gets us more workloads faster. It makes customers more confident in expanding and getting both fit into our platform. And so that's great. So we support that. We invest in it, and we're glad it's there, and it's actually a tailwind for us. So we feel great about that.

If you look at our history, we've been pretty good at maintaining margins. We've been pretty good at maintaining the amount of value we can deliver to customers because, again, when you sell, there's only 2 reasons customers buy your product, they save money or they make money. And we're very good at making that case and demonstrating it. And having customers see it with the first few products they add up from us, which drives them to consolidate and send us even more in the future. So we've been very good at that.

As a business in general, I think our business model and our metrics are a differentiator. We have a very, very efficient go-to-market. We have a product that has a fairly high gross margins. And as a result, we are in a position where we can reinvest about 30% of our top line into R&D. And that's what allows us to remain relevant and build the future for our customers, but also to broaden the platform and be more of a consolidator of all of our customers' needs.

If you look at what makes us the best company 5, 10 years from now and what got us where we are today from where we were when we took the company public 6 or 7 years ago, it's that, the fact that we can keep investing and we can be the best in the long term. And our customers recognize that. That's why they partner with us because they understand that we'll be there for the long run. And they won't just get a low price for something today, but they'll have to switch that in 2 years because the vendor will not be the best vendor at that time.

Just on this notion of competitive differentiation because there's a lot of facets here that I think we can tie together that would underpin your competitive differentiation. But what I specifically want to ask you, Oli, is where do you believe Datadog's competitive differentiation has actually widened the most in the last year. So if we were to stack rank between product depth, your AI capabilities, your distribution and go-to-market, I mean, how would those filter into your ability to actually continue to widen?

Yes. Yes. So there's 2 things I would give you. So the first one is the one I just mentioned, which is the structure of the business. And by that, I mean, very lean, very efficient go-to-market, which allows us to deliver both investment and profitability, which I think is very unique. Some companies don't have to do that. If you're a private company, you don't have to do that and you can get away without doing it for a while. But in the long run, you have to do it. And that's actually fairly differentiated.

The other part of the business structure is that we serve a very wide customer base. We serve everybody from the small start-ups individuals all the way up to the largest companies in the world that are paying us tens or even more in the millions of dollars a year. And all of that is fairly unique. Most companies have to choose either one or the other, they can't be both. And that gives us a lot of pretty impressive flywheel in terms of how we can get into new customers, who we can expand and also who we can get a reading on what's coming up next in the market, thanks to the long list of smaller companies and individuals and start-ups that use us. So that's the first one.

The second one, and I think that's the one that's becoming more and more important. is the fact that we're a SaaS vendor with access to a lot of extremely clean, operationally relevant data that we can use to then train models to automate all of the operations, security and all of the other problems we deal with for our customers. When we started the company, it was always the intent. It was, hey, we're going to be -- to build a company that's going to be SaaS, so we can do that.

When you start building kind of companies, you get all sorts of offers from customers that say, "Hey, we're a big bank, we use you, but we'd like an on-premise instance that we can manage ourselves. We've always said no. And the reason was always no, no. We actually want to have the data. We actually want to be able to iterate over it. And also, by the way, when we're going to build more functionality, we'll be able to do that so much faster if we have direct feedback loops from what we see customers using and not using and what we see work and not work in their environment. So we've done that.

Today, I think, is when this promise comes to life because we're at a time where, of course, we see the Frontier Labs build incredible models, but there's different kinds of models we can build. We can build those models ourselves. We can tailor them to our needs. We can build those models based on very specific types of data that are relevant to observability. So think of it not so much in terms of words as you see in the LLMs, but more in terms of traces and metrics and logs and network topology and all of those things that take a slightly different shape that we can ingest into our systems and that we can turn into predictions.

So we've shown some version of that with our open source time series models. So we've released 2 models of Toto and Toto 2.0, which we released a few months ago, both of which were state-of-the-art at the time of release. And these are fairly small models, like these are not gigantic models that consume gigantic server farms. But the quality of those models is completely based on the quality of the data we have.

And we also announced a few months ago, the acquisition of Adaptive ML, which is a company that was building reinforcement learning to tune and build custom models. And that team has actually joined our research team to accelerate the development of next-generation models for us. So, you should expect to hear and see more from us in the future on that topic.

This is a good segue into the next question I wanted to ask you. Still sticking to the competitive lens here. But I think what a lot of the investor community and myself included, have been positively surprised about, Oli, is this idea of the DIY or the in-sourcing kind of debate, right? You have, as customers, some of the most innovative companies in the world. They are doing very avant-garde things in changing the world at the jagged edge.

And so what's been very interesting to see is that they've knocked on your door to solve some of their most pernicious challenges, right? So can you talk about why in terms of their decision and decisioning and their reasoning around, hey, we want to align with Datadog to solve these challenges where these are actually the most financially and intellectually, most well, financially and intellectually resourced companies in the world, right? So how do you bury that debate around, at some point, some of these very large companies may in-source? They may do it themselves.

Yes. So a few things. So one is every single engineer thinks that they could and should be building observability. And I know because I've done it, that's...

Spoken like a true engineer.

Yes. I mean that's -- when I was confronted with that problem, I decided to go build a company around is Alexis Le-Quoc. So I can't blame people for thinking that. And so the -- our business is to make sure that we -- people understand how much more value they're going to get by not having to do that themselves, so they can focus on other problems. And it turns out that for most businesses, I mean, pretty much every single business except us, there's some other problem they should solve and observability and automation is in service of that, and they should focus their energy on that. That's number one.

Number two, it's not economical for almost any company to do so. And we know that because, look, a lot of the people we have at Datadog who run our operations, who run some of our products actually used to run observability at very large companies internally at some of the -- the top 5 hyperscalers and they know exactly what it costs in these companies to do that at scale. And actually, it turns out it's not any different from what it cost to completely leave it up to us.

And we know from all of those people who used to work in those places that we offer a much better value, much better experience, and in the end, better outcomes that they could get at their previous employers. So that's number two. Number three, like when you -- when people reach a large scale, there's often folks who want to build it themselves. And in some cases, they will. We've had some customers that say, we're going to build it ourselves. And a number of them have churned off, and some of them have come back.

We've seen that. We called some of those out on our calls. It's a very small fraction of our customer base. If you look at our aggregate numbers, our gross retention is extremely high. It's in the high 90s, including the whole business from SMB all the way to enterprise, which is incredibly high. There's basically no upside there, like you can't imagine of anything that's substantially better in terms of retention.

And what we've seen on a fairly recently, I think we called that out a couple of earnings calls ago. We've had actually a number of the largest hyperscalers come to us. So these are companies that were completely homegrown, that had built their own systems that don't run any software from other providers that came back to us and said, "You know what? For our AI build-out, we need your help.

And the reason they did that is that they're in a situation where they're all competing extremely hard, and it focuses the mind. They are not under the assumption anymore that, yes, they have enough resources to do everything themselves. And it's okay if they dilute their impact, or if they have -- they some percentage of their smart people and some percentage of their compute and that problem and that problem, and that other problem, that are not their main problems.

Now, what we are hearing from them is their teams need to ship as quickly as possible, and they want the best so they can do that and they don't want to waste their time, whether they're on the user side because they have to use inferior solutions that are built in-house, or on the higher management side, wasting resources to work on problems that are not core.

This brings up an interesting other angle that I wanted to have you opine on, Oli. This idea of dogfooding, pun intended, I myself would say drinking your own champagne, but very consistently, Datadog has kept a high R&D envelope, about 30% of revenue, as far as I can remember, notionally about $1 billion plus per annum, just going back into the business, pushing the envelope forward on the technology road map.

Can you talk about your own usage of AI internally, agents how that's improving your own shipping velocity? And then relatedly, you talked about Adaptive ML earlier, and I think you were sort of flirting with the idea of the benefits of having small and contextually rich training models and running inference off of those versus the large language models. So, how does all of that factor into your R&D strategy and how you allocate R&D capital from here?

Yes. So I mean, look, obviously, we use a ton of AI everywhere in the company. In the -- I would say, in the non-engineering side of things, we're not any different from many of our customers. There are so many things we can automate, and we're well on our way to doing that. So, for example, -- we used to scale the number of people we had to react to support tickets and things like that, along with the size of the customer base. Now that curve has gone the other way.

And instead of having all these reactive work, now we can allocate people to do proactive work with customers. We can build Forward Deployed Engineering to work with customers, all sorts of things that we didn't have before. And all of that -- all of those doors are opened by the AI automation. On the engineering side of things, we're transitioning the whole organization to build with agents. And we're obviously dogfooding for that. And so we're finding, again, what most of our customers are finding too, which is that, yes, you can accelerate the development quite a bit.

But the hurdles after that remain. Okay, now we need to ship that into production. Now we need to make sure it actually works. Now we need to make sure that actually delivers the right outcomes for our customers. And by the way, on the way to doing that, we're seeing our compute bill or token bill explode, and we have no idea how much of that we need, don't need, et cetera, et cetera. So we're heavily dogfooding all of that.

So we're building the right products for our customers on that. And there's a lot of opportunity in each of those areas, whether that's shipping to production, keeping it running on production, measuring the value with the end users, are we actually shipping the right things, understanding what to work on, or optimizing the costs, token costs of writing code and automating.

In terms of how we allocate the spend, the mental model you should have is that the overall envelope doesn't change all that much. So we still -- we invest 30% in R&D., but the makeup of those 30% might change. And so, we're obviously spending more and more on tokens. These engineers are using a lot more agents and doing a lot more there. And we also started spending a lot more on GPUs. I mentioned earlier that we are training models. The first of those models were pretty small, but now we're starting to scale them, and you should see more spend there also that's related to that.

And on the labor side, and this is going to end up becoming a little bit of a philosophical debate, the role -- the classically defined role of the software engineer now fusing with the developer with the site reliability engineer, the cloud infrastructure engineer. So, the average engineer persona or the technical persona inside Datadog, are they now just wearing a lot more hats simply because there's more scope for one individual. And maybe you can put to rest the debate around, hey, does the labor productivity of the average technical person in your organization surmount your need to hire more heads? And how do you balance that equation for yourself?

Yes. So I mean, look, we definitely see that people can wear many hats. And we also see that you can empower much smaller teams to build with AI. So where you needed to build a team of 8 with a lot of different roles, now maybe you can do with a team of 3. And so that completely rewrites the equation in terms of how you structure the teams, how you scale them, all these sort of things. So we definitely see that.

I don't think we know exactly what the end state is because we're still working on the process for that. The models are still evolving. So I think we will take maybe a few years before the dust settles on that. But yes, in terms of the -- sorry, some other part of your question was on...

The human labor component, right? You said the envelope is going to shift, right? Token costs are going up. But...

On that side, no, we're definitely still scaling the engineering teams. So when you think of our business, we're limited by 2 things. We're limited by how much of the right products we have to sell for our customers and then how much -- how wide our distribution is, how much sales capacity are we deploying across the world in front of the right customers. And so we need to scale both.

And so, right now, if we get more productive, if we produce more with every dollar invested in R&D, we will just do more R&D, and we'll produce more products, and we'll go deeper with our products, and we'll expand into neighboring categories, and we'll help our customers consolidate more. We see no end in terms of the demand on that side from our customers. We're more limited internally by what we can produce.

What are some of these neighboring categories that you have aspirations to have a stronger foothold in?

Well, I mean, look, we've mentioned security before. There's quite a bit we need to do also in terms of going up to the -- like further into the business side of things. So we started with infrastructure monitoring, then we went up to application, then went up to end-user monitoring on top of that and what are they actually doing with it. From the end user, we're getting into the business value.

Are these users creating value? What are they doing? Are they buying more? Are they spending -- are they staying longer? And I think we can get further on to the business side from that. We also announced at our conference, a data agent that you can use to query your data and to actually get -- we run all of our internal data analytics on that at Datadog now. So, I think there's plenty of opportunities for us there, too.

On the cybersecurity side, because you put my antennas up on that, just talking about the product portfolio and strategy here -- it's about 1/4 of the base penetrated with security SKUs. It's a $100 million ARR franchise. So you've made a ton of progress, but I think there's still a lot of opportunity. And I think by your own admission, there is still work to do there, right? So, relative to your initial expectations, how is this unfolding? What has gone well? And where do you see opportunities and scope for improvement?

Yes. So, if I look back at the opportunity, so first of all, we think the opportunity is huge. if you look at the fundamentals, if you look at what I mentioned earlier in terms of the roles getting smushed together, like security is a big part of that. And so, it's extremely clear to us that selling security to security people is not going to be a thing. Like it's -- everybody is going to have to take the responsibility, and that's just the way companies are going to be building products in the future.

And so the DevSecOps is going to -- is not going to be an emerging trend anymore, but it's going to be the way most companies are running things. So, that's number one. Number two is -- and that one is even newer. -- we've seen over the past 3 months that security tooling has been completely upended. We've seen that the models are so good and they move so fast that pretty much anything that was built more than a few years ago in security is obsolete.

And the very notion that you're going to have 12, 15, 25 different security products installed that are all contributing to a central repository that then humans review and prioritize that's just not going to work anymore. I mean, now you need everything to be fully integrated, you need to respond at machine speed. You'll have agents constantly probing everything that can be probed on your end.

And so, we think it completely flattens the space and negates any advantage there was to be an incumbent in security. So, we think it's a tremendous opportunity, and all of that is going to be developing over the next couple of years. So, we feel very good about investing and building more in that department.

And how does the playbook from a go-to-market messaging standpoint change when you are going to be interfacing with a completely different new class of competitors who, by the way, are, in some cases, also meandering into your core territory, right? Some of the largest cybersecurity companies are more assertively talking about observability as part of their mandate, right?

So, how do you interface with that changing dynamic? And frankly, other very large tech companies like a ServiceNow, for instance, who have actually also been acquiring their way into cybersecurity? And how do you interface, or how does the messaging playbook change?

Yes. Look, the thing that's most important to us is to deliver true platform experience. So where we are different from everybody else is that -- so we target maximum usage by the maximum number of users, so the developers and the operations people and folks that -- and that's typically not something that you find in security products. Security products have a very small user bases. And two, we offer a fully integrated platform.

So, whether we build or we acquire, we always through platform, and we have a fully integrated platforms. Most of the companies that have scaled in security tend to be asset allocators and consolidators. So they are going to buy a number of companies, and they're going to integrate for the sales channel. What we think is that when you need to get a large number of users using you every day, and also when you need to act at machine speed across different streams of data, you need an integrated platform.

You cannot have siloed products that just happen to be bundled together by the salesperson. So, our advantage, what makes us special is that we are this integrated platform. That's the DNA of the company. And at the same time, we have a business model that also lets us be acquisitive and accelerate the growth of the company.

You said you're essentially constrained by your -- well, I'm paraphrasing, you're constrained by your imagination in terms of your output because your velocity in R&D is just infinitely higher now. Two questions around that. What do you expect to be Datadog's next $1 billion product franchise, number one. Number two, because your ability to build and innovate is orders of magnitude faster than 3 to 5 years ago, how does that alter or modify, the way you think about your build-versus-buy decision? You've been very disciplined. It's very tuck-in oriented. So, kind of 2 different sides of the same coin as it relates to R&D.

So in terms of the next $1 billion product, I mean, look, there's a number of candidates internally. The one thing I will tell you is that when we brought together the various trends of observability together, we called ourselves 3 pillars. We said metrics, traces, and logs, like infrastructure applications and logs, like however that these are the 3 pillars. I think now we consider that we have 4 pillars. We're breaking out the digital experience into the fourth pillar, which is Real User Monitoring and Synthetic Testing, and it's growing fast, actually accelerating growth over time as it gets bigger.

And it's a critical part of all of our customer stacks, whether that's the non-AI companies or even the top AI labs. This is something that is extremely valuable and absolutely everybody needs, perhaps even more so in the world of agentic AI. And so, we see that as a great growing part of the business. But yes, there's many parts of the business that could be $1 billion business, and we want them all to be in the end. Whenever we start with a new product, we definitely want those products to be $100 million-plus products, and we have hopes that they can cross the $1 billion mark. On...

The build versus buy.

Yes. On the build-versus-buy, look, we're very -- there's no category where we think, oh, we need to buy there. Like we always are very opportunistic in that we look at absolutely everything that we might want to do. We're ready to build everything, but we look at the opportunities we'll have from accelerating from some of the assets and the companies we see out there.

Historically, it's mostly been about teams and product advantage. So we have -- we're going to take a shortcut of maybe 2, 3 years by having a team that's been doing that for 2, 3 years and has learned the market and we can get to product-market fit a lot quicker by acquiring them. So that's most of what we've done. We might also do the same thing for distribution. If we think that a specific company will give us differentiated distribution in some -- with some buyers, some part of the market, we might do that, too.

But the underlying principle is always we're shipping an integrated platform; we're bringing the users together into that platform. We'll bring the use cases together, and that's what makes us different in the long run.

On the continuum of size, just to reiterate, you have been doing tuck-ins. The checks have been smaller, but you do have very robust share price currency. So, in thinking about where the bar is to do something bigger and bolder, naturally, the rubric is going to be different. Wondering, Oli, if you can help share what the parameters of a potentially larger-than-historical Cadence transaction look like?

Look, there's no hard-and-fast rule. I think by definition, the larger deals are fewer and far between. You -- the chance that you find exactly the right assets with the right economics, and the right alignment, and the right everything are very, very low for larger deals, but nothing is out of the question. I think the point here is as long as we believe that it can accelerate the path to someplace we want to go and that we can deliver on a unified platform, that's -- everything is on the table.

Oli, I wanted to end our conversation and have you bring out your crystal ball. If you materially outperform financial expectations, investor expectations in the next 3 years, and there's a lot of goodness that we talked about, right? So barring execution, what will investors have materially underestimated rather in your abilities or in the market environment at large?

Well, I think the opportunity as a whole, like it's pretty clear to me that observability is a major part of any transformation story and the AI story in particular. It's also pretty clear that observability is the last frontier. That's what remains, like keeping tabs on AI, keeping tabs on the machines, keeping tabs on the agents, whatever the job is, is going to be absolutely key in the long run. So, that's a huge opportunity, like a gigantic opportunity.

I think it's also pretty clear that the question is not just -- it's not who's going to -- whether we're going to be there. The question is who's going to be there with us in the end? Like it's a market, great market, great opportunity. It's going to be us, maybe 1 or 2 other companies. So with that in mind, I think it would be foolish not to invest in Datadog. That's my take.

I think it's a good place to end the discussion. Thank you so much, Oli, for your thoughts and insights. This was a fascinating discussion. I appreciate it.

Thank you very much.

Thank you.

Hi, everyone. Thanks for joining this -- one of the last sessions at this conference, probably the last thing between you and Smith & Wollensky, so we'll try to make it interesting. I'm Kingsley Crane, technology analyst here at Canaccord Genuity. We're really excited to have Datadog with us here today. David Obstler is CFO. David, thanks so much.

Thanks for having us. I hope it's a meaty discussion before Smith & Wollensky.

So let's kick it off. For those of you that aren't familiar, let's start with the quarter. 36% growth, $1.1 billion of scale, accelerating from 32%. You've accelerated the past 5 quarters. Just what were your takeaways from the quarter? And is there any simple way to describe what's going so well right now for the company?

Yes. It's really a combination of our investment in the platform, which has expanded the product line. That's enabled us to cross-sell and also take market share. We're seeing strength across all the way from SMB to enterprise and globally. And in order to distribute that, we've also successfully expanded our go-to-market. And then the environment is pretty good where we have the AI natives, but in the overall customer base, there's a strong investment in platform right now partially to take advantage of what's coming with AI. And that any time we have a replatforming and a modernization of tech stack, that's complemented Datadog and their growth. So all of that came together, and it's been compounding over 4 or 5 quarters, which produced the results in the quarter.

So there's this idea that because you've been able to monetize AI natives so well and it has worked well for you, that growth is top heavy or concentrated. But in reality, that's not all that true. You've seen non-AI natives accelerate for multiple quarters from high teens to high 20s. What is going so well in that segment for those incumbents? And is that driven by AI adoption?

Yes. So that's -- so our -- this has been 4 or 5 quarters, as you mentioned, of acceleration up into the upper 20s, and that's enterprises and stuff. That's the adoption of the Datadog platform. That's winning market share. For instance -- in the last quarter, I think we said sequentially, we grew $115 million of revenue. So if you just want to, on a quarterly basis, think of that as over $400 million of business and you look at that versus the competitors, you'll see that the market share gains are very substantial. And it's all because of the adoption of the platform, the fact that our end market wants to look at single pane of glass and in real-time observability and security, it is a real premium on having it all knitted together.

And that's really resulted in pretty strong adoption in a number of different areas. It's any time technology has changed, et cetera, we've also had all the modern workloads tend to go to Datadog -- for observations. So there's probably also going on more of a weight towards modern workloads, some of which are AI-enabled. It's been very broad-based. And as you said, very strong. The AI natives themselves, where we've also won a lot of market share, have complemented that. And in that, I think we said it's pretty diverse. We had over 750 names, over 30 of them having $1 million, 10 of the top 10. So that means that, that group of cohorts is adopting Datadog on top of the overall environment and the overall business accelerating.

So we've had this stat for a number of years of how to judge breadth of product adoption. And we've moved from 4 to 6 and one of the premier stats is 10, and that's doubled in the past year. What happens when a customer moves from 4 to 10, like, kind of, give us a sense of that adoption timeline and then when a customer is consuming or buying 10 products, like, what does that look like as a percent of their IT spend?

Yes. Well, in terms of the land and expand, so our clients, when they land with us, they have other vendors that have been there for a while, right? So we don't tend to have everything switch all at once, but what's been happening, this has been going on for 5 years. The weight has been to when those other contracts come up for renewal, to use more Datadog products.

We sell on a credit basis, we sell 2 million of capacity, and they can use the platform. So we find when we get that is we find that adoption. Our cohorts are very, very long, meaning our cohorts signed 5 years ago are still expanding. The reason they can with us is they consolidate on Datadog. And part of it is that we've expanded the product line so much. So that's, kind of, what happens on an ongoing basis. That produces the net retention that we talked about in the 120s, low 120s. And that's been pretty persistent over a long period of time. That's something very important to look at. Like you said, each year, we expand the definition of cross-sell to the number of products, and then we tend to fill it up pretty fast.

So you were early in winning AI natives versus many of your peers, and that's grown wonderfully for you. Some investors weren't really sure what to do with that category, they should underwrite it to 0. We've been particularly excited about the category, and it's grown well. But how about for you? I mean, like, what gets you excited about AI natives? And do you think that, that, kind of, helps prove that you're just getting to where the puck is going in terms of those product.

Yes. I mean, it definitely does. It's AI in AI natives, and we can talk about that momentum in AI natives. So we've always been very successful in what we used to call cloud natives. Now they're AI natives. This group is a much smaller group as a percentage of ARR than when COVID happened, but it is growing very fast, and it's the who's who. And it basically says that if you're essentially investing in modern technology, those companies do not have legacy systems. They've been invented over the last few years.

There's no displacement of something else. It isn't there. And so they're adopting Datadog early on. And I think it's a very strong endorsement of where the puck is going, as you said, because this is being pervasive, whether you're talking about the model providers, the database providers, the verticals, the GPU providers, they're going to be tool companies, infrastructure for the overall digital economy, and they're choosing Datadog for monitoring. So it's a really good forward-looking sign. In terms of a question that's come up, is there going to be volatility there? Yes, there might be -- there might be -- I mean, there's volatility in all these markets. But what is most important is you're compounding where technology is going over 5, 10, 20 years, which we've been doing successfully.

So I want to build on that. And then there's been -- before AI, the largest technology organization would always buy, procure, and build technology very, very differently. And so the fact that you've been able to win at hyperscaler labs for some large AI natives is a big testament that your product is that useful if they want to use it rather than build it themselves. That being said, talk about, like, maybe usage trends among largest customers, like, if you get to that scale and then just even how that conversation goes when you get to 8 figures or 9 figures of spend.

Yes. I mean, we've always been with larger customers. It's always been not one or the other, they do a portfolio of things, and they tend to put their more modern workloads and mission-critical workloads being observed by Datadog. So, there's always going to be the back and forth. The net weight has been towards not doing it yourself but buying Datadog. That's what's produced from 0 Datadog to over $4.5 billion. That's produced Datadog from not being in the industry to being the largest player.

So the weight of this has been that way. That doesn't mean every client is going to do it. But with very traditional enterprises, we have a very long set of data on the net retention, i.e., the expansion of that. That's a very, very powerful motion. So anything from car companies to banks, insurance companies, metal benders, video, all of the media, all those companies are essentially weighted average expanding with Datadog if they're using Datadog. They may do some things themselves, but they're putting more and more of their workloads into the cloud and more and more of that is being observed by Datadog.

So as you point out, one of the strongest elements of Datadog is that consumption can increase that quickly and customers often consume ahead of commit. And so how do you manage those customer conversations in that regard? And then also how do you guide and manage the street in that way?

No question. So we have take-or-pay contracts, and they tend to be a year or more. They tend to be out to 3 years. And then we have a base of commitment, so they can't spend less than that. And then for the most part, this is really about the same thing with AWS and the hyperscalers. There's essentially capacity planning we do together. And then for the most part, you can tell with the net retention that they used more than they've committed to. And so the motion is we then work with them to figure out what their next consumption is going to be.

And they have an incentive to do it because we're volume-based pricing and term pricing. So, most of the time, we have within the contract, they have a set amount, but it's in their advantage to expand that. And so what we do is we have long histories by customer of what happens. We know because every day, we see the usage. I can see the usage when I wake up in the morning. I can slice and dice it by customer. So it's almost perfect information. And then we work with that. We've gotten pretty good at helping clients use it. There was -- I would say, in the bubble after COVID, there was -- in the ZIRP side, there was probably some usage that was over usage.

But we step in and we help clients then use it effectively. We help them figure out what logs to put in and what logs not. And then in terms of guidance, what we're able to do, and that's what how we can beat and raise is we take that growth rate that we see over a long period of time and we discount it so that we have that cushion. And then what's been proven out in the beat and raise is that the clients spend more than the amount we're discounting. So, let's say net retention is blank. We discount net retention. And then we have a long time series, we can't be perfect, but in what net retention is broken down by clients and sector, et cetera. And so that's how we work with the consumption model in making sure that we meet the obligations to the investors.

So there's 2 main, kind of, ways we're looking at AI for Datadog that we described as Datadog for AI and AI for Datadog -- and on the Datadog for AI front, thinking about how a customer would move from monitoring microservices into GPUs or training models or agents. Exactly. Is that -- I mean, maybe on the agent side, that's very pervasive. Is that something that every customer could do? And then, like, what kind of consumption changes do you see? How does it make them stickier?

Yes. We're basically setting it up. So whatever they're doing, if it's a human, it's an agent, it's a coding agent, if it's a large language model, we're set up to monitor it. And we generally -- for the most part, we monitor production environments. We're starting to do more in training. So we basically set that up, and we've been seeing very good growth in that area. So we have to do the investment to set it up.

And then as clients introduce them to production environments, there are a lot of metrics we've been giving out on the growth of agent monitoring, the growth of MCP calls, lots of metrics, if you read the scripts, et cetera, you'll see these things are growing at a very high rate. It's still early on. So if they're training models and they're not putting in production and it's in-house or they're using it for their marketing collateral internal, it doesn't tend to be our market, but more and more of it is being our market.

So we're monetizing it through pricing we publish as we put into GA. And it generally is on -- like everything else, it generally is on the amount of data consumed or the amount of investigations and things like that. Still early days, but really good growth signs. That's Datadog for AI. Then there's AI for Datadog, which means when you're using the platform, are you able to automate more quickly, investigate, use models, figure out what's going on, root cause and things.

And that's what we're putting in the model itself. That's a lot of what the Bits product is. And again, it's early on, but we're seeing traction in that, which we're optimistic. Those of you that know us for a long time know that we don't call it. We don't go like we're going to have $1 billion of this. What we do is we say that we're getting traction. And then when we get to certain amounts, we tell everybody we've done it. And so we're seeing that high growth.

Within that Datadog for AI, I'm thinking back to, like, 2024 before a lot of this infrastructure spend took off. We've got LLM monitoring. We have GPU monitoring. What -- so you've had some recent wins there. Like, what do you think is driving maybe more interest or an inflection there? Is it just maturity of the AI ecosystem? Is it maybe the rise of open weight models? Or, like, what's...

No, it's that our clients are putting LLM-enabled applications in production. So we're following that. So essentially, as most of you know from following it, the early part of it was very training and research. And most of the early applications, this is what our consumer or training, et cetera. And now we're starting to get to the next stage.

That's why you're seeing all this information about enterprises. I'll use Datadog as an example, developing their models, not just doing API calls out to the large foundational models, but also using open weights and their own data. So, for instance, Datadog itself is in the evolution of its own models, which we have a research lab. We just made an acquisition.

And so all of that is very, very typical of what's happening across enterprises in getting to the next stages of moving beyond API calls to create their own models, et cetera, then tune inference on their models. That's what's happening at Datadog, and that's what -- where we are -- that's most of what we're doing, and that's starting to accelerate in terms of our clients as well.

You talked about this, the research lab, I think it is Adaptive ML. You have a world-class R&D team. I would describe them as primarily organic, and an ability to integrate some inorganic R&D. Datadog was actually one of the first companies that we ever launched on, and it became apparent. We always knew it was a world-class team, but it became apparent over time how that could compound. But when you look at what you're doing with Bits AI, how and what is -- like, is there an additional advantage that you gain by being able to post-train using the data asset that you have today?

Yes. I mean it all goes back. So the platform has tremendous weight. Like if you have a model and it's not part of the overall operation of the platform, it has much less value. So we have, one, the data sets on observability, right? We have a large customer base. We have a platform that is already being -- we used to call it ML, that is already being used with analytics.

So there's just a tremendous competitive advantage, not in training models for legal processing of contracts, but in observability. And that's the competitive advantage where you're going to have specialized intelligence that we're investing in. And our view is that, that's going to wide -- deepen the moat and it's going to be something that's going to move towards self-remediation, meaning in some cases, you'll have enough intelligence that you know what's going wrong and the client will push, "yes, self-remediate" and the thing will -- and without humans, or less humans, you're going to have it. We're on the journey there. We're not quite there yet, but that's what the vision is, and it's happening.

Yes. We've been really excited by the Bits AI product. And it's continued to broaden. I think when it first came out, you were pricing on a per investigation basis. That's evolved a little bit. It's broadened; it's on a token basis. What have you -- maybe talk about what if you've heard anything from customers or trends there and just how you think about pricing and balancing gross margins with the token.

There's 2 things going on. One is you're right. We are -- we have sort of -- and this is very typical of Datadog. We play around with pricing, see where we think we can add value. And so that is being launched and early signs are a lot of good reception and use. We also are expanding what we're doing with Bits. So we started out with these investigations, right? And now we're also doing it on to the left with development and security. So the first was basically reliability engineers handling cases. And now we're also investing in security and software creation. So there's a number of vectors here with this, which is why you're probably getting signals of excitement because it's broadening out the end market, it's broadening out the workloads that are being pushed into Datadog.

So I brought this up in another fireside today, but so, one of the quotes from Matthew Prince recently was that "Humans are going to be a rounding error for traffic on the Internet over the next decade." So this right now is still a nascent portion of the business. Like how big do you think that could become? And then do you think that in some ways, you're not your user, but your customer could be changing?

Yes. I mean, we think all we can say -- I think that I'm not going to say humans are going to become. What we are seeing is that the weight between models and human and compensation is shifting so that you're -- this is happening in coding agents, this will happen in observability so that the customer will get more intelligence, more automation. There should be a lot of cases that should be able to be largely automated. There probably are going to be at least in the next bit of time, humans that are going to have to take that information and make decisions because some of this is about the delivery of their product.

So we're still in evolution. We're early on. Definitely, the weight is going to change. Where it's going to wind up, I don't know. But the good news is it probably doesn't matter, meaning we're not a seat model. We are basically monetizing based on the workloads that go through. And it really doesn't matter if those workloads are created or looked at by agents versus not. You're going to have to -- we believe you're going to have to observe and secure this. And probably all this is going to accelerate the pace of change, which means you're going to have to move from legacy. Legacy still is the biggest part of the market because it was there for 50 years. So it's all probably going to be our friend if history repeats itself.

So if it's becoming easier and cheaper to build any kind of software, and in theory, you could build some kind of internal observability tooling, not that it makes it necessarily workload, but we've seen that impact SaaS. I mean, why -- maybe just like a simple reason why that could strengthen the role of observability rather than displace...

Yes. I mean, it's basically our platform is the integration of data. So is the data publicly available or not? No. Is it data that is consumer, not mission-critical? No. How many integrations and curation of the data integrations do you have to do? And can that be done like in a public way? No. What about security of the whole thing? No. So essentially, there -- I think the world is seeing that things that are basic to the infrastructure, closer you are the infrastructure layer, the more -- it's not like you go into Claude or ChatGPT and you create Datadog. And so I think a lot of it has to do with Datadog. A lot of it has to do with how it's all integrated. A lot of it has to do with mission-critical and all security. And so we feel that what will happen here is software gets developed more quickly.

You're still going to have to quality control it, curate it, put it into production somehow. You're going to have to observe it. You're going to have to still manage your loads of provision of GPUs or CPUs, et cetera. All of that still has to be done. And anything that can be done to have more of that flow into modern platforms or modern infrastructure is going to create more workloads for the modern companies like Datadog to observe. And that's what's been happening. I mean, you can see it in the numbers. And we're not predictors of the future, but it looks like that's happening again.

Right. Yes. I think the AI native success too helps give conviction there. As we near time, I just want to make sure that the audience has a chance to ask a question if they'd like. All right. So I mean you're accelerating revenue. You're holding margins roughly flat, free cash flow in the high 20s. Can you just give us a sense of how you plan on investing in this, again, the world-class R&D organization and how you also manage the pace of hiring sales and the kind of dynamic consumption environment?

Very good question. So I think for right now, people -- there needs to be salespeople to sell, meaning we haven't gotten agents to be able to sell. So we look at that as what's the TAM, what are we covering? What do we need to continue to be able to sell our software? And right now, we've done a good job. We have successfully ramped quota capacity on a global basis, and it's been roughly in line with revenues. And we're not at the point where we saturated the market. We still have many areas. I can spend a lot of time on that.

So that we're going to continue to invest in. When it comes to R&D, I think we're going to continue to invest at a high rate, but it's very likely, and it is happening, that the percentage of that, the allocation of that is going to move more towards tokens than it is to humans. We're playing around with it. But you asked about margins. You can see our margins haven't changed, right? But we are also -- we're not token maxing. We're basically using tokens in order to create good software and products to release to clients. But you can see it's more of a distribution than a margin erosion. And that's what we see right now. So I think that's what's going to happen. In many ways, it's a little easier because to have to basically -- to all depend on humans and recruitment of humans takes more time than if you could find a way to use coding agents and everything to speed things up.

Okay. Final quick question.

Somebody is out there.

[indiscernible]

Yes, Datadog is observing is basically knitting together all of that to observe software in production. And there have been -- I'm now 8 years at Datadog, we've been public for over 6. There has been the same thing. It's Snowflake in your market, it's Palo Alto in your market, it's Splunk in your market, it's blah, blah, blah. Okay. So easier said than done. And it hasn't changed the competitive dynamic because when you're spending that amount of R&D and you're focusing on a problem and you're good at it and you're relentlessly investing, it's not -- the market is not there.

Now the same could be said for us, what about us? So we're trying to pick the areas like in security or in service management that have a lot of synergies to what we're doing with our observability platform. So we're not saying we're going to secure desktops. We don't use it. We're not saying that we're going to be a coding repository. We're not saying that we're doing email security. We're basically saying that we are working on the security of modern cloud workloads where it's using a lot of the same raw materials in the case of cloud SIEM logs, and we have a right to win.

So I think that is not saying I'm going to take over the whole security market. It's saying, I am going to expand into the parts that are around the management of cloud workloads. That's how we're approaching it. And I think you see when you look at everyone else who is trying that it hasn't affected the market, broadly speaking, in this period of time. Who knows 20 years from now? But it's pretty much resulted in a strengthening of the competitive position of Datadog, not the weakening.

David, I wish we had more time. Thank you so much for joining us.

Thanks a lot. Thank you. Thanks, everybody.

Good day, and thank you for standing by. Welcome to the Q2 2026 Datadog Earnings Conference Call. [Operator Instructions]

Please be advised that today's conference is being recorded. I would now like to hand the conference over to your first speaker today, Yuka Broderick, Senior Vice President of Investor Relations. Please go ahead.

Thank you, Lauren. Good morning, and thank you for joining us to review Datadog's Second Quarter 2026 Financial Results, which we announced in our press release issued this morning. Joining me on the call today are Olivier Pomel, Datadog's Co-Founder and CEO; and David Obstler, Datadog's CFO.

During this call, we will make forward-looking statements, including statements related to our future financial performance, our outlook for the third quarter and the fiscal year 2026, and related notes and assumptions, our product capabilities and our ability to capitalize on market opportunities. The words anticipate, believe, continue, estimate, expect, intend, will, and similar expressions are intended to identify forward-looking statements or similar indications of future expectations. These statements reflect our views today and are subject to a variety of risks and uncertainties that could cause actual results to differ materially. For a discussion of the material risks and other important factors that could affect our actual results, please refer to our Form 10-Q for the quarter ended March 31, 2026. Additional information will be made available on our upcoming Form 10-Q for the fiscal quarter ending June 30, 2026, and other filings with the SEC.

This information is also available on the Investor Relations section of our website, along with a replay of this call. We will discuss non-GAAP financial measures, which are reconciled to their most directly comparable GAAP financial measures in the tables in our earnings release, which is available at investors.datadoghq.com.

With that, I'd like to turn the call over to Olivier.

Thanks, Yuka, and thank you all for joining us to go over our Q2 results. Let me begin with this quarter's business drivers. Our revenue growth in Q2 has accelerated across our customer base. On one hand, our [ AI native ] customer cohort continue to grow and diversify, both in the number of customers we serve and the scale of those customers. But on the other hand, and as a great illustration of the breadth and strength across our business, revenue growth for our non-AI customers also accelerated again this quarter to the high 20% year-over-year, up from the mid-20s last quarter and 18% in the year-ago quarter.

Overall, we continue to see healthy trends in customer demand. Our broad base of customers from the most nimble start-ups to the largest and most established enterprises are all adopting AI. We think this is accelerating the usage of cloud and modern technologies as well as their usage of the Datadog platform to observe, secure and act on their cloud and AI workloads.

Regarding our Q2 financial performance and key metrics. Revenue was $1.12 billion, an increase of 36% year-over-year and above the high end of our guidance range. We ended Q2 with about 33,400 customers, up from about 31,400 a year ago. We also ended with about 4,720 customers with an ARR of $100,000 or more, up from about 3,850 a year ago. These customers generated about 91% of our ARR. And we generated free cash flow of $279 million with a free cash flow margin of 25%.

Turning to product adoption. Our platform strategy continues to resonate in the market. For example, 58% of our customers now use 4 or more products, up from 52% a year ago; 37% of our customers use 6 or more products, up from 29% a year ago; and 13% of our customers use 10 or more products, up from 7% a year ago. So we're landing more customers and delivering value across more products, and our products are broadly delivering strong growth in usage and ARR.

As an example, RUM or real user monitoring now exceeds $200 million in ARR and accelerated at its scale to over 50% growth year-over-year. Our customers are [ sending ] more user sessions and using RUM in conjunction with our newer product analytics to optimize our business outcomes.

Moving on to R&D. We held our DASH User Conference in June, where we announced over 100 exciting new products and features for our users. So let's go through some of the announcements. First, we expanded Bits AI to accelerate and automate the DevOps loop. This is the look that goes from detection to investigation to remediation that engineers go through each time something breaks. At DASH, we announced a lot of new Bits capabilities for the DevOps loop. Bits can now create and maintain monitors, identify root causes within minutes of [indiscernible] signal, recommend and implement fixes, follow guardrails to add safety and controls, continuously learn and improve from prior incidents and detect symptomatic behaviors early to repair infrastructure issues before they escalate.

Second, we announced Bits AI products to address the development loop. This is the loop that goes from coding to delivery to evaluation that develop and navigate to get code to production. For this loop, Bits Release now acts as an AI release validation agent, analyzing the impact of co-changes, running end-to-end checks and verifying production rollouts. Bits Code generates code fixes, running every fix in reproduction behavior, and Bits Testing also automates synthetic test generation and maintenance.

Third, we expanded Datadog for AI, our products that observe, secure and optimize the AI stack from end-to-end. Data observability enables companies to trust the data being used by AI with line edge quality monitoring and jobs monitoring, Bits Data Analysis uses our rich data context to accurately answer business questions and Agent Console provides visibility into AI usage, cost and effectiveness.

In Agent Observability, our patterns capability automatically clusters user interactions into behavior groups to identify quality or cost issues, and Bits Evals handles the repetitive parts of the agent development loop in order to improve the outcomes of agents.

Fourth, we are broadening our platform to ingest, correlate, analyze and act on more data, whether on-prem or in the cloud. In Network Monitoring, we launched network path and network configuration management to track changes that cause complex network issues. Within database monitoring, Bits Database Optimizer now automatically simulate and evaluate the impact of AI generated changes in order to optimize logistics queries. In log management, federating logs enables users to acquire external data stores, including Databricks and [indiscernible]. And with Bring Your Own Cloud or BYOC, customers can now use a full Datadog experience on logs that are kept within their own infrastructure. And we've also announced that we are bringing BYOC to Metrics and Traces as well.

In the digital experience space, [indiscernible] monitoring automatically gives a single shared view for every critical user flow. And for custom metrics data, we introduced infinite cardinality metrics, which allow our users to answer arbitrarily complex questions as they generate larger amounts of data with AI agents without incurring any extra costs.

Finally, we launched a number of innovations to secure the AI stack and defend against a new class of AI-powered attacks. AI Guard agent discovery, finds and maps every known and unknown custom agent so security teams can see what is protected and what is not. AI Guard for custom agents provides run-time protections to block attacks that can only be detected with real-time observability data. AI Guard for coding agents applies the same deep observability to block [indiscernible] and packages in code. And we also announced Runtime Prioritization Engine to cut vulnerability noise by over 95%. And finally, we expanded Bits Security Analyst to run unknown Datadog teams, so customers can benefit from the smart central learnings of our broad data set regardless of which [ SIEM ] they deployed.

As we continue to innovate, we are being rightfully recognized by independent research. We are pleased to see that for the 6 year in a row, Datadog has been named the leader in the 2026 Gartner Magic Quadrant for Observability Platforms.

Let's move on to sales and marketing and look at a few of the deals our GTM teams have closed in what has been a very strong quarter. First, we landed a 6-figure annualized deal with a Fortune 10 company. This company is expanding its e-commerce business and they plan to use Datadog Log Management alongside 10 other Datadog products to improve customer experience and business outcomes. This win validates our expected go-to-market approach to focus on the world's largest companies and win opportunities in the most complex environment.

Next, we landed 7-figure annualized deals with 2 new labs. These AI labs are rapidly scaling their AI model training workloads and preparing for major product [ launches. ] By deploying observability using Datadog, they gained visibility across our training infrastructure and GPU fleet and can iterate faster on their AI models. They are also using Bits AI to rapidly build monitors, dashboards and alerts for deep observability context.

Next, we landed a 7-figure annualized deal with a South American bank. This bank's fragmented legacy monitoring stack and manual triaging caused significant application downtime that was often called in by customers. By consolidating into Datadog with 11 products, this customer enables visibility from their mainframe all the way to their micro services and has already reduced [ mean ] time to resolution on live production incidents. They are adopting Cloud SIEM and data security and evaluating other Datadog security products to improve their security posture.

Next, we signed a 7-figure annualized expansion for an 8-figure annualized deal with a Fortune 100 health insurance company. This customer's biggest pain point is to deliver a great experience to their members throughout their care while protecting PII across dozens of business units. Datadog's HIPAA compliance and PII handling in RUM, log management and Cloud SIEM allowed us to differentiate and win over our competitive solutions. And Bits AI investigation is already speeding up the incident resolution and reducing expensive escalations. This customer will expand to 19 Datadog products.

Next, we signed a multiyear over $30 million TCV deal with one of the world's largest online media companies. This customer [ shows a standard ] from Datadog across its business, displacing 4 commercial and internal tools. Datadog also proved value beyond core observability with product analytics, CI visibility, data observability and [indiscernible] cost management. This deal includes our largest win to date for Bring Your Own Cloud, displacing their legacy commercial logging tool at a [indiscernible] scale.

And finally, we signed a 9-figure renewal with a leading AI company. This long time very large customer uses 17 Datadog products to enable unified visibility on production workloads at a very large scale, albeit with a usage reduction starting in Q3, which were considered in our guidance and which David will speak to.

Before I turn it over to David for a financial review, let me offer a few words on our longer-term outlook. There is no change to our overall view that digital transformation and cloud migration are long-term secular growth drivers for our business. But we now have an additional growth driver with AI as we help our customers deliver value with this transformative new technology. We are tremendously excited about the opportunities in AI.

To summarize where we are and where we're going. First, AI is a tailwind for Datadog today as cloud consumption grows and drive more usage of our platform. As of Q2, over 750 AI customers use Datadog to monitor and improve their tech stacks. When we look at the largest companies driving AI, all 10 of the top 10 AI leaders are Datadog customers.

Beyond AI native, we see AI activity growing across our broader customer base. We're also seeing signs of rapid growth in agentic activity with a number of MCP [indiscernible] quadrupling again quarter-over-quarter and growing more than 22x when compared to Q4 2025.

Second, we are delivering AI for Datadog to deliver more value and greater platform capability to our customers. This includes our Bits AI products, chat, investigation, detection, code, testing, readings and many, many others.

Third, next-gen AI introduces new complexity and observability challenges. We are addressing this with what we call Datadog for AI to observe and secure the AI stack from end-to-end. This includes the GPU monitoring, agent observability, agent console, data observability, AI Guard and many other products.

Finally, our AI research team and our large volume of rich data using critical workflows enable us to conduct groundbreaking research. We have shown some of our work already with the second version of our time series model, Toto in May. Toto version 2 was exciting for 2 reasons. First, we've shown it to be state-of-the-art on key benchmarks. But more importantly, we've demonstrated for the first time true scalability for time service models, allowing us to target the same improvement path language models have followed since 2020.

So now beyond Toto, we are working on larger and more ambitious dedicated models, [indiscernible] models to power Bits AI and bringing other modalities beyond time series data into world models that we think can lead to a step change in capabilities for our customers. And we plan to accelerate these research efforts with the acquisition of Adaptive ML, which we closed in June.

Because of all of that, now more than ever, we feel ideally positioned to have customers of every size and every industry as well as all types of users, whether humans or AI agents, so they can transform, innovate and drive value to AI and cloud adoption.

And with that, I will turn it over to our CFO, David.

Thanks, Olivier. Our Q2 revenue was $1.12 billion, up 36% year-over-year. Within that, our 11% quarter-over-quarter revenue growth is the highest since Q2 2022. And our quarter-over-quarter revenue added of $115 million is a record by a significant margin. We continue to see robust usage growth from existing customers as well as a strong ramp in our new customers.

Revenue growth accelerated with our broad base of customers, excluding AI customers to the high 20s year-over-year, up from the mid-20s percent last quarter and 18% in the year-ago quarter. We saw robust growth across our customer base with broad-based strength across customer size, spending bands and industries.

Meanwhile, our AI customers continue to grow rapidly and diversify in the quarter. This 750 strong customer group includes a broad range of AI start-ups as it has in the past, but now also includes hyperscalers using Datadog for in-house AI labs. In Q2, this includes 31 customers spending more than $1 million annually, of which 8 customers spent more than $10 million annually.

We also achieved strong new logo dollar bookings with particular strength in enterprise, where new logo annualized bookings more than doubled from a year ago. And we are seeing new logos ramping faster, contributing more to revenue growth. The portion of our year-over-year revenue growth that relates to new customers was about 30% in Q2, up from 25% in Q1. Geographically, we are performing well in all regions with growth acceleration across the regions. We see particular strength in the Americas as much of the AI activity is occurring in the U.S. as well as -- in addition, we are executing strongly in LATAM.

Regarding retention metrics, our trailing 12-month net revenue retention percentage was in the low 120s, similar to last quarter, and churn remains low with gross revenue retention in the mid- to high 90s. We believe this metric highlights the mission-critical nature of our platform for our customers.

Now moving on to our financial results. Billings were $1.18 billion, up 38% year-over-year. Remaining performance obligations or RPO was $3.47 billion, up 43% year-over-year. Current RPO grew about 40% year-over-year and RPO duration increased year-over-year. As we previously mentioned, we continue to believe revenue is a better indication of our business trends than billing and RPO.

Now let's review some of the key income statement results. Unless otherwise noted, all metrics are non-GAAP. We have provided a reconciliation of GAAP to non-GAAP financials in our earnings release. Our Q2 gross profit was $892 million for a gross margin of 79.6%. This compares to a gross margin of 80.2% last quarter and 80.9% in the year ago quarter. As we've discussed in the past, our gross margin varies from quarter-to-quarter with investments into innovations for our customers, offset by efficiency efforts. There's no change in our expectations for gross margin, which has been in the 80% plus or minus range historically.

Q2 OpEx grew 26% year-over-year versus 31% last quarter and 36% in the year ago quarter. We held our DASH Conference, User Conference in June, and as expected, the event cost about $15 million. Q2 operating income was $257 million for a 23% operating margin compared to 22% last quarter and 20% in the year ago quarter.

Turning to our balance sheet and cash flow statements. We ended the quarter with $5 billion in cash, cash equivalents and marketable securities. Cash flow from operations was $316 million in the quarter. After taking into consideration capital expenditures and capitalized software, free cash flow was $279 million for a free cash flow margin of 25%.

And now for our outlook for the third quarter and the fiscal year 2026. Our guidance philosophy overall remains unchanged. As a reminder, we based our guidance on trends observed in recent months and imply conservatism on these growth trends. Regarding our largest customer, we have seen a usage reduction, which is incorporated in our Q3 and full year 2026 guidance. As Olivier noted, this customer has recently renewed with us.

For the third quarter, we expect our revenue to be in the range of $1.135 billion to $1.145 billion, which represents a 28% to 29% year-over-year growth. Non-GAAP operating income is expected to be in the range of $260 million to $270 million, which implies an operating margin of 23% to 24%, and non-GAAP net income per share is expected to be in the $0.63 to $0.65 per share range based on approximately 378 million weighted average diluted shares outstanding.

For the full fiscal year 2026, we expect revenue to be in the range of $4.45 billion to $4.47 billion, which represents a 30% year-over-year growth. Non-GAAP operating income is expected to be in the range of $1.01 billion to $1.03 billion, which implies an operating margin of 23%. And non-GAAP net income per share is expected to be in the range of $2.50 to $2.54 per share based on approximately 376 million average diluted shares outstanding.

And for some additional notes on guidance, we expect net interest and other income for the fiscal year 2026 to be approximately $180 million. We expect cash taxes in 2026 to be about $30 million to $40 million. We continue to imply a 21% non-GAAP tax rate for 2026 going forward. And finally, we expect CapEx and capitalized software together to be in the 4% to 5% of revenue range in the fiscal 2026.

Now finally, to summarize, we are pleased with our execution in Q2. Our investments in R&D and go-to-market are yielding positive results, and they position us well for continued execution. I want to thank all the Datadogs worldwide for their efforts. And with that, we'll open the call for questions. Operator, let's begin the Q&A.

[Operator Instructions] Our first question comes from the line of Sanjit Singh with Morgan Stanley.

Congrats on the acceleration in revenue growth again this quarter. David, thank you for giving us the color on some of the guidance assumptions, particularly headed into Q3. With respect to the largest customer, I was wondering if you could share any additional details in terms of the new contract? Was it a similar duration? And in terms of the lower usage. Is that a function of the customer getting lower unit price because of making a new commitment? Or is there some churn or downsell that we are seeing through, not only for Q3 but for the balance of the year?

Yes. So maybe I'll take this one. I think we -- So overall, we, as usual, we don't want to comment too much on any specific customer because we're also not really in control what's happening with any specific customer. We wanted to be transparent about this on the call because we did see a reduction in usage, and we took the liberty to fully derisk the guidance for the rest of the year with respect to that customer. And again, the reason for that is we don't control what's happening to a specific customer, but we do have a great amount of control on what's happening to everything else in the business, and the business is booming, and we don't want that to overshadow basically the acceleration we see pretty much everywhere else in the business.

So the -- as we mentioned on the call, we renewed the customer. It's a long-time customer, uses many of our products. But there's not a lot more we can share.

Yes, I think just to get specific on the guidance, we, last quarter and previous quarters, said that we essentially have a level of commit, and we can derisk our guidance by using that. And then as you know, in most of our large customers, we have variability relating to the commit. So take that into consideration.

Yes. The last thing I will say because I know it's on people's minds is if you back out our largest customer from our growth, you get pretty much the same growth rate as the rest of the business has been accelerating very steadily. Actually, we've seen, I think, now 5 quarters of continuous acceleration from the rest of the business. And we feel very good about the -- what we see in the market.

Yes. No, I appreciate the thought. Let's talk about maybe the rest of the business. What we've seen in the past couple of years, sort of AI Native is sort of leading the charge. It sounds like the enterprises are getting on board with their AI initiatives. And so just in terms of like the enterprise AI app dev cycle, what does that look like for Datadog over the last couple of quarters?

Well, we do see broad adoption, and we see it in 2 ways. One is we see it manifest itself in just more transformation, in more cloud adoption, more workloads, more [ modernization ] from customers. And that's what drives the majority of the known AI customer acceleration. So we mentioned also we've seen continuous acceleration from customers that existed before AI and are not majority AI businesses. And it's been pretty remarkable, like the acceleration that we gave the numbers on the call, but their acceleration since last year has been constant and very significant. And we -- it keeps happening as far as we can tell. So it's a very positive trend there. That's the first thing we see.

The same thing we see is a very rapid increase in the usage of all of our AI first surfaces. So that would be the products that measure agents and LLMs, where we see an explosion of traffic in terms of the LLM and [ tool calls ] that we're getting. That would be the amount of closed book gain to our MCP endpoints. So we see that explode completely over the past 2 quarters.

Our next question comes from the line of Raimo Lenschow with Barclays.

Perfect. Could I stay on that AI theme, please? At the moment, like if you think about the large customers, there's a lot of model training, et cetera. But if we broaden it out, the inference is really becoming the bigger part. Can you talk a little bit about like how much more observability is needed? And I'm thinking there, if I do inference, I need to think about vector databases. I need to think guardrails. All of these agents are going to be in containers that need to be monitored, et cetera. So what do you see in real life at the moment in terms of if some people do more inference, how much more observability gets triggered by inference? Is that kind of an opportunity that we should probably pay more attention than that one renewal? And then I had one follow-up.

There's opportunity at every layer of the stack in inference. So we do think at the end of the day, inference will be the dominant workload. That's -- any time you train, you probably would want to infer more than you train as a rule of thumb. We see opportunity at the low level when it comes to the infrastructure, the GPUs and the consumption you have there. There's opportunities at the very top end, when you measure what the agents are doing and whether you're getting the right outcomes and whether you're getting the right alignment, and there's opportunities at every layer in between, just looking at the LLM itself, just looking at the tool calls and the applications that are being called by the agents, like everything is an opportunity in there.

We see growing adoption from the products we already have there. We mentioned our GPU monitoring product is actually getting quite a bit of usage in a number of new labs and AI first types of customers. We're also seeing an explosion of volume in our agent monitoring products. And so we're well positioned there. But we think this market is going to change quite a bit. And the [ occupations ] of customers, they also change over time. So for example, last year, our customers were mostly trying to validate correctness and validate that they were getting some form of [indiscernible] scale-up. I would say, 3 to 6 months ago, the focus has moved quite a bit towards cost. Now customers were spending a lot on AI and they were wondering how to optimize costs. And I think we'll see some variations in the concerns over time as customers get further into the adoption and new products emerge for that.

Just want to add that when you look at what we described as some of our deals in the quarter and you look down our script, you'll see that a number of them have the AI products included. And so that is indication that those large enterprises are using the platform and buying the AI products as well.

Okay. Perfect. And then David, one for you. It's like, obviously, you're always in a tough position if you have to guide these large contracts. How did you do it historically? So did you always kind of put in the base level and then what happened, happens? Or has that approach changed? Or I don't [indiscernible] you on having to do this.

No. We essentially use, as we've talked about over the many years, we kind of use the inputs of what we see. And what we said, I think, in the last quarter or 2 is that we have certain base levels. As you know, we have a commitment and a usage model. And we've factored that in, in providing our guidance. So our -- as we said in the prepared remarks, our methodology for guidance hasn't changed. We've always used those inputs and looked at the commitment in usage in doing that.

Yes. I mean one thing I'd say is in this case, we did chose to fully derisk our largest customer. And the reason for that is we don't want that to be an overhang on what is otherwise a business that is accelerating and performing extremely well. So we extended that. We have the same overall conservatism as we always do when we look at our numbers. But in this case, we also weighed this one a little bit differently.

Our next question comes from the line of Gabriela Borges with GS.

I wanted to ask you both about one of our observations of DASH, which is the [indiscernible] love the pace of innovation. They talk very positively about the product. The CFOs love to complain a little bit about their data on [ bills. ] So my question for you is talk to us a little bit about how the CFO level conversations are evolving. Clearly, the ROI is there, but maybe give us a little bit more on where the budget is coming from? And something like infinite cardinality, is that now part of the conversation with CFOs in solving some of those very particular cardinality cost portion?

I mean, look, at a high level, there's only 2 reasons people buy software. It makes them more money or it sells them money. And any time we sell, any time we got a renewal, we've got an [indiscernible] or we land a new customer, that's because we do one of those two things for them. And we always have to make that case. So I wouldn't say that's any different from what we've seen before.

What we do for our customers today, especially as they keep adopting AI is we help them sell a lot of the money they would spend on building, running operations or running AI agents. When you [indiscernible] customers, that's the one thing they kept mentioning, hoping you help me running my AI cost. This is growing very fast. I don't have any control on it. And I don't know whether I'm reaching the right outcomes with that. And so that's one of the reasons we've invested in all these products we've mentioned earlier. And also, we're seeing some of the great returns on that product already.

In terms of Infinite cardinality, that's -- I would say it's been one of the longer-standing source of frustration for customers when sometimes they send more data or they send more fine-grain tags with their data and to get some unpredictability on the [ bills ] because of that because it increases the [indiscernible] of the data we're getting. And we saw that from a technical perspective and from a commercial perspective by packaging our metrics a little bit differently. And we think it's particularly important and relevant as customers are building more applications with AI and as they want to send basically more [ tech, ] more information and ask more complex question and get more fine grain answers to those questions. And so that fits well within their plans basically. So we've got great feedback on that so far, but it's still early. Sometimes we get it right, sometimes we get it slightly wrong, and when we get it slightly wrong, we fix it. That's not different from what we've done in the past.

Our next question comes from the line of Mike Cikos with Needham.

I wanted to come back to the significant size of the lands that you had this quarter, and it's great to see the sustained traction especially with those AI labs. But if I'm thinking about the 2, 7-figure AI labs that you landed this quarter and then going to David's commentary around winning some of these in-house AI labs with the hyperscalers. Are those one and the same here? Or are those two separate customers that we're talking to?

These are different customers. The ones we mentioned on the new lands are [ neolabs. ] So these are companies that didn't exist a few years ago. And what's interesting about them on the use case there is that very often, we land customers when they go into production and they release products and they start selling their customers. In this case, these are customers we're getting as they are training models, and they're using us to observe and improve and optimize the training of the model. And so that's an exciting new area that was not really a business area for us a couple of years ago, and we've seen a number of new proof points around that.

In addition to that, and we've mentioned in previous calls, we've also landed the AI labs or super engineers labs or the number of hyperscalers. And I would say the workloads are similar in that it's largely training of the models, but the customers are a little bit different. These are very large companies that in that case, previously had a lot of in grown -- a lot of homegrown technology to observe and run workflows.

Excellent. And for a follow-up, I know you had cited the new logos ramping more strongly than what we've seen historically. And correct me if I'm wrong, but I feel like that's a newer phenomenon that you guys are calling out this quarter. When I think about those new logos ramping, is that a function of pull-through where maybe some of these AI capabilities are pulling through the broader platform? Or is it vice versa? Anything you can do to help us think through what is creating that catalyst, if you will, when the new logos are contributing to the model?

It's been happening and building up some with the number that we have in our Qs, which is the percent from customers of growth that we didn't have a year ago. That number, we said, has gone from 25% to 30%. So this has been building, and we wanted to point that out because of that disclosure indicating that the customers that we're landing that it's not only the new logos, but it's also the growth of the new logos that we've added over the last couple of years -- last year, sorry. So it's a compounding of that.

Our next question comes from the line of Alex Zukin with Wolfe Research, LLC.

Oli, maybe first for you, just on the -- a lot of headlines around security over the course of the last few weeks, particularly AI breaking containment. And it occurs to me that with your positioning and observability and security increasingly, the notion of a guardian model and development around that could meaningfully increase kind of your ambit on what you can do and achieve for clients, both AI natives and legacy. Can you maybe talk to what -- the increasing opportunity around this crossover in this AI agent and what that means for Datadog? And then I've got a quick follow-up for David.

I mean look, there's a complete switch in the way the AI security products need to work. So you can't wait basically for putting humans in the loop. You can't have the typical path when you have 12 or 15 different products that are going to aggregate signal and then you get it linked into a system [indiscernible] prioritize them for humans, then humans will leave them when they can. Like you need to integrate everything a lot more. You need to operate a lot closer to the application into the infrastructure. And you need to have AI agents solve the issues first.

So it's a complete rebuild for most of the industry. And I think it plays into our approach, which is to have an integrated platform and have all of the different data streams come directly from observability straight into the security agent and have all that integrated from end to end.

So obviously, the data field is moving very fast. We see new classes of issues pretty much every week at this point. We are quite busy building that up, but we think it displays into our strengthening to where we are basically already on and what we're building for our security products.

Perfect. And then, David, maybe just for you. On the largest customer renewal. Is there anything you can tell us around maybe just any changes around the duration or anything that makes this new contract maybe a little stickier in terms of the discounted rate card, the amount of products that they're able to kind of use for better value? Anything that increases the conviction level around stickiness?

I'll comment on this other than to say that most of our enterprise customers, as we talked about for a long time, have annual plus and then the pricing is generally volume-based pricing. So I would say, overall, our customers transact in that way. And then we have that level of commitment. And then as we talked about over a lot of years, then there's usage and then we transact. So similar to what we have with most of our larger enterprise customers. Oli, anything you want to add there?

No, I think clearly there's a lot of continuity in that renewal. I think that's what you gain. That's one way to put it.

Our next question comes from the line of Eric Heath with KeyBanc Capital Markets.

This is [indiscernible] on for Eric Heath. I would love to get more color on your 3Q, guys, specifically. It seems like it's a little below your sequential levels of how you've guided your previous 2Q. So I would love to just hear more about what trends you're seeing going into 3Q and maybe what some of the assumptions in the guide are?

Yes. I think it's similar to the methodology we take, what we see and provide some conservatism. And I think we had mentioned in the script that we've been -- renewed our largest customer, but we've seen user declines relative to the previous quarter. We said that. So that's all taken into consideration in trying to develop a guidance that is consistent with the methodology of conservatism that we've used as a public company.

Got you. And if I could just ask one more for Oli. I'd love to just get your thoughts on the impact of diversification of AI model usage in your customers and what you're seeing there?

Well, we think it's great. Like there's a lot more options for customers to choose from. And that creates -- that opens up a lot of doors and opportunities for them. It also creates a lot of complexity, and we're here to help them with that complexity. So for us, these are great opportunities.

And by the way, we see -- like we've had that thesis since the early days of AI that we would not just end up with 1 or 2 big AI companies and everybody using them, the same way we didn't just end up with 1 or 2 big cloud companies and everybody just using software from them. Like the ecosystems are very, very, very rich. There are a lot of providers. There are very large providers, there are smaller providers and everything in between, and there are many compositions of those different systems that are used by any given customer. And so we think the same is going to happen in AI.

We think also that the [indiscernible] application of model as an open source model in particular opens the door to customers doing a lot more training on their own. And so that's a new market for us. We see some signs that we have a very good role to play there. And so we're building towards that as well. So overall, it's very positive for everyone.

Our next question comes from the line of Koji Ikeda with Bank of America.

Just one for me here. I wanted to ask on Bits AI. All the commentary that you guys are saying on Bits AI and all the work that we've been doing in the quarter sounds like Bits AI has really taken off for you guys. And so just thinking that Bits AI is going to be increasingly automating activities that historically has created observability workflows. I'm curious and really wonder, how do you ensure that greater automation that might be driven by Bits AI doesn't eventually reduce the volume of activity that traditionally drove Datadog consumption?

Well, look, if we provide more value, we'll get more. As I was saying earlier on the call, like we sell more software by helping customers make more money or save money or both. And I think if we can automate more and let them do more, we'll provide more value. That's as simple as that.

I think the future of observability is not just observing, it's fixing. It's not waking your people in the middle of the night because something -- bot is fixing it for them. It's not letting people do damage control on the security in [indiscernible] because [indiscernible] is in. It's preventing the [indiscernible] from getting in to start with by auto remediating issues, and we're very, very busy building all of that. And we're super confident that this will yield great business outcomes for us in the end. And that's what we see from customers in the market. Like when they use Bits AI, they use more of our product. They deploy more of it. They create more, that bots and alerts and everything else. They have more users inside of our product, like it's not a zero sum game.

Our next question comes from the line of [ Samik Chatterjee ] with JPMorgan.

Maybe just on the non-AI part and the acceleration that you're seeing related to the non-AI part of the business. I just wanted to sort of get your thoughts on the sustainability and whether this acceleration that you're seeing is driven by some of the new customer logos that you're pointing out or more usage going up? And as CFOs get more sort of cautious around their budgets, do you see more sensitivity around non-AI eventually related to some of the AI products and how they're doing at this point? And I have a quick follow-up.

So I mean, from what we can tell, it's very broad-based. And it's largely driven by existing customers because that's the majority. Like when you think of what it takes to move that number, that's basically the majority of our business. But we're not just going to move that with a few newer customers. Like it's largely driven by the existing customers, and it's driven by both increases in volume and because they are moving more and more close to the cloud and adoption of our newer products as they consolidate onto us.

We think it's sustainable for one thing, if you compare to what we have seen in the early days of 2021, the growth rates are accelerating, but they are still far below what we're seeing at that time. And so we don't create the same issue of customers having to digest very large increases multiple years in a row. I think in this case, we're very well within the range of sustainability.

And as has been [indiscernible] in this call, remember that when customers adopt and they consolidate, they have an eye towards the financial side of the equation, basically, how much money are they going to make or save by doing that at the end. And we are very good at helping customers understand that and making that case and helping them save money at the end of the day. So we feel good about that.

And I want to just add one thing, and we talked about this last quarter that some of this has to do with the investments that we're making in our platform and our product, but it also has to do with the investments that we're making in our go-to-market. We've successfully expanded [ quota ] capacity, the geography of it. And essentially, that's as we talked about last quarter, providing returns. So that's also being a growth driver in our non-AI or enterprise type business.

That's right. And you see also continued investment there. So we keep investing in R&D, obviously because we're seeing more products are successfully being adopted and consolidated into our large number of existing customers, but we also are adding to our go-to-market teams. We are still not at the scale we want to be in terms of getting to all of the customers worldwide in all of the segments that are relevant to us. So we're investing as we see the return of those investments.

And for my quick follow-up here. You talked about the FedRAMP High certification last quarter. Just curious if there's anything to sort of update us on the pipeline and how -- if there's any momentum on that front on the pipeline yet?

Yes. Well, we're investing quite a bit in the buildup of our federal and government [ sales ] in general. And we see pipeline there. In general, these are not deals that happen overnight, but this is a very large market and we see great traction there and we're investing to take full advantage of it.

A lot of that was a buildup to get to the right level of certification so we can deliver SaaS to various levels of government. And we've done quite a bit there. There's actually even more we're planning to do there. But we're happy with the results so far.

Our next question comes from the line of Howard Ma with Guggenheim Securities.

Great. And congrats on the strong quarter and the full year guidance raise. I have two questions. I'll just ask them together. The first is on Bits AI. I'm curious how adoption and contribution compares to previous major future expansions in the past. And then my other question is the $30 million TCV deal with the -- I think you guys said it's the largest online -- or sorry, one of the largest online and media companies. I'm assuming this company did mostly DIY before. So if you could share some light on the decision-making process and if they're using multiple Datadog products? And why now? That would be really helpful.

Yes. I'm sorry, I missed some part of your second question.

It was, are they taking multiple products, I think, right, Howard?

Are they the nature of the sale and why now? That client, yes.

Yes, yes. So I mean, I would say -- so first on Bits AI. So yes, and one thing that happened is Bits AI used to be fairly specific. It used to be dedicated to alerts, like Bits AI would pick up an alert and would do an investigation for you. Now the surface of contact is a lot wider with the customer. So Bits AI, you can access it through chat. You can run -- you can, of course, still do the investigations and we've done quite a bit more there. You can have Bits AI manage your monitoring and manage your detection for you. You can have it code for you, you can have a generate managed test.

There's all sorts of use cases that we built into it that broadened the surface of contact, and we see a lot of adoption across all of those different areas. We also are changing the way we package it. So we have a new model with AI credits that we're rolling out just because the surface of contact is so much wider now in a specific feature. So we -- there's quite a bit that is going on there. The explosion of activity that I mentioned earlier about other parts of our other AI services is happening also in Bits AI. So that's something we're looking forward to.

[indiscernible] on that. On the second one on the products that are being updated in the sales. I mean, look, we typically land with 2 or more products that -- the balance we try to strike there is always to land enough of the platform without slowing down the deals too much because the more you try to do it once, the more stakeholders you get and the longer it takes. And so we found that 2 products in general is a good land, and then we can expand from there.

On the call, we tend to mention a lot of consolidation deals because they tend to be the larger ones, like if you land with 12 products, you're going to be larger than if you land with 2 in general. That's not the majority of the deal. The consolidation typically happens later than when we land, but this make for very interesting examples of what our customers are doing when they consolidate all at once.

Our next question comes from the line of Andrew Sherman with TD Cowen.

Congrats on the core growth acceleration. All these CPUs have had a renaissance lately, driven by agentic AI. It would be great to hear your thoughts on this topic if it can be an incremental growth driver for your infrastructure monitoring? Have you seen any evidence of this yet? That's it for me.

Look, we do see an acceleration of consumption of our [ infrastructure ] products in general. And also that's at a high level, we do see that across the customer base. I don't know that if we see specifically the CPU that get attached to GPUs in the new build-out, I think a lot of it has more to do with the fact that the AI agents are largely spending a good amount of their time, like sometimes the majority of their time calling tools, and tools are just applications that already existed and those applications typically run on CPUs. And so we see quite a bit of that.

Our next question comes from the line of Brad Reback with Stifel.

Oli, given your commentary around how strong the core is and that your largest customers are not additive to growth here in 2Q. Should we assume that if we ex out the sequential downtick in that customer that the core guide would have been probably 300 or 400 basis points higher?

Well, I can't speculate. But what I will say, look, the business overall is growing at the same rate if you expect that customer, as I said. And all the business has been accelerating overall. So that's why we feel good. Like when we look at whether we're getting the right returns and the right outcomes for our investments in R&D or investments in go-to-market. And when we look at our pipelines and all of the funds we have about the business, we feel great about business. It's a good time to be in the business.

Yes. I think we commented in the remarks that the non-AI has has accelerated and the AI, excluding the largest customer continued. So I think we gave those the trends in in describing the business.

And of course, customers are growing a lot faster on AI.

And that AI is growing. Yes, exactly.

Our next question comes from the line of Ittai Kidron with Oppenheimer & Co.

Congrats on the great quarter. I wanted to ask about new customer additions. This probably was the weakest quarter I ever remember for you guys, especially in the quarter, we had DASH where historically, that has been an accelerant of new customer additions. Any color there would be great.

Yes. Yes, I think we -- essentially, it's very similar to what we talked about before. Our gross customer additions continue to be strong and on trend line. And that's the vast majority of our revenues. We have, at the very low end, the border between free and contract and that has variability very low effect on revenue. So if you -- that accounts, as we talked about in many quarters, that accounts for the variability of the customer count and it really has to do with something that has very little effect on revenues.

Yes. When you look at the customers above thresholds that we know whether it's above $1 million or above $100,000, all those are trending very well.

Very good. And then as a follow-up, Oli, for you perhaps. I want to follow up on the questions around Bits, which sounds super interesting. I guess, longer term and as you try to push deeper also into the security side of things, could this be evolving to some broader AI SoC automation kind of platform? Is that a reasonable direction to think that this is where it's going to go?

Well, there's definitely -- we've taken move towards that, right? So we initially, so we built a scheme first from that, then we built the agent into the same. So Bits AI Synergy Analysts. And now we've actually separated the agent from our SIEM, so customers can use it with other things. And we knew that because the agent performs so well and it's been such a differentiator when we created [indiscernible] that we think we're limiting our sales market-wise, if we just go after customers that want to replatform their SIEM and that can be -- it can have a much border appeal as an AI function. So we are definitely taking more of that.

Our next question comes from the line of Andrew DeGasperi with BNB Paribas.

I just wanted to ask a question on the non-AI natives, specifically in terms of the growth that you saw in the quarter, I was wondering, did you see rising demand for the AI monitoring tool, particularly with open source tools being deployed across enterprises?

[indiscernible]

I think he's talking about within that, the AI, what we used to call AI monitoring, I think you're asking LLM, et cetera, the growth trend there.

And look, the volume -- there used to be very little volume a year ago. It started growing quite a bit into the second half of last year, and now it's been very rapidly accelerating over the past couple of quarters. So we've seen an explosion basically of the volume we're getting there. And we get more usage from different kinds of companies. So we definitely see that. We see it also across traditional companies and some more recent AI native. So we see a little bit of both.

I would say, for that category, it's still super, super early. Like we expect the products to change quite a bit. We expect the data usage and maybe also the packaging to change over time quite a bit.

All right. So I think that was the last question. So I want to thank all of you for attending the call today. I also want to again thank the teams everywhere at Datadog. I think everybody has been doing a fantastic job both in the product side and the go-to-market side. I know we have a lot more lined up for the end of the year on the product side. And I know also we have very large and very happy pipelines to then do on the go-to-market side. So I hope to talk to you again in the quarter. Thank you all.

Thank you for your participation in today's conference. This does conclude the program. You may now disconnect.

Please welcome to the stage Datadog Chief Executive Officer, Olivier Pomel.

Good morning, everyone. I'm Olivier Pomel, Co-Founder and CEO at Datadog, and I'm really excited to welcome all of you to Dash today. If you've seen our keynotes before, you know I won't be very long as we have a lot to show you. But first, I'd like to thank our partners and sponsors, and you can meet them all on the expo floor. I also want to tip my hat to our Datadog ambassadors for the amazing work they are doing with our community. And most importantly, I want to thank all of you, our users and our customers. I want to thank you for your trust and for building with us.

And there is more of you every single day from every part of the world and every corner of the economy. You represent some of the largest companies in the world as well as the top teams that are building the future of AI, whether that's building the models, the coding agents or the apps we all started using every day. So it is a truly inspirational peer group and an amazing opportunity for all of us to exchange and to learn from each other. Now the past 50 or 60 years have been pretty good for software engineering. And you can see the increase in productivity behind me. My dad was a software engineer. And he actually started programming on punch cards when he was in school. So that would be all the way out to the left there.

And ever since then, we've been through revolution after revolution with higher-level languages, with the Internet, the rise of open source, the cloud, and now with artificial intelligence. Progress has kept accelerating, and we do expect that to continue. And so what this means is that we can now build technology in weeks or even days that would have been science fiction just 5 or 10 years ago. And that's really amazing.

But it also means that we're shipping apps and systems that we don't fully understand with code that we mostly didn't write and a lot of components we don't really have a good mental model for. And so what we do end up with is an increase of complexity that is even bigger than the gain in productivity. And because of that, we keep going faster and faster, and yet it doesn't feel like our jobs are getting any easier.

This explosion of complexity is precisely the problem we set out to solve at Datadog. And for that, we can't slow down. In fact, we are accelerating our investment in R&D. We invested more than $1 billion last year, and we're going to show you some of what's coming next. So a big theme today is going to be how we can help you win that race against complexity and help you build autonomy into your operations so that you can ride that beautiful productivity curve and fully enjoy its benefits. And to get us on that path, I'd like to invite on stage my co-founder, Alexis.

Thanks, Olivier. Hello, everybody. I'm really glad to be here with you at Dash today. Last year, I told you that we've always been focused on answering a simple question: how can we help? I also told you that we saw AI as a powerful tool, and we still do. None of that's changed. What's changed is I feel that every team in this room and beyond is building more, writing more code, shipping more changes, launching more experiments. In short, teams are building at AI speed. So back to my question, how can we help? How can AI truly help? Last year, we introduced Bits AI for SRE, and your feedback has been overwhelmingly positive. Thousands of you use Bits every day to help you understand and fix issues faster and with confidence. This year, we're teaching Bits AI to be helpful in a lot more cases across every layer of the stack, so that everyone across the organization can move at AI speed, too.

We'll cover a lot of ground today, so let's get started. We are all writing more code with AI. That's awesome, but writing code is just a small part of the whole process because operating software has to keep up with that pace and volume. You cannot review every single line of code, or keep track of every new changes, or understand every new failure mode that's appearing. So the question concretely becomes, how do you ship faster and still deliver experiences your customers can trust.

If you're running software like we do, you're in this loop: detect, investigate, remediate. And it can take minutes, hours, or even days. And for years, we at Datadog have worked to help you navigate this loop as quickly and as easily as possible. With Bits AI SRE, we fast track the investigation step, accelerating one of the most high-pressure and high stakes parts of SRE. Bits is already helping thousands of teams find the root causes of issues faster. You can also chat with Bits natively in Datadog outside of an investigation. Bits Chat lets you ask, understand, and act across Datadog, creating dashboards, notebooks, triaging issues, making sense of your entire production environment.

If you have not had a chance to use Bits Chat, this is what it looks like in action. It's meant to drive the whole platform as you work through a particular problem. And good news, Bits Chat is generally available to all of you today. But Bits AI SRE and Bits Chat are just the beginning, because we know you need more than that. The amount of code, the complexity of modern software stacks are just rising too fast to keep up. So we've built a lot more in Bits. And to kick us off, let's hear from Samantha.

Thanks, Alexis, and hello, everyone. I'm Samantha, and I'm a product manager here at Datadog. For years, Datadog has helped teams answer one of the most important questions in operations: is production healthy? Traditionally, detecting when it's not depends on our teams creating monitors and tuning thresholds and keeping them up to date as our systems evolve. But as we just heard from Alexi, AI is now helping all of our teams ship more code faster. That means production is changing more often. And with all of our amazing new features also comes more surface area for something to go wrong.

So production monitoring needs to keep up. It can't be limited to the monitors that we knew to create ahead of time. It has to detect when something is unhealthy and determine if it matters as production changes. That's why today, I'm excited to announce that Bits Detection is now available in preview. With Bits Detection, Bits uses your historical telemetry, your service topology and ownership, and source code context to manage your critical monitoring automatically, continuously, through every change. Bits understands what's impactful to your business and what warrants response.

But that's enough talk. Let's see it in action. We're looking at an e-commerce app in Datadog, specifically at checkout API, the service behind the final purchase flow. And at first glance, checkout looks fine. Requests are coming in, latency is reasonably stable. And at the service level, error rate doesn't look catastrophic. But there is an active alert, and it looks like Bits AI caught something our manual monitoring missed. Last week, my team shipped a new feature powering personalized offers during checkout and launching this new endpoint, promo eligibility. Because it sits directly in the purchase path, it quickly became tied to conversion.

But Bits didn't wait for us to know what to monitor. It saw the new endpoint appear, understood that it sits within checkout, calibrated its baseline as traffic ramped up, and detected when errors moved outside of that baseline. That is why this is red now, not because my team predicted the failure, not because we manually created a monitor, because Bits kept monitoring aligned with how the service was actually changing. And as I dig into this alert, I can see that Bits didn't just turn the endpoint red and wait for me to investigate. It kicked off the investigation automatically.

And that's where autonomous operations begins. Bits catches the right problem early with enough context to move directly from detection to investigation, while our teams stay focused on building. We just saw how Bits continuously monitors production to detect impactful issues. With that, I'll hand it off to Aneesh to show how Bits continues autonomous response.

Thank you, Samantha, and good morning, everyone. I'm so excited to be here today. My name is Aneesh, and I'm a product manager on Bits AI. What Samantha just showed us is Bits detecting the checkout issue completely on its own. But being alerted on the issue is only the beginning because now we know that checkout is broken, but we don't know why. It could be because of a traffic spike, a logic bug, a connection issue, a bad deploy. All of these are plausible. But our customers are still trying to buy. So every minute we spend guessing between them is just more failed purchases.

For the past year, Bits has helped take this guesswork off of our hands by investigating the root causes of production issues using the signals around our system like telemetry, service context, deployments, and source code. But anyone here who's been on call could tell you that the hardest production incidents aren't solved by system data alone. They depend on operational knowledge. The things our teams have picked up only after years of running production, like knowing what signals have mattered in the past, or remembering all the things that have failed before, or understanding the nuances that make our environments unique from everyone else's.

The problem is this knowledge is scattered. It's usually in Slack threads, runbooks, postmortems, tickets, docs. So the next time an incident happens, your team may have already learned something in the past that could help you know where to look and what to do this time. That information is probably already out there, but it's impossible for any of us to remember where it all lives and remember which pieces will be helpful, especially in the middle of an incident. And that is exactly why we're giving Bits Memory. With Memory, Bits now learns from all of the work that our teams are already doing, and it remembers lessons as they land. It also knows what's worth remembering, like signals that my teammates share in Slack conversations during an incident or recommended actions that are listed in runbooks, any of my team's hundreds, or root causes written down in postmortems.

Bits notes these lessons down so that it can bring them back the next time they can help. And memories aren't just for investigations either. Bits brings its memories to all of its work. So the next time it's helping your team with anything, it's starting with your team's knowledge and experience.

Now let's go back to that checkout issue that we were looking at earlier, and let's see how Bits investigates using everything it knows. Bits starts investigating by generating several hypotheses at once, and it tests all of these against the live signals around checkout API. The first thing it sees is that the promo eligibility endpoint is failing in bursts. And then as Bits continues, it remember something, a different incident with a different endpoint from weeks ago, where the incident initially looked like the root cause was a connection issue. But my team dug in and they realized that the real culprit was checkout API pods crashing.

Bits captured that lesson when it happened. So right now, nobody is digging through old Slack threads trying to look for that incident's details. Nobody is manually pasting any context into a prompt box. Bits just brings that knowledge directly into the investigation, which tells it exactly where to look next, pod health.

From there, the rest of the investigation falls into place. The error spikes line up with pod restarts, which started after last Thursday's deploy. So Bits goes ahead and checks the code. And sure enough, it sees a new session cache for promo eligibility that's growing with outbound. And now within minutes, we have what we're used to chasing for hours, a root cause that we can actually trust.

Bits also lays out the whole story step by step. Thursdays deploy out of the cash, memory climbed and forced restarts, and the restarts interrupted promo eligibility. So now Bits knows exactly what happened, but our customers are still failing to buy. In other words, the investigation is done, but the incident isn't over yet. And after everything you just watch Bits do from detecting the issue on its own to investigating it, we thought it'd feel a little bit odd if we stop here and had someone log on and manually run a fix, which is exactly why we're giving Bits the ability to act today with Bits Remediation.

With our checkout issue, Bits looks for the fastest safe path to bring our service back to green. And it lands on 3 immediate options. We're starting the pods, rolling back the deployment, or scaling up. But this is production, so before Bits takes any action, it checks the guardrails that my team has set for checkout API first. These guardrails tell Bits which actions it can take completely on its own and which ones still need a human sign-off. And we'll show you how these were defined in a bit. But for now, the important part is that my team has already given Bits approval to restart pods completely on its own because it's scoped and it's low risk.

So Bits goes ahead and restarts the affected pods to clear the memory pressure and then it waits to follow the recovery all the way through. Within minutes, memory drops and checkout is back to healthy. Our customers are back to buying now, but Bits also knows that the bug that started all of this is still sitting in our code base. So Bits goes ahead and writes the code fix as well and it lets me open a PR with one click. Now production is back to green and my team has a durable fix in hand.

I want to do a quick postmortem on the incident that we just saw. Bits caught the problem on its own, investigated it with Memory, and acted safely to bring the service back. What Samantha and I just showed you is Bits for SRE. It's incident response that nearly runs itself, and that's exactly what we mean when we say the words autonomous operations. But as good as it feels to fix an incident in minutes, the best type of incident is still the one that never happens. So now I want to hand it over to Jessica to show you how Bits can also catch the routine infra issues that can turn into tomorrow's outages. Thank you.

Hi, everyone. I'm Jessica, a product manager here at Datadog. You just saw how Bits detects, investigates, and remediates when something breaks. But operations isn't just instant response. It's also the routine infrastructure work that teams do daily to keep everything running. Restarting Kubernetes deployments that crashed, troubleshooting network latency and congestion, adjusting Lambdas to prevent time outs. Individually, these issues are easy to fix.

But infrastructure issues are increasingly changing faster than we can keep up. And what you're seeing here is just a sample. Teams may have hundreds of issues in flight that don't always page. And together, these issues compound and get worse until what you're left with is a business impacting outage. And this is what Bits is always watching for, not just incidents, but the infrastructure issues that lead to them. Bits helps you get ahead of these with purpose-built autonomous operations for your infrastructure at scale. Every action Bits takes is closed loop and deterministic so your team stays in control.

And today, I'm excited to announce Bits Infrastructure Operations. Let's see it in action. My infrastructure is complex. And here in Datadog, Bits is detecting, investigating, and remediating all the issues across my environment that have been quietly building, crashing Kubernetes pods, ECS task failures, expiring certificates. Remember a checkout service that Samantha and Aneesh just walked through? We may have averted an incident earlier, but there are still issues degrading checkout stability. It looks like there's a critical issue. So let's deep dive into the investigation.

Here, I can see checkout API pods are getting repeatedly out of memory killed because the memory limit is set too low for peak usage traffic. Each pod restart spikes latency before traffic fills over to healthy pods, making the service less stable over time. So when that memory leak hit earlier, this is exactly what would have made that incident worse. Bits has done the investigation and the fix is ready, increased memory to 1 gigabyte. But safety is paramount for any production change.

Before I approve this Kubernetes patch, how does Bits know when it can and cannot take action? We know this question is top of mind for teams adopting autonomous operations, which is why today, we're introducing Bits Guardrails. Earlier, Aneesh walked us through guardrails in action for the checkout incident. Now Datadog is already embedded in my environment collecting telemetry. But with Guardrails, I can now safely automate fixes for my infrastructure at scale with Bits. Here, I can create a guardrail for any resource, environment, and action such as a Kubernetes patch in this checkout example. I can ensure my teams are notified in Slack whenever Bits makes a change, whether it's auto-approved deployment restarts or a production patch that requires user approval.

And Bits only makes changes within the scope that I've defined. So now I can have Bits patch the deployment to immediately stabilize my cluster and open a Terraform PR, so the change persists in code. And every action is logged with full evidence and outcome. But I'm still in the loop here. How can Bits work autonomously for me? Bits sees that I've approved the same action for the service before. Now I can tell Bits to update my guardrails, so this action is auto approved and Bits will autonomously resolve these issues for me next time.

This is how Bits learns from the changes I've already made in my environment. And now Bits is also available wherever you work, whether that's in the Datadog app, in Slack, or in Claude Code and Codex is powered by the Datadog MCP server. So that's how Bits helps you autonomously close the loop on infrastructure issues before they become an incident.

And now Bit is a true autonomous teammate, detecting earlier, investigating with full context, and remediating safely on your behalf. And so the entire DevOps loop accelerates. What used to take hours now takes minutes. And that's the shift to autonomous operations for your infrastructure. We're excited for you to try this out, find us on the expo floor to learn more. And now I'm going to hand it back to Alexis.

Thanks, Jessica. So we're taking you out of the day-to-day toil that AI can handle, so you can focus on more complex and critical tasks. With autonomous operation, we help you accelerate this loop and free up time for you to build what truly matters. And today, over 3,000 of our customers are already using Bits Investigation and thousands more are using our other AI capabilities like Bits Chat. And among them is Nasdaq. You all know Nasdaq. It's one of the world's largest stock exchanges with massive volume, very high rate of transactions, all of which is business critical to the clients they serve. And so it's my pleasure to welcome Donald Beery, Chief AI Integration Officer at Nasdaq, so they can share with you his experience with Bits AI.

All right. Thanks for having me. I'm here from Nasdaq to share some of our insights on this innovation economy. At Nasdaq, data is at the foundation of how we operate our markets. We serve our clients and deliver trusted intelligence at global scale. As data volumes and complexity continue to grow, our ability to observe, manage and act on that data in real-time is more important than ever. Let's take a step back and look at Nasdaq's business and how we serve our clients.

At the core is Nasdaq's modern intelligence platform built on data. We generate deep gold source data across our markets, index and investment businesses, and then embed that intelligence directly into our solutions. This creates a network effect. It allows us to deliver deep insights and outcomes. As we operate our clients' most critical solutions, we lever this intelligence platform to power their trading, risk management, analytics, and investment decisions.

And all this happens at extraordinary scale from markets generating billions of messages, surveillance systems managing trillions of records, to platforms representing tens of trillions of dollars in assets under management. That scale, combined with trust and intelligence, allows Nasdaq to deliver mission-critical capabilities to clients worldwide. Remember, Nasdaq was the first electronic stock market, and we spent decades in advancing and modernizing our best-in-class trading technology. These platforms serve us, our customers, and economies worldwide. To support these platforms, we've always had logging.

But logs, they're just an archive of history. As you know, Datadog brings us observability, real-time telemetry, and actionable insights, the ability to look at our system real-time and understand what's going on at a transaction level? And since they've added Bits, we have the ability for anyone to interact with their platform in natural language. At Nasdaq, Mario Diaz leads our global derivatives platform. These platforms handle around 100 billion messages a day. Every message is in sequence and without data loss. Mario tells me, if you're not carrying Bits on your phone, you're not managing your systems. Because Bits is integrated into the Datadog platform and lets anyone on our team ask questions in plain English from which servers flow all the way down to where is this transaction getting stuck?

So at Nasdaq, we see markets evolving toward a more connected, always-on, and intelligent financial system, one that moves faster, settles sooner, and operates with greater transparency and, of course, resiliency. But right around the corner, our 24/7 financial systems and tokenization. And these will be the defining features of the innovation economy. And this represents a significant opportunity for Nasdaq and Datadog. Just think about the support that 24/7 platform will demand. It makes that Bits application on Mario's phone even more important.

Again, realizing this potential will require what Nasdaq is renowned for, trusted exchange-grade infrastructure, reliable, available, scalable, and secure, providing global access to connect investors and issuers to new pools of capital. Today, more than 135 markets and regulators work with Nasdaq as a transformation partner. So Nasdaq is well-positioned for this transformation. And with Datadog as our partner, we're excited to accelerate innovation, strengthen our platform and deliver even greater value to our clients. Thank you.

Thanks so much, Donald. We're so happy to be working with you and supporting Nasdaq. Okay. So we're on our way with Bits Ai towards autonomous operations. But to keep pushing towards more autonomy, we need to keep improving our platform because the complexity of software systems and the amount of observability data that's needed are rising exponentially. So we want to share with you our latest innovations to ingest, correlate, analyze, and act on that data, whether it's across your networks, your databases, your logs, whether it's on-premises or in the cloud. And for that, please welcome Angelina to the stage.

Hi, everyone. I'm Angelina, Product Manager here at Datadog. We've seen how AI is helping teams operate faster than ever. And as our systems become more autonomous, some things still remain true: every application, every service, every AI workload, everything still runs on the network. And as our systems become more distributed across containers, clouds, and on-premise data centers, understanding what's happening underneath only gets harder. The problem could be anywhere, a noisy cloud workload, a congested circuit or a misconfigured firewall. To users, it all looks the same, slow applications. The challenge is knowing where to look.

Today, I'll show you how Datadog Network Monitoring helps any engineer pinpoint, understand, and resolve issues across cloud and on-premise environments with confidence. Now let's look at a scenario that will feel very familiar. My team operates a fintech application and at 1:00 p.m. today, purchase app fired an alert for an unusual latency spike. With Network Path in Datadog, I get a hot-by-hot view of exactly how traffic flows from my application to its destination across cloud, on-premise, or hybrid environments. I can drill into every hop along the path, whether it's a transit gateway, VPC, cloud resource, or on-premise device. This means I can quickly see where latency, packet loss, or connectivity issues are being introduced.

And right away, the problematic hop stands out. Ny-edge highlighted in red, introducing 550 milliseconds of latency. On this path, this seems highly unusual. To validate that, I'm going to compare the current route with the path traffic was taking before the incident. And now the problem becomes clear, same source, same destination, but traffic is now taking a different route, which clearly identifies ny-edge as the culprit.

And because my team monitors our on-premise network infrastructure with Datadog network device monitoring, I can dig deeper into that on-premise ny-edge device. Checking the health of connected dependencies takes seconds in case there are also issues downstream, and I can drill straight into ny-edge's interfaces to spot any spikes in errors without wasting time.

Even better, Datadog has already done the hard part for me. It looks like bandwidth utilization on an ny-edge interface dropped from 42% to 5% right at 1:00 p.m., exactly when purchase app latency spiked. And there's more. At 12:58 p.m., someone made a QOS policy config change on ny-edge, and that change ended up incorrectly reclassifying critical traffic, causing packets to be dropped just 2 minutes before the incident. That's our root cause. And this is where Bits can really shine. Bits walks through the evidence step by step and explains exactly what's happening. So even if I'm not a network engineer, I can quickly understand the issue, the impact, and the cause without having to decode complex network behavior myself.

And it doesn't stop at diagnosis. With Datadog, I can roll the device back to its last known good config with a single click. No SSH, no separate tool needed. I can clearly see the current configuration and compare it against Datadog's recommended rollback state and apply that change with confidence. The change is pushed immediately, automatically logged in the network configuration management time line, and we can watch in real-time as latency returns to baselines.

You just saw some exciting new features we're launching today like how with Network Path, you can instantly compare route changes and pinpoint the exact hop causing latency. With Bits and Network Device Monitoring, you can quickly identify the root cause of issues across cloud and on-premise infrastructure in plain English. And with Network Configuration Management, you can roll back config changes and get your environment restored to a healthy state. And there's so much more we're announcing today.

When something goes wrong, Datadog Network Monitoring gives you full stack network operations, taking you quickly from alert to resolution. You can trace traffic across every network hub from applications, to cloud services, to on-premise devices. And you don't have to be a network expert to understand what's happening. With Bits, anyone on the team can pinpoint the root cause. And when it's time to act, Datadog helps you fix issues across every network layer automatically, so your whole team can resolve problems with confidence in a single click, less finger pointing, faster resolution, and more resilient application. But the network is not always the issue. To tell you how to fix another common pain, here is Alex.

Hi. My name is Alex Weisberger, and I'm a senior software engineer here at Datadog. I've definitely had outages caused by network latency. But you know what's been giving me the most problems recently? My databases. Like many of us, that's why I've been using LLMs to help me debug and optimize my slow queries. And they help a lot of the time, but I think we're all seeing how inconsistent they can be. In fact, just the other day, I had to roll back an LLM suggested optimization because it actually regressed performance.

My team and I want to quantify this. So we took the top 500 queries from across our services and asked an LLM to optimize them. We got back around 300 suggestions. 300 is way too many for me to roll out to prod with any confidence because I know that there are outages lurking in many of them. So I would have to spend days running benchmarks to figure out which ones to weed out. So is the LLM saving me time or causing me more work? I need to know which changes I can trust and which are just theories, which is why today, I'm really excited to show you Bits Database Optimization. We built an AI-powered query optimization system where every change gets validated on a simulated database. Let me show you an example.

Here I am in database monitoring, looking at my query metrics. The top query has been a thorn in my side ever since it's been deployed. It has the vast majority of total execution time and block reads on this database, and it keeps timing out. But I see that Bits is saying that the query can be rewritten to be much more efficient. Let's take a look. The current query has an or predicate that references both tables in the left joint, and this is preventing the current indexes from being used. Bits is saying that we can instead union each side of the or together, which should allow those index to be used.

Now this sounds totally believable, but I've seen way simpler optimizations not pan out in practice. So here's where I normally would run a benchmark, but now I have to go through the pain of finding reasonable seed data and interesting params. But below, I see that Bits has already run a performance analysis for me. Let's see how it works. Database monitoring already collects my schema data along with comm statistics that give information about my data distribution. Now this is the cool part. Bit spins up an ephemeral Postgres container and populates it with generated test data matching this distribution. This means I'm getting realistic benchmarking data without touching any sensitive customer data of mine.

On this simulated database, Bits runs my original query against the optimization and measures latency, logical reads, and blocks [ dirtied ]. And Bits doesn't just run the queries once and call it a day. It explores the parameter space to capture a diverse distribution of performance in a variety of scenarios. Here, I see a huge performance improvement across multiple dimensions, including average execution time and P95 logical reads. This gives me a lot more trust in this recommendation because I know it was run on a real database engine.

However, I still want to understand the cause of the performance improvement under the hood. And for that, I need to look at execution plans, which Bits also collects. So by coming up and clicking the compare plans button, I can see the move from sequential scans in the before version to index scans in the optimized version. Now I have the complete picture and understand why performance improved. But we still don't stop there. Bits has already generated a pull request, applying this change against my code base. So whether I'm using an ORM, query builder, or any other SQL generation pattern, Bits Database Optimization has me covered all in one flow from detection, to validation, to deployment.

Back to our case study, we can see how Bits Database Optimization avoids wasted time from LLM hallucination. Instead of you sifting through hundreds of optimization candidates, Bits presents you with validated changes. And we found that this validation harness took those original 500 queries with 300 blind optimization candidates, and produced 30 validated optimizations ready to merge. That's 90% less noise with full confidence in what's left. And we think this is the key to employing AI for query optimizations and making sure your time stays on the work that matters.

To recap, Bits Database Optimization finds optimization candidates with AI. It then filters out invalid optimizations with a validation harness that runs queries on a simulated database. And finally, it opens a PR that deploys -- that applies this optimization to your code base. That's query optimization end to end. Try Bits Database Optimization today and see us on the expo floor to learn more. And now to tell you about our latest innovations in log management, here's Zara.

Hi. I'm Zara, Product Manager here at Datadog. Over the past couple of years, you're innovating faster than ever. And while that's exciting, this has led to your telemetry volumes exploding. This also means that, that telemetry is scattered across more places than ever. It's in Datadog, it's in your data lakes, and it's in your cloud storage. And when your data is scattered across that many places, every question becomes a full day workflow. That slows you down and you end up losing the critical context that you need, which is why today, I am excited to announce Federated Logs, search across your observability data, your data lakes, and your cloud storage, all without leaving the Datadog Log Explorer. Let me show you.

I'm an engineer on the payments team on an e-commerce platform, and I just got an alert that the payments API error rate has jumped from 0.4% to 6% in the last 90 minutes. I start my investigation by checking what type of errors actually coming back. And I can quickly see that almost all of them are FraudCheckTimeOut exceptions. So the payments API isn't just broken. It's actually waiting on that FraudCheck and then giving up. I'm going to capture this as a finding so that I don't lose track of it.

Now because the FraudCheck is timing out, I actually suspect that there's something wrong on the model itself. And this is where my investigation normally breaks. The FraudCheck calls a model which runs in Databricks, and that's where the model's logs are at, too. Normally, I'd have to switch tabs and restart my investigation all over again. And that is assuming that I have access. But now with Federated Logs, I don't have to leave the Log Explore. I can quickly switch the source to Databricks and then to the fraud model logs. I group by confidence score and -- by merchant type. And I can quickly see that the score for gambling has dropped by almost half. That's likely our culprit.

I want to capture this as well. Everything I've captured along the way lives in Findings, a new feature in the Log Explorer, which allows you to save, sort, and analyze your investigations. And now Bits AI can take these findings for further analysis. And as you can quickly see, Bits has not only help me determine the root cause, but business impact and recommended next steps. And that is the power of bringing all of your data together regardless of where it is with Federated Logs and Findings.

But that's not all for when you want to turn unstructured data into AI-ready data sets for platforms such as Databricks and your cloud storage, you can now use Observability Pipelines. Observability Pipelines allows you to standardize in open formats such as OTel and OCSF, enrich with contextual data from ServiceNow, redact PII, and transform as needed. With this, Datadog has you covered. With Federated Logs, Findings, and Observability Pipelines, you've got everything you need to search across your data wherever it lives.

And to do this, we're partnering closely with Databricks to bring all of these experiences natively to you in Datadog. And here to tell you more is the Databricks CEO, Ali Ghodsi.

Datadog and Databricks actually are very similar companies. We both started around the same time. We were multi-cloud companies, and we wanted to give enterprises optionality. These days, the world is changing. We have AI everywhere, and we're seeing, now, massive operational logs, metrics, and traces coming into these systems. And enterprises and companies are facing a really bad trade-off. Either they have to make lots and lots of copies of their data to make things work across all these different systems or they have to silo it, in which case, it's not accessible.

With this partnership, we intend to change that. You can use Datadog and Federated Logs and access the data that's in Databricks in real-time and do your exploratory work right there. You can use Genie and ask questions in natural language from the AI. You can ask questions like what incidents are hurting my revenue, or what endpoints are causing drop in conversion rates, or what latency spikes are impacting the highest revenue customers that we have? Customers are already moving to this architecture from leading AI companies all the way to the most iconic enterprise companies out there. And the Databricks team is, this week, at the Dash Conference, and next week, we'll be at Data and AI Summit. Please come visit us and have a chat.

Thanks, Ali. We're excited about this partnership and doing more together. Moving on, as your teams heavily invest in AI and you're doing business across multiple geographies, many of you are seeing skyrocketing telemetry volumes at petabyte scale. This, along with the compliance requirements across regions, presents many challenges. You need more control and flexibility in terms of how you collect, store, and analyze those data sets. And many of you have asked for Datadog's ease of use and reliability deployed in your own VPCs which is why today, I am excited to announce Bring Your Own Cloud for logs, metrics, and traces.

Your data stays in your environment and everything else is exactly how you know it, the same dashboards, the same alerts, the same Bits AI, and the same workflows that you already know. Let's take a look at how it works. Lately, I've been seeing intermittent errors of my authentication logs. And just like searching Databricks, I can search the cluster from the Log Explorer and filter to those time outs. And because this is Datadog, everything is connected. I can pivot from the log, to the trace, and then to the host metrics. And you're probably thinking what's different here? And that is exactly the point. It's the same Log Explorer and the same easy experience, but running in your cloud and in any region.

So wherever your logs live, Datadog has a pad, standard indexing and flex for the best-in-class SaaS log management. Federated logs and Observability Pipelines are there for when you have data which lives in other systems and Bring Your Own Cloud for petabyte scale or data residency requirements, one platform, one experience wherever your data lives, and we're excited to get this into your hands. Thank you. And back to Alexis.

Thanks, Zara. That was very cool. So let me summarize what you just saw. You can understand and fix issues with Network Monitoring even if you're not a deep networking expert. You can optimize database queries with Bits Database Optimizations with the confidence that you're not in your low mode. You can also search your data wherever it is with Federated Logs, whether it's a Datadog or other observability providers, data lakes, or cloud storage. And finally, you can extend the data platform in cases where the data just can't leave your own environment with Bring Your Own Cloud.

And there's one more thing on the topic I would -- I'd like to share with you. It's about metrics. Metrics intake has climbed relentlessly over the past decade. You have kept us very busy. And with that growth comes more questions you want to answer. You want to break down metrics by container, endpoint, by user, by version, by model, many more dimensions. And the result is that cardinality has been outpacing data volume by a wide margin, and that's the problem but one we've been hard at work to solve for you. So today, I'm happy to introduce Infinite Cardinality Metrics.

Infinite Cardinality Metrics is built for agentic exploration and querying, gives you the freedom to capture every attribute and dimension, the ones that matter, no matter how high the cardinality. And it's priced by metric [ name ], and it scales with your data volume, not cardinality. So it gives you predictability. If that sounds like something you want to use, please talk to your Datadog account team to give it a try.

Okay. Back on track now. You're all working hard to build and run great software products, but that's not the end goal. That's just a means. The goal is to delight your customers. And that means deeply understanding user experience, which may include millions of users or even billions of users, all using different devices, application versions, networks, et cetera. It's not easy. And if that's a problem that resonates with you, you're going to like what the next speaker has to show you. Let's hear from Mael.

Thanks, Alexis, and hi, everyone. I'm Mael, and I'm a product manager at Datadog. There is something we all fear, 5,000 of votes, 700 comments. Somewhere on social media, a user is selling the world, your checkout is broken again. And you are hearing about it hours after the fact, SLOs are green, monitors are quiet, services are up and yet, here we are. Where do you even start looking front end, back end, UX? Today, you check each layer of your stack and each one might just look fine, but users are still reporting issues.

That's the gap. Each component in your stack is only one piece of the puzzle. It cannot tell you the full story on its own. The only way to know whether users can actually complete the jobs of your application is to monitor journeys end-to-end and a journey at its core is just 2 things: an intent, how users target like opening the log in stream, and an outcome when they succeed and are redirected to their profile. Simple in concept, right? But we've heard you tell us a struggle it was to instrument a journey with all the possible steps and variations, and how time consuming it was to do this for all the journeys you had to track, and how frustrating it was to see the instrumentation you wrote going stale the day after with the next release.

And even when you managed to keep up, you still had 3 questions that felt surprisingly hard to answer. What journeys are my users actually taking? How healthy are they like right now? And when something breaks, what changed? Today, we're closing that gap with Datadog Journey Monitoring. Let me show you how an ops team would use it. Journey Monitoring brings REM, synthetics, and product analytics into one new experience. And here is our journey map that shows every user journey listed on a single canvas.

Log in, search, add to cart, checkout. These are neither services nor endpoints, but the things users actually do. Take log in, for instance. As we mentioned, each journey is just 2 pieces, the start when users reach our log-in page,, and the exits when they redirected home. How healthy is it right now? Well, at a glance, we see how many times users try logging in, the overall conversion rates, and uptime from synthetics looking good across the board. We even get context from other journeys upstream in the flow, telling us where users are coming from and downstream, where they're navigating next, like searching for a product in our catalog.

But log-in is one journey we already track. The harder problem is keeping up with everything our team ships, new features, new flows, every sprint. And that's why Datadog continuously scans real traffic in the background and surfaces journeys nobody configured like this one, where users add to their card some product or AI chatbot just recommended. We did not instrument it. We didn't ask Datadog anything. It just automatically appeared and it's not looking good. Conversion is trending down. Fewer users are completing the flow than before. SLOs covering the journey are breached. Datadog is telling us we have technical ARs interrupting our existing tests and real users alike. And because this journey is failing, we can even see the traffic on downstream journeys taking a hit too. So what changed? The answer is just one click away.

Bits AI has already kicked off an investigation. Let me open it. And Bits AI found the root cause, the recommendation cards are timing out. It turns out a recent model update pushed latency past our 3 seconds cutoff. So most users actually never see the recommendation. Normally, this would be hours of cross-team triage, product sees a conversion drop, engineering sees a healthy service. It takes a war room to connect them here. They're both looking at the same investigation. We're all back to the previous model as a quick fix. Conversion recovers and the journey goes back to green. That whole loop from business impact to technical root cause to fix happened on one surface.

Now let me sum up because Journey Monitoring brings our products together, you always know what journeys your users take. Datadog put them on the map automatically when your team ship those new features, even the AI flows that are so hard to track. You always know how healthy they are right now and both from a business and technical standpoint. And when something breaks, you always know why, because Journey Monitoring sits inside Datadog alongside your backend telemetry, Bits AI can trace the root cause across the full stack. Engineering and product are finally looking at the same story.

That's Datadog Journey Monitoring and know you can use it, too. We can't wait to see what you build. And with this, I'll hand it back to Alexis.

Thanks, Mael. Okay. So up to this point, we've been talking about the ops loop. But now I'm going to switch focus to the developer side. As you've heard dev, you too are going through your own loop, write code, ship it, test and evaluate the results. And with AI, the speed of development is accelerating, but not evenly. For more on that, let's hear from Ruxanda.

Hi. I'm Ruxanda Lueck, Group Product Manager at Datadog. In 6 months, AI attributed commits have gone from almost nothing to millions, and it's accelerating. In the dev loop, AI is changing not just how we write code, but how we safely release it. AI agents compress the time from idea to PR. But trust doesn't accelerate automatically. The bottleneck shifts downstream to reviewing, releasing, and evaluating it safely. How do we reconcile the speed of creation with the speed of trust? At its core, you want to know, is the code working as intended? To answer this question, today, we are previewing the new Bits AI capabilities for developers.

Bits AI follows every code change and validates its behavior in staging, monitors its rollout to production, catches regressions and fixes them, and establishes production guardrails. Basically, Bits AI safely accelerates the entire dev loop. Let's see how it works. I'm a product engineer working on Couch Cache, an e-commerce site for finding the perfect couch. Using AI coding agents, I created an AI shopping assistant. The customer opens the chat, they describe what they want, and they see personalized recommendations. I'm pretty happy with the feature. So I open a poll request and ask for reviews.

What can go wrong? I know, famous last words. So I start doubting myself, what if the coding agent missed something? What failure modes it didn't consider? This is when Bits steps in. When a PR is open, it analyzes its expected behavior and builds a validation plan across staging and fraud. Bits just validated my PR in staging. I want to learn more about it. Today, we are previewing Bits Release as part of the new capabilities Bits has for developers. This is the Bits Release page, which summarizes my PR progress. Instead of bouncing between GitHub, my CD provider, feature flags, dashboards, and Slack to figure out the state of my change, Bits Release brings it all into one place.

At the top left, I can see what Bits automatically detected about my PR's intent and expected behavior. On the left-hand side, I see the validation plan that's going to focus on 4 criteria: correctness, error rate, performance, and conversion. At the top, I can follow the progress of my feature from staging to production. For my AI shopping assistant, the first thing Bits checks is correctness. Does the assistant return relevant recommendations or not? To answer this question, Bits uses its new testing capabilities to create an autonomous goal-driven test that navigates the UI just like a real user would.

Today, we are previewing Bits Testing as part of the new capabilities Bits has for developers. Bits Testing opens the staging gap and tries to see if we can achieve this goal of my feature without following a predefined set of steps. Instead, it dynamically reacts to the responses of the AI shopping assistant. To achieve this goal, Bits is exploring all possible pathways in the application and autonomously develops very complex testing scenarios.

Next, I see that there are no back-end or front-end errors, latency looks stable as well. Check-in conversion will have to wait until the feature reaches production. Across the board, this looks great. So I merged the PR. The feature is behind a Datadog feature flag, which Bits can enable to 5% of users in prod. This has been a pretty productive morning so far. However, in a dramatic twist, Bits detects an issue in production. So what happened? Bits we ran the same autonomous test, which was validated in staging, but this time against production. And I see now that the AI shopping assistant does not return any recommendations at all. At the same time, Bits has been investigating the issue, and I am curious to learn more about what's going on.

Bits found that the root cause is a missing IM role and a PR fix is ready to go, which I merged right away. Okay. Now I'm done for the day. Later, I do get confirmation that everything is working as expected. I can see my conversion rate in production, and people are moving through the funnel, and total checkouts are trending up. Finally, Bits converts the temporary test it created for this PR into a persistent test powered by Bits Testing. This is the Bits Testing page, which displays all of the runs of my autonomous test. This isn't just a one-off test. Bits is going to monitor this going forward. I want to take a look and see how Bits does this.

The goal of the autonomous test is to continuously validate in production that the chat still returns relevant recommendations. Moving forward, Bits Testing will continuously run in the background to try to achieve this goal. This is how Bits Testing catches the edge cases as your product changes over time. So validation doesn't stop at the rollout. It becomes a permanent guardrail for your features.

To recap, when AI accelerates co-creation, trust becomes the bottleneck. Bits AI for developers closes that gap through 1 connected workflow. Bits Release creates and runs validations plans across staging and fraud. Bits Testing autonomously identifies and executes complex testing scenarios. And finally, when something breaks, and something always breaks, Bits Code generates the fix. The results team move faster through the entire dev loop from idea, to PR, to trusted release. We're excited for you to try the new Bits can do for developers. Find out on the expo floor to learn more. And now I will hand it back to Alexis.

Thank you, Ruxanda. So with Bits AI, we're making the dev loop tighter and faster. And really, the dev loop and the ops loop in the end that both need to speed up. And as we deliver on both of these loops, we help you with what you want to do, moving faster from development to production, from idea to outcome. Wherever it makes sense, we're building AI for Datadog, all in service of making Datadog more helpful to you. But that's not all. We have a lot more to show you. And for that, I'd like to welcome Yanbing to the stage.

Thank you, Alexis. Good morning, everyone. I'm Yanbing. This is my second Dash, and I couldn't be more excited to be here with all of you. What you've just heard from Alexis and the team is how we're using AI to help you accelerate your software development and your operations loop. All these Bits AI capability are production-grade AI that we build to help you with your production system. And we've learned some important lessons as we build them.

For example, AI can look amazing until it hits production. Before you take an AI system into production, you need to answer these questions that matter. First, does it work? Not just in the demo, not just in a benchmark, but consistently, reliably in production for real users. Secondly, is it safe and secure? Can you trust it? Can you govern it? We all know AI is creating new attack surfaces from prompts to tools, and data, and code. Third, does it have the right economics? Can you scale it in a cost-effective fashion?

We've all heard the horror stories. Teams exhaust their full year AI budget within just a matter of months. This happens at Datadog. And also agent may get stuck in a loop. Suddenly, the experiment becomes a 6-figure bill. It is time to move beyond tokenmaxxing into valuemaxxing. And to answer these questions with confidence, it requires full visibility into your entire AI stack. And the time to take action is now. It's no longer optional. You need to observe it, you need to secure it more, not less, than before. And we have seen this shift happening in our customers who's been adopting the LLM observability product from Datadog.

In the past 12 months alone, we've seen a 30x increase in the number of LLM spans sent to Datadog. These organizations are moving AI into production and critical workflows such as software development, customer experience, and critical business operations. And they do require more observability and more security, and that is what we are building for, because AI introduces an entirely new stack. At the foundation layer is infrastructure where precious resources like GPU powers your training and inference. And there is a data layer that fuse training and real-time decision-making. Then comes the model layer where intelligence is created and refined and applied. And finally, at the agent layer, where AI comes to life in products, applications, and workflows.

Every layer generates new telemetry, new risks, and new way for you to get paged at 3:00 a.m. This is why the second part of the keynote is all about Datadog for AI, how we are helping you observe, secure, and improve your entire AI stack end-to-end. How about we start with the data layer? Please join me in welcoming Harel to stage.

Thanks, Yanbing, and hi, everyone. I'm Harel, an engineering manager here at Datadog. Over the past year, something has changed with the way people query their data, 9 out of 10 organizations now use AI directly on their data warehouse, including us at Datadog. But your data always has issues. Pipelines can break, schemas can change, and data becomes stale. And now with AI on top of it, garbage in, garbage out turns into a blazing dumpster fire. And that's why last year, we released Data Observability, and it's now generally available. Let me show you how it helps.

You can now track every pipeline, every run, and every failure, all in one place across Databricks, Snowflake, Airflow, and many more integrations. Datadog will also find anomalies in freshness, volume, schema, and value distribution so that you can catch the bad data before the business does. And with end-to-end lineage from ingestion, through transformation, all the way to your BI, so you always know what's happening upstream and what's downstream. And when something does break, Datadog clearly points out what's wrong.

We proactively diagnose issues, provide optimizations to your pipeline, and we even draft the fix, so you can apply it directly from Datadog. So now you can trust your data end-to-end using lineage, quality monitoring, and jobs monitoring. But having trusted data is not enough anymore because now we have a new challenge: all those AI agents that most of you are using to query your data, they don't have access to shared definitions. They don't have visibility on pipelines. They don't have any data governance. And as a result, you get AI slop at scale. Let me show you what I mean.

You have a review meeting in 1 hour, and you need to know your revenue by marketing channel in Q2. So you ask your agent. The agent finds a revenue table, runs a query, and returns a number that looks fine. So you walk into that meeting, but only later, you find out that number actually included unprocessed returns and test transactions, wrong number looking just right enough to go unnoticed. When you're using AI agents with your data, some answers are right and some are not. So it's possible that right now, in your company, an agent is handing someone the wrong number. And that's why we built Bits Data Analysis. I'm really excited to show you how it works.

Now because we built Data Observability, Bits now has access to your data warehouse, your queries, your data pipelines, it can understand how you model your business. Now some BI agents might have access to that. But Datadog has a much wider context, your applications, your source code, your product analytics data, and that makes a huge difference. Okay, enough talking. Let's open up Bits Data Analysis, ask that question again and see what it can do. I'll ask, "What was our revenue by channel in Q2?" The agent grounds the answer in the data context, automatically built from everything Datadog already knows about your data, lineage, quality, business definitions, not just exposed tables. It reasons about all this data in one chain, excluding the unprocessed returns and the test transactions, and it confirms the source table is fresh. I'm asking the same question, but this time, I'm getting the correct answer.

Now with Bits, I can push further. I noticed a drop in revenue last week, but I don't know why. I'll ask, "Tuesday afternoon, revenue dipped. Is that another data issue?" Bits starts by confirming the data is clean using Data Observability. The dip is real. But then it does something no other agent can. It looks at the checkout flow using product analytics data and identifies a drop-off in conversions on the checkout page. Then because it knows which service is powering that checkout page, it can access APM data, and it pulls telemetry for that checkout service and identifies that checkout service latency went up from 200 milliseconds to 4 seconds at 4:00 p.m. on Tuesday. Finally, it overlays deployment events and ties that spike to a specific checkout that happened 5 minutes earlier, slow checkout, drop conversions, real revenue impact.

And that's great because now I can go from a vague business question to a specific deployment in a single conversation. But even with trusted data, agents still produce the wrong answer sometimes because some context only lives in people's heads, and that's where you can step into help. With Bits, data teams can be part of the feedback loop. You can observe and improve quality by updating the business context Bits has, for example, excluding test transactions from revenue queries. Your company is now curating context and continuously improving how your business answers questions. Bits will keep evaluating the reasoning chain and not just the final answer. So data teams can measure quality and improve it. And since you own the context, you're now using the correct revenue definition and your agents will stay honest about your data.

Okay. Let me recap what we just saw here. As you keep using AI in your data, make sure your foundations are solid, and that's where Data Observability helps with lineage, jobs monitoring, and quality monitoring, and it's now generally available. And as you continue asking agents questions about your data, we're really excited for you to try out Bits Data Analysis. It grounds answers across all of Datadog. It evaluates the reasoning chain and not just the final answer and allows you to curate business context so that you get the right answers. So now that your agents have the right context, we also want to make sure that you're using agents efficiently in your organization. To tell you more, here's Matt.

Thanks, Carl. Hi. My name is Matt Lee, and I'm an engineer here at Datadog. Over the past year, developers everywhere have been rapidly adopting coding agents from Claude Code to Cursor to Codex and more. And we've seen examples of developers using these tools to great effect, identifying and addressing previously uncovered security of vulnerabilities, and accelerating development cycles. But broad visibility remains a problem.

As we adopt these new tools, 3 key questions remain top of mind. First, what agents are being used in my organization? Second, how much is it costing us? And third, can I help my organization use these agents effectively? To answer these questions, we built the Agent Console. The Agent Console is a single pane of glass experience that now gives teams visibility into their organization's usage of coding agents. Let me show you how I answer the first 2 questions.

Here, in the Agent Console, I see aggregate details for Claude, Codex, and other coding agents like total spend, users over time and number of sessions. I can break all of this data down by team, user, repository or coding agent. I can also look into details like the number of AI-assisted commits and AI-assisted PRs. All of this helps me understand how my team is tracking over time. But that's not all. See details of my own coding agent usage, I can drill down in the analytics view. On this tab, I can look into details of coding agent usage in my teams. And in the user detail tab, I can map my agent usage and associated costs to my PRs. And if I want to understand AI usage at the organizational level, this scatter plot helps me identify different user cohorts.

For example, the cohort in the bottom right may have insights to share with the rest of the organization about using coding agents efficiently and the cohort in the top left may be using coding agents for tasks other than coding and hopefully aren't simply tokenmaxxing worth a follow-up just in case. And to make sure we avoid runaway costs, I can set up alerts on my team's usage, helping us to not spend our entire 2026 AI budget in 1 month.

With these views, the Agent Console gives me a full inventory of my team's agent usage footprint, which agents are being used for what purposes, by whom and how much it's costing? This is great, but as a developer, I would really like to be able to help my team use these agents more efficiently, not just look at cost and usage dashboards alone. Let's take a look at how I do that directly in the Agent Console. In this page, I'm seeing detected problems in my coding agents because I have set up to detect via metrics, patterns of agent behavior. I can see how these detected patterns map to coding agent sessions on the left and across repositories in my code base on the right. So let me click and see what the issue is.

I've specified a marker to detect the pattern of behavior where my coding agents try to commit or a PR code without testing first, which we're calling skip checks. Looking specifically at sessions in my team's Helix API repository, I can see that my teammates have run into the same issue. So how can we fix this? The Agent Console enables me to see a collection of fixes authored by teams across my organization to address this agent behavior. I can explore these fixes, understand how they work, see which agents they impact, and decide which fixes I'd like to propose to my team. This fix takes the form of a pre-tool use agent hook to remind the agent to test code before committing. And the fix below is an agent skill for testing.

And if I want, I can also author my own fix by clicking this button. Once I'm happy with the fixes I've selected, it's just one click to create a PR with my selected fixes to my team's repository where they will take effect once merged for future agent sessions. And that's how the Agent Console can help us make our teams more effective and impactful with agents, along with helping us understand more about our AI usage and spend. We're just getting started. We'll be expanding to business agents like Claude Desktop and ChatGPT next to ensure you have full visibility into your agent stack. We're excited for you to try out Agent console, sign up today and learn more by visiting us on the expo floor. Now I'll hand it back to Yanbing.

Thank you, Matt. I love how Agent Console is providing the needed visibility and governance in how we can help you use AI efficiently and productively. Few companies understand the challenge of running AI at scale that Modal Labs. Modal is at the forefront of the AI infrastructure, helping customers scale, training, and inference without any operational complexity. To put it plainly, they know how to make GPUs behave. So it's my honor to welcome Akshat Bubna, CTO and Co-Founder from Modal to the stage to tell us how they're doing it. Welcome, Akshat.

Thank you. It's great to be here. I'm Akshat, Co-Founder and CTO of Modal. When we founded Modal 5 years ago, this is before the whole explosion in Gen AI, we set out to build infrastructure designed from the ground up for how data, AI, and ML applications really behave. We wanted to build infra that scales up to thousands of GPUs with a developer experience that feels like running code on your own machine. And it turns out the bet landed at the right moment because the world needed a lot of GPUs and CPUs. And over time, Modal became a whole platform of primitives for the model life cycle from inference, to training, to batch processing, and the agent stack, which is sandboxes and persistent file systems for agents.

And this means we run the infrastructure at significant scale. We run hundreds of thousands of containers across 17 cloud providers, and we move more than 100 petabytes of data every month. And we do this to power workloads for companies like Suno, Cognition, Ramp, and DoorDash. And as we built Modal, we realize that AI workloads break our mental model of observability. This is because requests last milliseconds. There are millions of them every second. And each one gets sent to a different container, on a different worker, to a different cloud, by a different proxy and so on. This means that there's a combinatorial explosion of tags associated with metrics.

And this is where the new Infinite Cardinality Metrics from Datadog have been game-changing for us. Before this, it wasn't really feasible to do monitoring at our scale, at least in a granular way. And beyond this, Datadog has been powerful for us in a few different ways. Now first of all, it's been truly a single pane of glass for us. We chase GPU capacity wherever it lives. So Datadog unifies our 17-plus cloud providers. And this means we get infra traces and logs, so our engineers are not context switching between consoles.

Second, Datadog Network Monitoring, which you heard about earlier in the keynote, matters a lot for us. Given how much data we move, which is 100 petabytes a month, it's really important for us to have a deep understanding of where the flows in our network are and how much they're costing us. And third, we use Datadog MCP really heavily. A surprising amount of internal SRE work for us now runs through agents plus Datadog MCP. It's actually all you need and it's changing how our team operates.

And finally, the Modal integration with Datadog has been really valuable for our joint customers like Reducto and Substack because they use it day in and day out to see how their model workloads are working in their own Datadog instance. So if you're building seriously on AI today, you're probably running fundamentally different infrastructure than you were 2 years ago. And it's not just about having the most compute. I think having visibility into your compute and your AI workflows is more important than ever. This is why we're excited to continue to partner with Datadog. Thank you for having me.

Thank you, Akshat. What Modal is building is a great example of what's required to run AI at scale, highly efficient infrastructure, optimized GPU, and also unified visibility across their 17 and growing cloud environment. We're proud to partner with Modal.

We have talked about data and infrastructure. Let's move up the stack to agents and models. How do we make sure models and agents are working well in production? How do we know they're generating the right business outcome? And how do we know they're making the particular type of decision they're making. This is where Agent Observability becomes essential. Let's find out more from Charles.

Thanks, Yanbing, and hi, everyone. I'm Charles, PM at Datadog. We're all racing to ship agents into our products. But the definition of an agent has changed. It started simple. You take an LLM, you give it an input, generates an output. But now agents are autonomous loops. They can use skills and tools and even take actions on your behalf before answering. And your agents have evolved, and so did we. We expanded the scope of LLMs to cover the full agent observability.

Now by nature, building an agent is hard. Users can ask them anything and agents to decide what they do when they want to do it. This means shipping an agent that you can trust requires a continuous workflow: develop, monitor, and iterate. To help you prove that your agent works before you ship it, we built annotation, data set, and experiment. And to see how it behaves in prod, we built authentic tracing and online evals. But the reality of it is that it is impossible to cover all of your users' behaviors and all of your agent behaviors. And so the right mindset becomes being able to answer these 2 questions really quickly. What should I improve first? And how do I fix it fast? So let me show you how Agent Observability answers these questions.

The team at TaskCruncher built an awesome AI assistant. We have tracing in place. We have a golden data set with realistic scenarios, and our evals are green. But this morning, I woke up to a nightmare. We had a $50,000 budget, and we've already blown it. I hope my boss hasn't seen it yet. Okay, okay, okay. So now I need to fix that and fast. But nothing has changed. I haven't updated the agent in weeks and traffic has been stable. So where should I look? To answer this question, we build Patterns. It auto clusters your traffic, surfaces what's new, and highlights what matters. So let's see it in action.

Patterns classified 15,000 traces into the 6 topics I was already tracking like task creation and status updates, and it surfaced 3 new ones in yellow. And I need to know where that cost is coming from. So I've set the graph to show volume versus cost and coordination is way above everything else. It's a new user behavior, and it's $20,000. This is bad. But at least I know now where it's coming from. So let's dive deeper. The description tells me that users are asking the agent to coordinate work across multiple projects. And this is not something that we've designed or tested the agent to do. And looking down, list task is called 15x more than any other tool.

It's weird, like the agent shouldn't be doing that. So what's wrong with my agent? Okay. So Patterns has now helped me figure out where I should look coordination. But I'm not done yet. I need to understand why it's happening and I need to fix it fast. And normally, there would be a painfully slow process. You retraces, make a hypothesis, update your prompt, update your data set, and finally test it with an experiment. It is at least a day of work, and I don't have that time. My boss is there. That's why I'm happy to announce Bits Evals, the agent that debugs your agent.

Now going back into Datadog, I can see that Bits has already analyzed the cluster and has found the root cause. This system is looping. It calls list task over and over again instead of fetching all that data in one go. So Bits fix it. In one click, Bits update the prompt. That's easy, right? But if I test this new prompt on my current data set, I will learn nothing. We got here in the first place because my golden data set didn't know about the coordination behavior. And this is why we designed Bits to flag and fix data set gaps.

So I have it at traces from the cluster. Perfect. Now the golden data set actually has what users are actually doing. I opened the PR. Because my experiment harness is already wired into CI, the agent automatically runs against every single topic in the golden data set. And now the experiment is complete. So let's see the results in Agent Observability. Now that's my type of screen. The new configs the baseline across every single evaluator. Costs dropped by 20%, toll calls are down and quality holds. That means we got rid of that looping and we're good to go.

So we went from that crazy $50,000 bill to fix in minutes. Patterns revealed our blind spot, Bits generated fix, and the experiment proved it. This is Agent Observability, one loop to ship reliable agents fast. And now back to Yanbing.

Thank you, Charles. What Harel, Matt and Charles showed you are just some of our latest innovations, and they're part of something much bigger. We have built a full set of capabilities to help you observe and improve every layer of your AI stack from GPU monitoring, to full lineage Data Observability, to Agent Console, and also Agent Observability. Because running AI in production isn't just about optimizing one component. It's about having everything work well together. And that is the power of full stack visibility.

But we also know full stack visibility isn't just essential for observability. It's also the foundation of security. We often say you can't secure what you can't observe. When observability and security come together with shared data, shared contact on a unified platform, magic can happen. One of our customers have just done that. Let me welcome Jason Taylor, Head of Cybersecurity at Arc XP, the tech platform at the Washington Post.

Hi. I'm Jason Taylor, Head of Cybersecurity at Arc XP. The Washington Post built Arc XP to solve its own publishing challenges. We evolved that into a platform that runs more than 2,500 sites for hundreds of media customers, serving billions of unique page views per month. A secure and stable Arc XP is more than an engineering responsibility. We are safeguarding the reputation of one of journalism's most recognized institutions and the credibility of every newsroom we serve.

Arc XP powers some of the world's biggest newsrooms. So an outage isn't just downtime. It stops the flow of news. That's why we stopped treating platform reliability and security as separate problems. There are 2 sides of the same customer promise. Every newsroom built on Arc XP can publish safely around the clock. This promise motivated us to consolidate on Datadog as a single platform for observability, security, and incident response. With Datadog, every team gained a real-time view of platform health and a single workflow to manage incidents. Complex investigations shrank from hours to minutes. Customer impacting incidents dropped 86% year-over-year.

The hardest part of running security isn't finding the issue, it's getting it fixed. So we built our security stack on the platform that our engineers were already using. Security, SRE, and engineering now work from the same time line. Once that foundation was in place, we layered on an AI SOC analyst. Today, it performs the initial investigation on every alert and resolves 89% autonomously. We're adapting to a world where the time to discover and exploit a vulnerability is shrinking to minutes. A consolidated platform is the foundation and AI is the force multiplier. The capacity we gained is what let us ship, pass the news, and our agentic Editor and Composer products. These are AI experiences that bring agents and humans to journalists together to tell richer stories with Agent Observability watching model behavior.

Our engineers now spend their time building the future of journalism, not managing the tools that protect it. The next chapter of media will be defined by how quickly publishers adopt AI safely, at scale, and with trust intact. We intend to lead it, and we intend to do it with a partner whose pace of innovation matches our own. Thank you.

Thanks, Jason. Hi, everyone. I'm Tim Knudsen. I lead our security products here at Datadog. We're proud to partner with Arc XP, the Washington Post, as they build the future of digital media platforms. What Jason described is a powerful example of what becomes possible when observability and security work together on a unified platform. And that brings us to our next topic, security. 2.3 years, that was the time it took an attacker to weaponize a security flow. With AI, it's a day or less.

Why does this matter to you? Congratulations. You now have 3 new challenges on your list. One, AI agents. They've created a new way for attackers to get in, prompts, data, the developer tools that are being used, coding agents. How do I secure them? Two, the vulnerability backlog continues to grow uncontrollably. But which of these actually matters the most? Three, threats unfold at machine speed. Investigations still take hours. How do I defend against AI scale attacks? Let me show you how Datadog security can help with all 3.

First, let's talk about securing AI agents. When an agent gains access to your company's private data, a single malicious prompt from an untrusted source can turn the agent into a target for leaking sensitive data, credentials, source code costing you millions, reputational damage, or data loss. This is known as the lethal trifecta attack. To help you defend against this, I'm excited to announce Datadog AI Guard for Custom Agents. Now available, AI Guard secures your homegrown agents with runtime protections.

See, custom agents, they interact with a wide variety of models, tools, data sources. What AI Guard does is it integrates directly into that agent flow. It intercepts every agent interaction. It applies AI, behavioral analysis to detect and block attacks like malicious instructions sitting inside of an API payload and other data sources. So the attackers can't exploit that agent. They can't exfiltrate data. So let's take a look at AI Guard in action. Let's look at an example of a fintech company that's building an autonomous expense review agent.

In AI Guard, the first thing I see is a prioritized list of security signals. So I know the security threats that matter the most. At the top of the list, we see AI Guard automatically blocked a high-severity prompt injection attempt. AI Guard detected the attacker had malicious instructions in a receipt in an attempt to exfiltrate data. But thanks to AI Guard, this was blocked in real-time. But how did AI Guard figure this out? In this view, we see how AI Guard is inspecting every interaction, both the input and the output. The expense agent loads the latest expense policy, looks at pending expenses, and finally, processes the receipt. But when I look at the receipt, it looks harmless, but this attack was a bit sneaky.

AI Guard identified payment instructions were different than usual. The receipt included instructions to override the default model behavior and send 90 days of transaction data and employee account numbers to an attacker controlled e-mail. But in this case, the attack was blocked, all thanks to AI Guard runtime protections. Now let's talk about a surface that has grown, or may say spread exponentially over the last year, the developers' coding agents. Coding agents like Claude Code, Cursor, Codex, they now execute source call skills, call MCP servers right inside your SDLC. That puts the developers in the blast radius for the supply chain vector, and this is creating a lot of nervousness amongst your security teams.

So to keep you moving fast with the right security guardrails in place, we're introducing AI Guard for Coding Agents. Let's look at how this works in practice with AI Guard hooked into Claude Code. Here inside of Claude Code, I'm running a skill I downloaded in my teams repo. It's an innocent-looking poll request summary skill. But hidden inside is a dynamic context command that runs before Claude even sees the skill, which could lead to critical supply chain compromise. The skill is designed to grab my GitHub token and steal it, but AI Guard intercepts and blocks this command before the token even leaves the machine, simple as that.

Okay. So we've secured a known agent. But to protect our full environment, we need to see every agent, and that's where AI Guard Agent Discovery comes in. It maps every known agent in your environment. So you have full lineage visibility across model endpoints, data sources, dependencies, and what is protected and what is not. In the Discovery view, I can see both protected and unprotected agents. Here, you see some of the agents are calling it unsanctioned DeepSeek model endpoint. Let's click here to see which ones. It looks like it's calling a set of DeepSeek models that we haven't allowed in our organization. Even worse, AI Guard is showing threat signals on this AI service. This is telling me it's actively under attack. It also has access to sensitive data, increasing the severity of the situation. With AI Guard, I can simply click, protect this agent, and remove the risk.

All right. Next up, let's talk about challenge 2. What vulnerabilities matter the most? Even with strong prevention, vulnerabilities are going to end up in your run time. And when that happens, the backlog explodes. Everyone here who's been handed a list of 1,000 CVEs told to go fix them knows exactly what I'm talking about. The problem isn't about finding more. It's about figuring out which ones matter the most. And to solve this, we need runtime behavior, reachability, business impact, ownership, all in one engine. That's why we're introducing the Runtime Prioritization Engine. Let's see it in action. Here is the cloud security summary page. 16,000 CVEs in my backlog, which I'm not going to get to. But the prioritization engine has already isolated down and reduced the noise and automatically found those that are the critical CVEs that have the high risk of being exploited are actually running -- and actually running on my crown jewels.

This is Datadog's real-time visibility in action. This is observability and security together. This is the Datadog difference. As a result, 16,000 becomes 9. Let me click in 1 and let's learn more. Let's see how the decision was made and let's see what else is provided to help speed remediation. Here, we see the Spring4Shell CVE. It's been identified as critical severity with very high exploit probability. Here, I see the live ownership details, straight from infrastructure telemetry, not from some stale codeowners file. Here, I see the shipping service has been flagged as a crown jewel. And this is not because somebody tagged it 6 months ago. It's because live APM traces are showing a handling sense of data right now. Let's look at the reasoning, production resource, exploit available, high exploitation risk. Based on this reasoning, it's super clear. 16,000 to 9. This is how Datadog ensures you're spending your time where it matters the most.

Okay. Let's talk about challenge 3. How do I defend against AI scale attacks? Let's be realistic. Even with prevention and prioritization done right, attacks will happen. And we all know that dreaded feeling of getting the call at 2:00 a.m. be pulled into a security incident. With agentic power attacks, dramatic increase in volumes, it's only going to get worse. That's why Bits Security Analyst, autonomous investigations help you battle AI with AI.

Today, it's generally available with the Datadog Cloud SIEM. But now I'm excited to announce Bits Security Analyst runs on top of non-Datadog SIEMs. So you can get the benefits of the Datadog Agentic SOC and start helping you today regardless of what SIEM you're using. Here's an example of it in action. An IM signal just fired. An AWS admin access policy attached to developer user account was detected. Bits pulls the user activity, checks whether the workflow looks console-driven, validates geographic consistency, and weighs this against other concerting signals, residential IP, high privilege grant, missing MFA. It generates hypothesis as if it were a senior analyst. Is this routine admin workflow? Is this a compromised credential? Is this an insider? Every step of the reasoning is visible. It's not a black box.

And here's the verdict, high confidence, benign. Reasoning, evidence, supportive signals, all there. Now imagine it's running on every one of your security signals across all your queues at machine speed. So let's recap. We started today with 3 new challenges on your list. AI agents are new way in. Now you have runtime protections. The backlog is still growing faster than you can fix, but now you only have to chase those that matter the most. AI scale attacks outpace your queue. Guess what? Bits AI Security Analyst is there to help you battle AI with AI. Three challenges, 3 solutions, so you can stay focused on shipping.

But there's much more we're announcing today across our security portfolio, including Bits AI Threat Hunter, API testing for app and API protection, as well as AI native SaaS. So come check out in the security zone. We look forward to seeing you. And with that, let me hand it back over to Yanbing.

Thank you, Tim, and thank you for sticking around with us. As you have seen, we are building a lot. We call all of this Datadog for AI, the end-to-end stack. that was observability security to help you have the confidence that when you launch your AI into production, you know it's performing, it's secure. And most importantly, you have full control and visibility into their performance and cost.

So let me recap the entire keynote. Alexis started with AI for Datadog. We talk about how we're using AI to help you autonomously detect, investigate, and remediate issues in your operations loop. We're shifting left to take that automation also into your software development loop with this code testing release, so you are shipping code at AI speed. We're also investing in our core platform with new capabilities like federated search, Bring Your Own Cloud, and Network Monitoring, Database Optimizations, and Journey Monitoring. Also, let's not forget, for years, there was a 4-letter word in observability called cardinality until we decided to put Infinite before it.

So there you have it, AI for Datadog and Datadog for AI, 2 flywheels, platform and lots of new capabilities. But wait, that was just an overview. We launched so much more, more capabilities and features all at Dash this week, actually 170 of them. Last year, we had a plain text slide. This year, we decided to turn them into stars in this beautiful galaxy. So this is an incredible moment for our engineering teams, but we're most excited about delivering all these capabilities to you. We're not here to just give you more tools. We are here, as Olivier mentioned earlier, to help you achieve autonomy in your development, operations, and security. Thank you for spending your time with us this morning. Have a great Dash and go Knicks.

All right, everybody, let's get this started. My name is Koji Ikeda. I'm one of the software analysts here at Bank of America. Absolutely thrilled, and thank you for joining us for the day 2 keynote lunchtime keynote. To have Datadog CFO, Dave Obstler with us. Thank you so much for doing this.

Thanks for having us. We appreciate it. Thank you.

Thank you. So we have a lot of people in this room. Thanks again for joining us. And I guess there are some people that know Datadog very, very well, but there's also many in this room that might not. And so let's start very, very high level. What is Datadog. What is the core problem Datadog is trying to solve today?

Yes. So Datadog has an observability and security platform that allows the deployment of modern software applications mainly in cloud environments in a safe and effective way -- most of our customers are those that have mission-critical software that's customer-facing. .

So think about video or credit card companies or banks or airlines or hotels, all of them have digital applications that interact with their customer and the Datadog platform is used to monitor and secure the effectiveness and operations of those platforms. Most of this is in real time and Datadog over the years has had an expanding platform to cover more of the surface area in examining what's happening in those applications and allowing them to operate in a good way for the customers.

So your results and fundamentals are accelerating. Clearly, good things are happening. And so why is observability and security becoming more mission-critical software complexity and AI adoption acceleration .

Yes, definitely. AI adoption is a part of it, but essentially, we are observing the development and the production of modern applications, mainly in the cloud. And as they get more and more complex, and as more and more of those applications are moving from legacy technologies to the cloud and be modernized, that is what Datadog does. So what is the effect of AI. One of the things is anytime there's been new technologies of which certainly large language models are one, there has been more of an impetus to modernize the tech stack and therefore, create more applications that are in the cloud and that creates business drivers and has historically created business drivers that have enhanced the data business.

This applies both to non-AI-native companies who are modernizing their tech stack in order to have large language models in their applications as well as a set of infrastructure companies who were called AI natives who are experiencing a demand cycle and have rapid product releases, they're cloud native.

Their whole stack is modernized. And they're using in a very significant way Datadog products to help observe the delivery of those products to their customers.

So 1Q results A little while ago, but still yes, --

Fond memories.

It has been a couple of weeks. But it's a -- I think our note started with low, right? A fantastic results all around accelerating growth. winning new customers, you never thought you would have before. And so let's tackle the first part about just accelerating growth. Looking over the past 6 months, what sort of inflections were you seeing from core observability using out there?

Yes. So this has been going on for -- I think we've been accelerating for 3 or 4 quarters. So we've been communicating this message that -- we have seen a good buying environment that the investments we've made in our platform, we can talk about the number of different products, we're resonating. So we're getting more platform growth -- and we also are having a significant demand cycle in AI native companies.

And this has been building on itself. We've also been investing substantially in our go-to-market over the last year in order to deliver this to our customers. So this has been building on itself. And essentially, in the first quarter, it continued that trend of acceleration in many areas, the non-AI natives, the AI natives, different geographies, enterprise down to SMB and then all those were all contributing factors in producing the first quarter, but the seeds of that started 3 to 4 quarters ago and have been building on itself.

Let's talk a little bit about the multiproducts. You guys give a lot of metrics 2-plus products, I think 4, 6, 8, 10. I think I got that right. So yes. So what are the products that are driving the most multiproduct sign out of those metrics, the 2, 4, 6, 8, 10, which is the one we should be focusing on .

Yes, I mean you should essentially -- the benefit of Datadog is that you can do all of your observation actions in a single platform. So in some ways, it's a single product, the platform. At the epicenter and this has been going on for some time, you have the -- what we used to call the 3 pillars, which are infrastructure or metrics, application monitoring or traces and then logs.

And we've had a very substantial demand cycle from what we call digital experience. And this is taking how an application interacts with customers from the back end all the way out to the mobile device, et cetera. That's called -- that's RAM and that synthetics, et cetera. And that -- those are bigger products for us.

Those have been growing very rapidly, and there has been a consolidation away from both point solutions and do-it-yourself towards our platform. That has been enhanced by some other products that we've put on our platform, including our cloud security products our service management, which is basically interacting with the users to be able to manage cases, et cetera, what we call products that allow you to A/B test on different applications and determine what's most effective.

So we've been adding on additional products on top of that. And then we started to add on what we call AI for data dog and Datadog for AI.

So Datadog for AI is most of our business is created because there are many things that impact an application. What I just mentioned plus databases plus network and now you have LLMs and other things. So we've been working to monitor those, and they are being adopted as well as AI for Datadog, which is how can the platform gets smarter, and service the customers, and these are things like our bits SRE.

And all of those have been developed and are starting to get traction, which is enhancing what had happened over the last 3 or 4 years in the core pillars.

So infrastructure monitoring, APM logs are all big ARR business?

They're all over $1 billion share .

Remind me out of the other products, what sort of metrics you gave on scale. I think there's a bunch that are $10 million plus -- and -- is there the potential for some of those to beach $100 million $250 .

Yes. We've been doing this. We've been giving a lot of metrics. Our Roman synthetics have passed through that, and we gave metrics on that. we gave metrics, I think that security passed the $100 million. What we've been doing is, over time, as we reach these metrics generally around 50 and 100, we have been giving those metrics.

So we had the passing of 100, which was the ones I mentioned, security, then we had a number of other products that were reaching the other -- the 50, which were things like database and network. I'm not sure I have all of this exactly right.

And then we gave metrics. So we have a lot of other products that are passing through 10 and multiples. And so yes, we have a lot of products that have been scaling this. And what we said was there are lots of opportunities. And what we're going to do is we're going to give -- as we reach these milestones, we're going to give those metrics so that everyone can follow along.

And the things that show a lot of promise are Bits SRE. These things like the product analytics we talked about, which is really about how an application is constructed and interacts with clients. These are all smaller products, but the TAM there and other point solutions are much larger than we're at today. And so we're optimistic we can scale those different milestones as well.

Yes. In the quarter 1Q results you guys won a lot of big -- I mean, you guys have many big deals for a while now. So what's going on there with the big enterprises? Is it the go-to-market motion getting better in the enterprise? Is it the enterprises need something more like Datadog? Maybe it's a confluence of both? I mean, help me understand the big deal activity.

Yes, it's definitely a confluence of both. So you have in a customer base that has been around for a long time. So therefore, they're not cloud native. They didn't just get birth. So they have legacy technologies. And they've they have a long way to go. There's somewhat 25%, 30% of workloads in the cloud right now and modernize.

So they are continuing to find use cases and modernizing Datadog's platform is getting bigger and better. So we're consolidating market share onto our platform. And we're getting better about delivering the enterprise service model whether that be channel partners, customer service, technical help the whole ecosystem to be able to deliver.

So all of that has been what we've been investing behind for some time that is bearing fruits. And that's resulted in some large lands. Some we still are largely a land and expand, some very substantial expansions. And when we've given our customer examples, if anybody is curious to go back and look at the scripts, what you'll see is a great combination of what the economy is, you'll see insurance companies, financial service companies, tractor companies, all sorts of different car companies, airlines.

So you'll see how this is evolving in the spread of the business towards cloud native and certainly AI natives, which we'll talk about, but also the cloud nativity within very large traditional enterprises.

Is there any limitation to the type of company or size of company that might not look at Datadog anymore? Or is Datadog available to all types of companies?

Well, so there are some companies that have in their past wanting to want to do this themselves. There are very few. I mean maybe I think Google is a good example of trying to do things themselves. So it's not -- that's not really core to our business, although -- and we'll talk about it. There's examples where we've gotten those types of businesses, and we'll talk about that.

I think the other thing is if we generally are delivering our product through the cloud, so if you require an on-premise solution for regulatory or other reasons that has, by choice, not been where we've concentrated our R&D. We are developing more of those products. So you can see companies which because of either their practices, or the regulation cannot have data leave their premise. That would be a company that is not core to Datadogs end market.

Last quarter, you guys talked about winning some AI labs within some large tech companies. Let's talk about that. So what happened there? Why did they come to you? What were they looking for? And how are you guys hoping to solve that?

Yes. So as background, so I think we -- there's a lot of information we've given a lot of information about how pervasive the AI business has been. And this includes some of the foundation model companies, companies and database companies that are vertical.

So already Datadog has been used pretty pervasively in the monitoring side. So production work environments, inference production. And I think we said over 650, and we gave a lot of statistics on spending over $10 million and 10-plus products. And what we added to that is that 2 larger companies, a hyperscaler and another very large tech company that have model creation within their businesses, foundational models had used Datadog for training as well.

Most of the time, our end market has been for production workloads. But as a couple of things have happened as there is a boundary between training research or training and when it has to go into production that we've now been able to have some customers buy from us.

And in this case, there were some large customers hyperscaler who traditionally does more things on their own, but use Datadog. And what we found is this is an example of the fact that they may not be using Datadog pervasively, but there are use cases where there going to be using Datadog, and that's a great voice of confidence that these companies whose whole businesses this are using Datadog, so it's sort of like a seal of approval.

I mean you mentioned earlier in our conversation that Datadog does well with Inference. And now you're saying training too.

Maybe we're saying we have, but we're not -- we don't believe -- we haven't said the -- we generally don't overpromise. So when we have like a certain amount of training that is spread out we'll tell you, right now, it's more of a sort of centralized. It's good, but we have said most of our revenues we expect to be from production and inference. But we'll see what happens. We'll bring everyone along.

As we think about AI in the future and more enterprises organizations, everybody building their own things, large small language autos for I mean -- how does Datadog think about that opportunity and maybe even going after it a little bit .

We'll prepare itself for that opportunity. I think that for the most part, when you think about production, so we're always going to -- I think we're always -- even if we're in training will always be somewhat proximate to production. So there could be market extensions. But essentially, you use Datadog when it can't go down. .

So if you're training, yes, I mean, if you're training and by definition in the sandbox, you're not putting in production. So you have sort of different impetus. Maybe that will happen, and we certainly are preparing our products. They're the same products. We certainly will have the products.

And we certainly are in touch with customers -- and we certainly will push that if it makes sense for customers, but we just don't know the answer whether that's going to be a core market or a specialist market for us.

So when I think about observability in the most simplistic way. It's data ingestion and analysis. But I know it's much more complicated than that and the architecture is very important in the way to address that. And so maybe help explain to those in the room that are again less familiar with Datadog what is it specifically about the technology architecture that takes this what could be a simple complement concept into a very complicated, something that you have to invest in and difficult to replace.

So 1 of the things before we get to the architecture, 1 of the things is data, Datadog, okay. But it isn't true. There's not one data. We have 1,000-plus integration. When you think about the operation of a cloud application in real time, CPUs, GPUs, databases, lines of code, network, all sorts of things. .

So one of the things that -- and this gets to the architecture, that Datadog did very early as they developed a common architecture to take all that data from all the different parts that could affect and organize it and put it in one place and make it transparent. So you can see that is very hard to do. And that's the reason why point solutions don't really make sense because none of the customers are saying, I want to see what happened to that line of code.

They say, I want to see what happened to the functioning of the application. So you have to see all of it. So that's 1 thing. So then the architecture is that data dog organized that data, knitted it together provided user interfaces and other ways to see it in real time. correlated.

Now this is being enhanced by AI now, which is using large language models to do diagnosis and maybe even 1 day sulfur mediate and produce it all knitted together. And that's hard to do because there are a lot of factors. And many other companies have tried to do this piece mill, and it's all gone back to that integrated data model putting the analytics on top of it, having a simple but not simplistic, meaning everybody can use it.

We don't charge by sea. We benefit when everybody goes into this utility and uses it, and that helps us. And so all of that architecture has been an architecture, which is -- provides the most value to our customer base in analyzing. And it all has to be real time because this isn't like, okay, I produce a marketing collateral, there's an error in it, I can fix said, this is your whole business on the front end.

So that's what's created it. And then all the different pieces, adding on all these different pieces and knitting together have been complex. Right now, we have a competitive advantage and that we have a very large platform. It's scale. It can handle anybody's scale. It basically organizes all this for everybody and it allows you to have a significant platform investment to amortize application investment on top of it, which is a big competitive advantage.

I wanted to dig in on the Datadog AI strategy, but also the data for our AI strategy. So let's get the data first. Right. What is so special about the data that Datadog has. And how long do you think it would be even if it's feasible for a competitor to amass that type of data to be competitive .

Yes, I wouldn't be feasible at a price point that would be competitive. In other words, large language models are by definition large. And what they're doing is they're looking at lots of data. So our strategy in the model side is to offer that, but also then have a set of models that are very specific to observability and security use cases and then to make them because of all the data we have, and then both have a higher functioning model and a cheaper model. .

Because if you're not -- if it's not a generalist model, you don't have to pay the cost of all that stuff that doesn't apply to your use case. So that's what we're doing in the model side. Then there's a whole bunch of different things.

So I'm speaking now about AI for data. We'll get to data for AI, which is all the DNA that goes in. Then there's also the ability to connect to all the software creation on the agent side, whether it be an agent or Marsh or a person, we don't care, whatever is creating and getting towards an application to see what's going on there and then correlate that with everything else that's affecting the application. So all of those are things that we're putting in our platform we're having our user conference in a week, the next week. And there'll be a lot of product releases and our Investor Day also had a very strong articulation of this. And what that will enable us to do is and is already is basically making the platform smarter being able to diagnose very quickly what's going on, being able to make recommendations on what to do about it.

And in some cases, to actually implement those recommendations. And that's the vision. That's what the whole service management vision is in the model. So that's like the investment in AI for data to make the platform enabled to see all those things and use large language models, whether they be the third party or our own to be smart. .

Let's talk about your AI products, the bits AI products Actually, I don't even know how many you said. What do you have? And how should we think about that.

Okay. So heading first into monitoring data flows. So we have LLM observability. So that is the functioning of an LA model Think of it as you have databases, you have code lines of code. So that is where you have LLM in a production model application. And you're using signals to understand is that affecting the performance of the application.

And I think we've said that, that is germane to having LLMs in production. And we have we've had significant growth there, still early days. So we're getting revenues from that. Then we have GPU, which is essentially like an infrastructure product, but instead of CPU, GPU, where you're seeing how the application might or the model may interact with GPUs in delivering.

And that's like similar to our other products where you have to see how the servers or the GPUs are doing -- and then you have that is sort of an example of data dog, monitoring things that affect. So that's data door A. Then you have AI for Datadog, and those products are the bets products, bits is sort of a general name. This is our mascot, our dog. So that's why we are using bits. It's very cute.

And so for different end markets, so SRE would be the systems reliability engineers, et cetera. So that product is out there. There are 2,000 customers using that. I think we have 100,000 or more investigations. And then we also have products in for development and for security that we're rolling out in the pit suite. We also have, as I mentioned, the ability to connect to the cogeneration through our MCP server.

When that will enable that information to get into Datadog and monitor that. And we also have a bunch of other products, agent [indiscernible] et cetera. That is being able to monitor on sort of the cost and management side, how much you're spending on tokens, who's using it. We have cloud cost monitoring and now we're extending that into agentic monitoring. So that's -- that was a long time. I apologize for that. There's a lot of names, but those are some of the product lines that are being put out to market in those areas.

I want to talk about platform consolidation. Good driver of growth for you guys. What is driving platform consolidation today with the large enterprises. And does that theme of what they're doing today for consolidation just continue into the future? Or does consolidation maybe change?

Yes. So you might wonder why this didn't all this happen already? Why isn't -- why didn't everyone just buy everything in Datadog to begin with, and it's already preconsolidated. The reason is, one, Datadog didn't have these products in 10 years ago. .

So there are other products out there and there are contracts that customers have with those products. So we're in an environment that it's happening over time. And why is that happening at all? It has to do with the -- that if you are a practitioner and you need to operate in real time.

You do not want to be context shifting. You don't want to be going into a lot of different data sources. It's slower. It's actually costly, more costly. So as we've developed these products and clients have looked at the utility and they've gotten off of contracts, this has been going on for some time. And I think as the product suite has been getting better and better and better, the components all getting to product parity and beyond.

It's been accelerating. And yes, I think we're very early on in this. There's proliferation, there's a lot of observability point solutions out there. Like I said, all of the decision-making is leading towards consolidation in the single platform. So we think we're pretty early on in that trend.

Do your buyers still think about -- and we'll talk about it in the 3 core applications Infra, APM, log analytics. Do they still think about that as 3 separate products or the customers beginning to come to you and saying, please solve this.

No, they've -- for a long time. We've been selling as credit, so you buy $2 billion of Datadog. You go in and you use all this. No, they've never thought as different products. We're doing that in order to provide transparency to you all. It's one product. They don't care what they're called. They only care about doing their job, systems reliability engineering. So this happened a long time ago. That's what made Datadog and they think of it as the platform.

I want to touch a little bit about security. 10 million-plus business -- it's been out there for 3, 4 years. A couple of years now. So maybe for those that are a little bit unfamiliar with here, you tell us about your security journey, what you're doing on the product side? And maybe more importantly, what are you doing on the go-to-market side? That's very a very good thing.

So first of all, you have to say what security, okay? So -- these are the things we're not in endpoint security. There's some really good companies. We're not in network security. We're not in e-mail security. What we're in is cloud security. Cloud Security has 3 components. It has how the cloud is working, it's called posture management and set up to secure digital applications.

Two is cloud SIM. How are you using logs and other information to investigate what's happening? And 3 is code security. How are you engineering security into the code. And so for -- so not on-premise, not legacy, but for cloud applications, we've built this suite of the 3 products. I would say the 1 that we probably made the most progress on is cloud SIM.

Some of the reasons are it's off of our logs business, and we have an over $1 billion log business. We have a great largest observability logs. And the end market has dynamics where we've been an innovator in it.

So one of our -- so we've had 2 major strategies. One is attach the whole suite to cloud-native emerging companies and 2 is attach cloud SIM to more traditional companies where we have a strong observability logs. So that's been the strategy and it's working. In terms of -- that's the product strategy.

In terms of go-to-market, particularly for enterprises, it's a different go-to-market in that one, you have the influence of a buyer that is not a traditional buyer, the CS and even though they're getting closer together, that is a different buyer than DevOps.

And for some reason, I don't know, I'm too young. But for some reason, there was a distribution -- bottoms-up distribution that facilitated direct in DevOps. But security is a sort of highly central -- has been a highly centralized and if CISO has the [indiscernible] So it's a different buyer who buys through channels.

So what we've been doing is we've been working on selling our direct to our champions, but also investing in product experts or specialty salespeople who are going to try to penetrate CSOs and others and channels where we have a channel program like the other security companies. It's still early. It wasn't our DNA, but we're investing behind it as part of the overall security investment.

One thing that we've heard over and over from customers and partners is I'm the observability buyer that guy is a security buyer. Right. Is that how it's always going to be in your view? Or do you think one day that might convert?;

We think the whole strategy is that it makes no sense -- there's certainly territorialism. There's certainly -- but essentially, it's better -- vol up here, he would say it's idiotic not to design security into applications and to have all this come through.

So dev tech ops, we have seen signs and we believe it will come closer to closer together. A thing that we think will accelerate that is Agentic coding because that's going to speed up everything, and then it's going to require everyone to work together. So even though you're right, there are these separate buying groups.

We're seeing signs of them coming together. And we believe in the future, since we believe that businesses are in the and logical that they will come together, and if we can use the gains in technology from agentic content of detection of penetration, remediation, that will most likely help us in our attempt to scale security.

So in that vein, when you say genetic coating, it's getting pushed out from getting from the software development side and security has to deal with it -- does that feel like the you're beginning to pull in some of the security budget because of where that velocity of...

Progressive companies. We have been for some time, meaning that we have clients where it's all -- it's like I said, it's all knitted together. But you have a whole big world out there. And so it's a process. I don't know if I can say that in the last 6 months, we would see a sea change. We think it's going to -- it's an evolution, and we're -- and we believe it's going to happen. It's going to happen over time.

You guys are highly successful with cloud-native companies and cloud native strategies -- at your Investor Day, you talked about Cloud Prem product that you have -- so can you tell us a little bit about Cloud Prem, what's the strategy there? And how are you leaning into go-to-market with that product?

Well, essentially, we're leaning into it and go to market in the same way, meaning that if a client would find utility or a cost benefit from having data kept in their -- on their own servers. We want to have a product.

And so far, it's been that we have the analytics or we have the log and you keep the data. So we've also have products that allow the data to get in an efficient way of everybody pipelines -- so -- but -- so that's what we're doing. Now we're investing behind it. We have customers that want it, and we've been successful -- it still hasn't been the vast majority of the way our customers want to buy.

But what we're working towards is in difference, meaning that if you want to keep the data, great. If you want to give us the data, great. If you want to have all the functionality over to your side. And so that's taking time to get complete product parity where we have exact functionality.

That's taking time, and that's what Cloud Prem means. We're working on that. That's mainly for the first use case lows.

I got you. I wanted to ask you a question on durability of growth. You guys have great growth trends. And so how do we think about the drivers of the durability of the growth, whether that's cloud migration, platform consolidation and AI and what are you guys seeing that's giving you the confidence that growth continues in the passion it has .

Well, the durability depends on what your time frame is. Okay. So we always have believed and we're seeing that this is a very long time investment cycle. -- because such a high percentage of applications and infrastructure are still on legacy technologies.

So we've always said that if you want to look at this, this is a secular trend. And it can be cyclical, but it's also secular and it's got a lot. I mean you have 70% plus of workloads that aren't in the cloud. So we believe this has very long legs. In terms of is it going to be a straight line, meaning on every month, is it going to have the exact same thing.

We don't know. It's likely if proven you're going to have periods of investment, you may have periods of optimization. It's cloud software, it's consumption cloud software. So we can't predict exactly what the line is going to be in every moment, but we have confidence that it's very durable and it's very long because we're attached to huge market drivers.

Now we also -- if history repeats itself, we found that new technologies, of which AI is happening is a huge accelerant of this conversion. And so again, it's still too early. But if history repeats itself, that is likely to inflect the line upward faster to get to the same place 25, 50 years out.

Yes. Yes, that makes sense. I wanted to ask you a question on how Datadog thinks about product velocity and R&D investment and frankly, your guys' own use of AI within the organization. So you mentioned -- I think we talked about it yesterday, you got about 4,000 developers with an organization. So how do you think about leveraging AI and investment?

Yes. So step back, we've always been a product-led company we've been out investing we're bending over $1 billion in R&D. That's had tremendous benefit, meaning it's allowed us to invest in the platform at scale and the functionality.

We also have our company that has no shortage of things of functionality that we want to roll out. So that's been like epicenter to Datadog. Out invest everyone in R&D. Look at the pipeline of what we want to put in the market at a profitable price and creating value for clients and just look at your R&D resources against that.

That's been what Datadog has done. So what about coatings coding aid agent. So that is essentially another piece of raw material that we are using to try to accelerate the launch of features where -- we're not -- what do you call it, code math, it's agent Maxine, what's the total in [indiscernible] we're not talking [ maxes. ] We've never paid anybody rewarded anybody on the number of lines. We basically think about products. So it's all product, all those software developers, we call product people.

And so essentially, to the extent we can become more efficient in speeding up feature release. It's good for us, and we're finding that. We're basically experimenting around the interplay within that envelope on R&D between tokens and people. And we have a lot of experiments. We have places where we have keep the people. Let's see how fast you are with less tokens.

No. Constrain the people. Let's see how fast you get. And we're doing -- we're pretty nerdy engineering, wonky. So we have all this stuff going on, and we're trying to see what's going to happen. We believe that the componentry of R&D will shift somewhat. We don't know exactly to tokens from people, but we don't know the exact amount.

Have you been able to use AI internally with all that development that you have to help out with your own homegrown solutions in sales and marketing and G&A at all.

So we have AI -- there's 3 or 4 major use cases. One, the product itself. We talked about the models and the products. So those would be components of the product. Two is accelerate product release through software coating tools. 3 would be use AI for productivity of the employees and 4 would be what you're talking about. .

Are we able to use a large language models or AI to improve the productivity and the effectiveness of the different functions. And yes, I mean, there's many, many examples. We're using AI to automate the creation of deals and the processing of deals. To enable our salespeople and our market people to understand propensity to buy and direct our salespeople in places that have a higher chance of success.

So we have all of those -- I'm not going to list all the names, but there's lots of them like that. How do we enable salespeople in product training and things like that? How do we speed up all of the back office functions, whether it be legal, accounting, cash cycle, marketing collateral, all of that.

Yes, we have experiments or projects going on in many areas to try to do that. And what we're really thinking about mainly is how can we convert repeated repetitive administrative tasks. Towards more insights so that we can act faster and inflect the business.

So we have all that going on too. I mean we're not -- again, it's not token. We're not -- basically it's -- you've got to produce something from this, but we have a lot of opportunity. .

And remind us, what is the investment strategy in sales and marketing from -- how do we think about that? .

Yes. The investment strategy in sales and marketing is bottoms up. to be able to cover the high potential customers around the world. It has to do with investment in enterprise, all the way up government and enterprise biggest entities all the way down to start-ups on a lot of different to be able to cover them comprehensively and to be able to get lead gens from them, et cetera. then it has to do with geographic expansion like many U.S. companies, we were most developed and followed where the investment in cloud software was most intense was the U.S.

And we're finding many areas with great payback Brazil, Korea, India, et cetera, where we did not have on-the-ground presence. And we're developing packages of salespeople. Sales engineers, marketing dollars, maybe even data centers to be able to do that. So that's sort of where it is. In addition, the way to reach customers direct and then also channel that we're expanding.

So it's a bottoms-up business plan aimed at that micro level and how we can reach the broadest swath of customers around the world.

I got you. I know we're running up on time here. And so maybe M&A strategy. You guys are not afraid to go out there and acquire things and how are you thinking about your M&A strategy? Any changes in what you've done in the past -- and how do we think about that going forward?

Yes. I think we basically to date, not acquired streams of revenue, but acquiring product capabilities and the people. And we continue to do that. So when you think about building product -- we often build product and then we can enhance that through acquiring a product capability.

And that's the epicenter of what we do. And if there's opportunities, we'll continue to do that. In addition, I think we're open to something that bigger where we do acquire some customers as well, but that will really be dependent on can we integrate it in? And can we accelerate? Does that help us?

And so we're pretty principled on that. One of the things we really insist on is that the R&D team and the product teams stay at Datadog. So that means for certain companies sell move. We don't want that. We make it in the incentive structure. So we only acquire companies where they want to stay.

So that's sort of what we've been doing. It's been very successful. We've created some very big businesses. through that with not that big acquisitions. And I think that's the core with the reservation that we could look bigger to the extent it fits in, in our disciplined way.

I got you. Yes. Last question for you, David, and thanks so much for doing. Yes, you got DASH next week and so -- we're trying to get a little bit out of you of what -- as investors in the room and myself as I look at the schedule, is there anything I should be focusing on just to make sure I don't miss anything big coming on.

Yes, was background, DASH is a place where we make a lot of product releases we do on product strategy, it's for users. And so I think you're going to get -- you get a pretty good road map. And then if you look at who's speaking and everything, I think you'll see is going to have no surprise, a lot of AI content. So some of the things that we will be discussing here the AI for Datadog, data the dog for AI, con, all that well, I think will be fleshed out a bit more at DASH next week.

Got it. Thank you. We're all out of time. Thank you so much, David.

Thank you. Good interview.

I can't even see any of you guys -- view in the middle -- Olivier is obviously a much more interesting and important person than me. But for those of you who don't know me, Peter Weed from Bernstein, I cover software. There's actually 2 of us from Bernstein who cover software that you might know, Mark Moerdler and myself. I do kind of IT infrastructure, dev type software, which is kind of my long-term background have been kind of around it since the late '90s, and you may know Mark, my colleague, he actually was a database entrepreneur back from the same time. So he covers databases and mostly like functional line of business applications.

And we are really fortunate today to have Oli from Datadog join us. Oli is 1 of the founders and the CEO of Datadog, a company that has been really instrumental in category that has really kind of emerged over the last decade that we now call observability, which is all about helping like when you ship a cloud application, you expect it to run. And this is a company that helps ensure that it's both up and it's running at performance before you even put it into production, you can predict how it might behave.

Now I recognize the audience there's probably going to be a variety of background in data and kind of where they are. I think it may be relatively obvious from some of the stock price action recently that things are going pretty well.

Oli, I thought it might be just interesting to kind of reflect on the last kind of 3 to 4 quarters and the acceleration that you've seen and kind of where has that really been coming from?

Yes. So I mean, look, First -- thanks for having me here. We have seen acceleration for the past few quarters. I mean, I won't bore you with the exact numbers. We deal in the filings. But we've seen acceleration. What's the most exciting to us is that it is not a specific customer, specific set of customers, specific side of the business. We've really seen acceleration across the board. And the way we look at the customer base today is we look at both the -- what we call the AI natives, which are largely companies that were started in the last few years built on AI or companies that were selling a little bit before that, that are the -- essentially the large builders and the most important building blocks of they are today.

So we see acceleration in that category, which sort of makes sense. I mean, everybody knows the AI is happening right now. So -- but what's even more interesting is we see an acceleration of the business outside of that. So in the non AI companies. And for us, that's a mix of the older cloud natives, which existed before AI and also all of the larger enterprises, whether they are large enterprises or mid-market, and we've seen acceleration across all of that. So on the AI side, obviously the AI companies are consuming a lot of infrastructure, building a lot of applications, running a lot of applications, and that's what we see on day-to-day basis.

On the rest of the market, we see also more applications, but we also see some growing adoption of AI itself. So we have -- we've disclosed in the last quarter that we've seen very large increases in volumes of our AI-specific products, whether that's the traffic we see to our MCP services or whether that's the traces we get into LLM observability product that really gets traces produced by AI. So we see inflection across all those points.

Oli, I think you've done a really nice job kind of helping people kind of understand like how you get paid and how you kind of ride the cloud wave. Maybe maybe describe like how you think about like where your value comes and like where your growth equation kind of comes from over long periods of time?

So the value comes from helping customers be on top of the complexity they create their applications. So meaning they ship applications, they have to understand what infrastructure is part of that application, how it's running, whether the application itself is running, if it's fast enough, what the users of that application are doing whether the application is creating the right outcomes for the business, whether the application is safe and secure. So all of that, that's what we do for our customers. One way to think about the growing importance of what we do is that, over time, developers have become way more productive.

So when you go back 50 years more than that, developers were hand coding every bit on punch cards and writing every single piece of code that was running-- then after that, there's been higher level languages, better interfaces, compilers, then after that, you've seen open source software. You've seen cloud, you've seen SaaS and internal -- at every single step of the way, you see an increase of productivity. What it means when you get higher productivity is that you produce less or you produce more in less time. which means that you don't really get to [indiscernible] as much to it, you don't understand what's actually going on.

And you see an even bigger acceleration of that today with the coding agents, where engineers can in a few minutes, with open application. They just have no idea what's going on. They probably don't even read the code, but they combine all sorts of different components into an application. So that increase of productivity creates a dramatic increase in complexity, and the problem we saw for our customers is we actually understand that complexity. We manage it for them. And in the future, we want to be even more an automate basically the way the applications run after the [indiscernible].

It's interesting. I think like 1 of the bare cases that has been pervasive around software is like now that we actually have these agents this complexity can now be handled and can't we just go ask whether or not it's Anthropic, Claude or any of these LLMs to just do it for us, how is my system doing? Tell me when it's down. Like, why does Datadog still have a role in a world where we have these AI agents? And maybe it's not today, but maybe 3 years from now, they get smart enough to do this.

So I mean, the 2 things I'll say. First of all, the companies that are building these kinds of models are also using our product to manage their infrastructure and their applications and everything else. So it is a hard job. It is consistently a harder job than writing the application itself. And that is why there's been this race has complexity over time. And everybody is going after it, including the leading companies AI [indiscernible] AI builders all the hyperscalers. So that's 1 thing.

The other thing I will say is that the kind of problems we saw are somewhat different in nature from what these fundamental models or financial models are built for. and operate at very different levels of data volume and real-time requirements. So we have 5 or 6 orders of managing more data than which is typically put in an we need to operate in the real time. So the analogy I would give to understand where we stand with the prime we saw versus what these models do is compared to self-driving cars.

And so mean today, it's working pretty well. You have self-driving cars everywhere. You could, if you wanted take a photo from your dashboard on your car and upload it to Claude and ask what you should do. And it will give you an answer, and it will be a pretty good answer. And we'll tell you actually you should make -- if you turn on the right, and that's why, and this is what's going to happen -- the problem though is that Claude will do that, but it will not drive your car in real time. If it did, it would do it poly, were worse than the algorithm that's or the mall that's running in your car.

And also, it would cost you $10,000 a day to drive your car. So that's a way to understand what these models do relate to what we do in our case, for observability, automating the life cycle of an application, automating the availability, the validation, the security of an application.

I think the other bear narrative like when you get through that, oh my gosh, it's actually a difference between like deterministic programmatic software and kind of probabilistic software, it's like, okay, well, I still have these code suggestion tools. And we've got some open sores like why can't I just like to ask it to build me this bet a programmatic software late, you're obviously not seeing that happen, like what's the disconnect between that narrative in the reality of like why the most innovative companies are turning to use Datadog as opposed to do it themselves.

Yes. I mean it doesn't make economic sense for you to do it yourself typically. And the reason for that is this is -- what we provide gives enormous leverage. So typically, for any dollars customers spend on us, they're going to spend $10 or $20 on their cloud provider. They're going to spend $20 to $100 on their engineering team. with the rise of cutting agents and [indiscernible] in general, maybe some of that $20 to $100 depending on their engineering team is actually going to go to these AI models instead -- so we're talking about a huge amount of leverage.

So when the -- and by the way, we help our customers make the most to optimize their spend on their cloud provider, on their engineering team, on their AI levels on all of that. So we're basically the small part of spend that helps our customers make the most out of the rest and accelerate and make more money, so at the end of the day, it makes no sense when you have them to do that. I'll give you a few more examples that we've had, so more recently. So we mentioned over our last earnings call that we also had a number of hyperscalers static adopting our products in their case for helping with their the development and training of their AI models.

And these are companies that culturally basically don't use any commercial software. They write everything themselves. They use open source, they have to completely homegrown by nature. Even those companies had to come to the realization of that, actually, it didn't make sense for them to do that. They would get better outcomes. They would get faster delivery of what matter the most to them. They would get better economics of it, and that would be better for the business in the long run. We didn't build Datadog to serve a handful of hyperscalers. That's not our business model. A business model is going is to go after every single company out there, have products with very broad appeal.

We think though that this is a great example of these hyperscalers just going for the first time, maybe through what every single other company has to think through, which is, okay, so what do I need to own? What should how do I get the best return on investment? What am I trying to achieve? I need to get growth, I need to get better economics at scale? What's the best way of doing that. I have finite resources, which should I focus on. And I think that's what we're seeing at play here.

So if -- can't be replaced by AI, just asking questions can't be coded, this place still has a tremendous number of start-ups, like I spend a lot of time on Hacker News, for better or worse. I would say, every month, there's somebody who's like celebrating the fact that they're trying to build some new observability platform. And you and I were talking about this a little bit, but beforehand, my history, I'm an old product manager -- years ago at Microsoft in the middleware category. And the 1 thing that has stuck with me this whole time and when I worked at McKinsey and other places, is that I always watch the buyer. And for me, what I noticed is that the greatest chance for a new entrant coming into a market was what the buyer changed, right?

So like you could say like, why does Salesforce exist and replace prior CRMs. Well, because the buyer moved from the CIO to the head of the function. Why is Datadog, I think, an amazing company that replaced prior generations of what we might have called monitoring software. Well, because we got the site reliability engineer and a new set of expectations and requirements when we move cloud delivery and management. If we look at AI, it appears, and I think you've made a good case for when we shed in the past, like the roles might compress again.

There may be several roles that need to get redefined if we're really going to have high velocity shipping security might need to be grouped in as well. And that looks like 1 of those scenarios kind of like the site reliability engineer that collapsed some old roles together and gave you an opening into the market, but I think you guys have been thinking about this almost from day 1 about like why that type of change actually is kind of like what you're almost hoping for because of the upside as opposed to being a threat.

So a couple of things on that. So the first thing I'll say is I mean there's a -- I think you understand when you say there's another new engineer building and observative product every month. I think it's every 2 weeks -- and it's been like that since the very beginning of the -- so I still remember the first fundraising meeting I took in the -- with VCs way back in 2010. And the first word that came out of the mouth of the investor where all monitoring because that's what it was called at the time, monitoring credit market. .

And then it went down here from there. The -- and it's always been like that. And I can't blame people for doing that, and that's what I did. I mean engineer, I started an absorbing company, and that's naturally I think we did approach it a little bit differently though. So our starting point was not, hey, let's build a better product [indiscernible] we're using that, we had it. We're going to build a better one. Or we're using that is too expensive, let's build something else. The starting point for us was, hey, I used to come from development. My cofounder used to come from operations -- technical operations. The 2 teams fought all the time -- and we thought, okay, so , maybe we can bring them together and have 1 platform that bridges the gap between the rules and things like that.

And that vision actually lended itself very, very well towards adding more roles, bringing more people into the mix, expanding the footprint over the lifetime of the company. And that's what we've done. We were actually quite lucky because -- what we didn't understand when we started that was that the collapsing of the role at the time between DevOps, we're at the center of cloud adoption. The whole DevOps movement was born at that time. And that's when you saw those rules become closer. And we saw that pull in the market. but you see even more of that happening today.

I mean you see securities being brought in, you see with the advance of cutting agents, now instead of large deals with specialists, you're going to have smaller teams with people wearing many hats across product, design, engineering, security and many other functions. So I think we're going to see even more on that. The last thing I would mention because you talk about resisting disruption. And we built also the company very deliberately in a way that keeps us in touch and honest when it comes to the low and bleeding edge of the market. So we have more than 30,000 customers. But the bottom half of those 30,000 customers represented about 1% of our revenue. So we do not serve that bottom half for money, it actually cost us money to serve those customers.

We do it because those customers are -- they are small companies, they're individual students. We also have people on the free tier that we only have included in that. And what those people do is they keep us honest with respect to the simplicity of the product. They also help us see what's coming in the market because we see they adopt all of the new stuff way before the large companies do that. So we see that coming. We get many of what we need to build for. And whenever we break things because we try hard to accommodate our largest customers that have very sophisticated needs and that make our products more complex. We see -- actually, we hear from those smaller customers, and we also see in the numbers that really something is not working as it should be working there.

The alternative is what most software companies do, which is they quickly understand that most of the money comes from the higher end of the market. As a result, they focus to care mostly about their large depressed customers. and then they end up being on the path towards future escalation. And before you know it, you end up with products that are what I call enterprise abomination, and we all live with them every day in the office. We know exactly where they are. And I think that that's -- you don't come back from that basically. That's where we become irrelevant over time.

I think we've focused a lot of this conversation is like some of these like bear cases around the people have. And -- maybe we should turn it a little bit to some of the opportunities, right? So you all don't rest. Obviously, I think you spend more in R&D than pretty much the entire rest of the sector put together. So you're shipping a lot of things. And if we look historically at the S curves that the business has gone through, maybe you started with metrics, actually really kind of collaboration and the metrics and then APM and then logs, and these have each kind of grown to being kind of $1 billion type opportunities. There are several things that you've been putting out that you, I think, have been very hopeful that can become some of those next desk curves. -- whether or not it's security, whether or not it's AI monitoring, the GPUs, modeling that you guys have been putting out. when you look at the acceleration that's going on right now, how much can you kind of thank those new things for this versus kind of your existing footprint?

What's interesting is most of the acceleration, like mathematically, it comes from the existing stuff. And when you think of -- like we have the I think we said like 14 of the top 20 AI companies using us, including the very largest of those. Like they use the same traditional products as the rest of our customer base. So by and large. So that gives you a sense of the shape of the market in general. When you look at the observed category, so there is gotten a number that we put every year, but we're #1 there was ability according to Gartner in -- or at least in the closest -- the category had as closer to observability, I forgot the exact name of the category. .

And -- but we only have about 13.6% of that market right now in a market that is growing pretty far -- pretty fast. So it doesn't take a ton of imagination to see where there's a 5x from there just from the core capability, looking at infrastructure applications, end users around those applications, logs and all of that stuff that's that. The new part of the market is everything that relates specifically to AI. Part of that is what we call AI for Datadog, which is who we can do more for our customers and really expand beyond the observatory category. So basically automating the whole for them, not just being in the business of telling humans where things are going, but being in the business of fully running and automating the application, it's promotion to production, it's repairing it when there's an issue, making sure it's secure at any point in time, there's a lot of automation we can do with that.

Another area is what we call [indiscernible] which is observing and understanding all of the AI components our customers are using. And mostly it's fall into 2 categories. One is the AI models our customers are using within their applications or within their agents, which are nondeterministic models and which can't just be tested in development and then validate a ship to production, you have to keep observing them all the time. Like they're like humans, you have to get to know them and you spend time all the time and you keep watching, and you see what's going on and you keep adjusting. So that's 1 big area.

Another area is adapting to the whole new world of agentic coding when developers can produce a lot more applications, a lot faster. But would they still come do faster is getting those applications to production. Now they still need to make sure that those work that they do what they're supposed to be doing, that they provide the right outcomes for their business that they are secure, that they keep working once they've been put into production and when other people are also making changes to the world around it, so that's a huge -- right now, we see a huge increase in the volume of co-change that are being made and applications that are being shipped, and there's a big opportunity for us there, too.

I think there's probably like 2 kind of natural places to take this. Maybe we start with the Datadog for AI. So if we think about some of the things that are going on, they're obviously like there's some very basic uses of AI, but as we start to get to agents look, maybe beyond coding copilots, there isn't like a lot of depth of use of agents. But as we get more and more of that, there are some kind of unique observability problems that come up with essentially monitoring like the behavior of the agents over time. And there are some companies that have come out, things like LangChain and BrainTrust and these types of things that are designed to kind of tackle some of that job, like how do you see Datadog working alongside of or instead of some of these kind of born and AI components that are there to observe kind of the behavior of agents?

Yes. So I think there's 2 extremes to that market. There's 1 side which is everything that's in production. And when you talk about things that are in production, it doesn't make any sense to separate the AI from the non-AI components. For 1 thing, what the agents do is -- so they have a lot of reasoning, but also they use a lot of tools. And the tools are actually apps. So when you monitor the whole thing, you basically end up monitoring like largely apps with a little bit of AI in sprinkle. So when you're on production, it doesn't make any sense to have different tooling, and that's something that we will -- like we develop, we have an element [indiscernible] product for that reason and something that we will want to own in the end.

On the other side of that spectrum, you have what happens in development. So when developers are tinkering, they're playing on their laptop, their testing various things. Historically, that side hasn't been very platform in that in a company, you don't have 1 of them and everybody plugs into it, but whether you've had many of them. And these were more a matter of choice and taste for the individual developers. So the equivalent there could be the IDEs, the developers we're using. I'm old enough to have leave some of the AMAX versus VIAs. So I remember people being very opinionated about the editors they use and things like that. And whereas absorbability, you have only 1 production application. So like everything has to plug into 1 environment to 1 thing in the end and that's platform in nature.

So what's not completely clear to us today is how much of that is they've clearly developed a purely local non-platform lots of it of usage in any company, some homegrown, some and source, et cetera. And what part is in of basically back there, we have to go to own fully the production side of things.

So when you kind of look at that opportunity going forward, I think perhaps 1 of the challenges is it's very dynamic, like how we want to do all of this stuff is somewhat of a work in progress. How do you think about your own investment in kind of being part of the frontier there, so you're in the right place at the right time when things start to solidify and become a little bit more common across organizations.

Yes. So I mean we spent a ton of time like watching our customers and what is they do and how they work. And so we have the luxury of catering to every single layer of the stack in the customer. So most of our business is with large enterprises. But we also serve a lot of start-ups in AI, whether they are super large-scale model makers or the companies 1 level below that are largely AI native and building on AI. So when you look at the largest, it's interesting to see how they work but it might not be completely transferable to the rest of the market because those companies have unlimited access to inference for 1 thing. And also for a number of them like almost vertical wall of demand in front of them.

So they are not in the situation that both of the companies are in. So they might make different choices and that might the way they operate might not carry across. When you look at the tier below that, though, like the companies that are, I mean, still very successful going fast in AI, but are still buying most of their influence from somebody else and have to make some hard prioritization goal in terms of what would they work on and not and have more relatable growth for the rest of the market. I think from there, there's a lot we see and we can learn in terms of where the world is going.

We see the trends pick up like a few quarters before the rest of the world might and we get some interesting insights in terms of how our products are being used and we need to be able to support those companies as they do that.

And I feel dumb. I feel like I should have almost even asked the question in a slightly different way because I realize for the audience, they might not fully appreciate your product-led approach to the organization, which you're kind of describing some of the outputs of product-led is something that I have been kind of in love with ever since like I got enamored with Atlassian flywheel approach background in 2010, and I tried to apply it to several businesses like building it out of Dropbox and things like this. like maybe help people understand what is special about that product-led approach to the organization with the growth function and with product management being so tight with discovering these opportunities and delivering and why that allows you to stay so innovative?

Yes. So the whole business is built around bottom-up product adoption. So we -- our product is typically adopted by the record files and then the people who actually do the work. We build around a unified platform, which is very important, meaning every single new product will build as part of that platform. Whenever we do M&A and we acquire company we do platform. So everything is part of that tightly integrated in 1 single platform. We charge per usage. So the whole business model is usage. I mean there's a few tiny parts of the revenue that are based. That's because we compete in categories that are purely based. But the vast majority of our revenue is completely usage-based, which also is interesting because right now, it allows us to I mean basically, we don't have any transformation to undergo to get into an AI edge where the notion of seed gets disrupted and you have to charge for outcomes for value per something.

We're already usage base, so we can very easily adapt our model. Everything around the adoption of our product is -- everything around the business is geared towards the adoption of the product. So when we land the customer, we're already deployed typically and we just see the usage start growing from that time. What this gives us as a business is a great way to understand which products to build and not build for our customers and what is valuable or not. Again, we use age based. We [indiscernible] about having that usage or having new products being attached to new SKU, meaning that we get very clear signal about what's valuable and what's not in our products. which allows us to keep developing a lot of the right things.

So that's -- and if you follow the business, were knowing the industry for very quickly expanding the product footprint very quickly, improving the existing products and be very good at growing new lines of business. And we have a lot of tasks in our earnings around the adoption of those businesses.

The other things we get from having this product-wide approach to the adoption and bottom-up adoption, is that we have a very efficient go-to-market. So we spend significantly less as a percentage of the top line on go-to-market, which then the typical enterprise business is which allows us to invest about 30% of the top line on the R&D, which then feeds the flywheel of building more product and building more differentiation and being relevant for the long term.

Maybe the asterisks on that now has been that you are actually starting to build some enterprise sales capability. like what drove that evolution where you keep kind of the product-led core, but you realized that some enterprise overlay was important. And how has that kind of worked in...

Enterprise has been around for a while, like we built that team many years ago. So the -- all that's baked in. What's important to understand is that it's enterprise but bottom up. So it's never the case or maybe I can happen once. But it's never the case that you call up the CIO and you're super impressive in your golf game and you end up with a deal that gets done, that's not will happen. .

In our case, we start with the teams, the teams adopters. And then we make sure that we connect between the teams that are adopting us and the leadership in those enterprises, so that they understand what is their buying and what the ROI is and things like that. And so that we can enter growing and longer-term agreements with those companies. But even with the large enterprises, we mostly land very small. -- the ASPs when we land are in the, call it, mid-5-figure to low 6 figures every year. which is very small for this kind of enterprise deals. And then we grow those customers for the very long term. And we announced on the earnings call, we typically have very interesting land deals or consolidation deals we make in the 6, 7, 8 figures annualized. But most of our deals start more and they grow and they -- we tend to have very, very efficient sales, thanks to that.

The 1 thing that we're doing today that is a little bit different is the new sales motion we're adding is that we're adding some overlays for selling security very specifically to the [indiscernible] because that's not the most we had before. And we find that it's a little bit different -- different enough to warrant a slightly different team to handle that. But that's specific to security to a CISO in the largest enterprises.

And by the way, maybe talk about that a little bit like why is security such a natural part of what you're doing? And obviously, you have security vendors that are also making some acquisitions into what looks like competitors in your space. Like how do you see that playing out between like your role in cybersecurity and what some cybersecurity vendors are doing?

Yes. I mean, look, the end of the day, the thesis we had when we started the company that you should bring Devon ops together also extends to security. It doesn't make sense for security to be separate. -- the teams would work well together, they should speak to same language, achieve the same role. And we see the same -- we see this accelerated by the adoption of AI, where you'll have the same person basically having to carry to wear many hats in care about security as well. So we think that, that bottom of adoption from the developers and the ops people is going to reach into security. And is, in the end, going to be -- it's going to become much more of a bottom-up motion than a upon motion in terms of how that adoption.

From a technical perspective, most of the signals you need to secure an application, you get observability. You have everything that happens in the production environment, you have everything that happens on the way to production in the core, in the testing environment. You have the behavior of the end users, you have the behavior you have the developers. You know exactly who's changing what and when, so all of that is the information that you need to fill into the security models, and we're uniquely positioned to get that as well.

So we went down the path right now of like some of the expansion opportunities organization and the commercial motion and why that's really helped you be successful. But I think you pointed out there's like the other side, which is AI for Datadog. And while it's still relatively early, there's kind of several investments that you're making there, everything from the its AI agent to some of the new modeling stuff with Toto that you guys are doing. Maybe help the audience kind of understand some of those things that actually can add incremental value, reduce the human time, add more predictability that you guys can do?

Yes. I mean, the high-level goal there is to automate right, and to do more than observed and to really automate -- so 1 thing a lot of our customers go through is so they use us to observe and monitor the applications and something [indiscernible] 30 a.m. and we work them up and then they start Zoom call with 20 people on it, that last 3 hours from 3 a.m. to 6:00 a.m. The site is down, business is upset. I mean you can imagine -- Yes, very costly, very disruptive, not anyone's favorite part of the job. So there's a huge opportunity to automate there.

And it used to be fairly hard do that. But with the new technologies in AI are really opening doors that were just not quite available before. So what we do today is we have products under operations side and on the security side, they do that automation. We have a bit, that's a name. So sorry, agent BI security agent. And what they do is they run investigations for you. So for example, in the case of the outage I mentioned earlier, instead of having like a 3-hour Zoom call with 20 people on it, 4 minutes in at an analysis that says, "Oh, by the way, I find out what was broken. This is what this is what it is. These are the 3 people who know how to fix that. This is the fix I propose, not sure we do it.

And I mean that's already life changing for those customers. And typically, this is the way they like they see they turn this on, maybe they forgot the tuning on and then the gain this incident. And then they get on to that long zoom call, and then after the Zoom call, they look at their notification and they were asked, "Oh my god, 4 minutes into it, you had it already. So -- and that's the light of moment from them and they start that. We have a similar agent on the security side that does security investigations. And there, the focus is mostly to cut the noise. So the teams that do security response have tons and tons and tons of investigations to run. 95% of them are more even are just benign.

And we -- a big part of the job there is to make sure to get to the super quickly, so they can focus on other things. And so we do that as well. So that's what we do today. If you squint, you could say, hey, some of that you could also do from you use cloud or and you give them MCP tools to acquire your data and maybe they can come to the to the similar reasoning and come to a similar conclusion. The next level of what we're building though is about having that intelligence directly inside the data plane at with all of our real-time data and to drive the automation itself, so models that are specialized that can run on very large volumes of data at a very attractive cost and with very low latency, so we can fully automate the operations.

There are some first outputs of that. You can see, we published 2 versions of a time series model like a financial time series model, which is called TOTO and you can look for it online. This model is very interesting because it is a model that is -- that was trained on observatory data almost exclusively, but that generalizes very well to other types of time series. So it can actually -- like it's state-of-the-art across domains, not just for viability. And so it's actually really good at what they're forecasting even though it was trained on mostly application of the ability data.

What's also exciting about it is that it's a model that scales, meaning that you can train a larger model from the same data set with the same architecture and even the same hyper parameters and you end up with high performance. And that's not something that has ever been achieved in time series models before. We have been achieving very famously in language models, starting like the most visible case of GP2, and we know what happened next. I think there's -- we start to see the same opportunity with these models. So now we're busy basically extending those models beyond time series, when I say time series. -- are mostly timestamps and values. We're extending that towards including the rest of observed data, whether that's logs and traces and application topology and network data and user data and all of that stuff, so we can really fully model and be predictive on our customers' infrastructure. And the idea there is if we build that properly, we can automate a lot of our customers' operations.

And I think the point being that like an LLM is designed to kind of represent human language. This is a completely it's a semi structured set of data that you would love to have a model like an LLM, but now this is a machine learning model optimized for that. As a result, its performance is much better than you can ever get out of LLM.

Yes. And also, these are -- like these are very different types of LLM are very bad time series for example. And so these are -- the architectures are even though the models are transformer based, so a lot of deep learning, transforms a lot of it is similar, but the architecture of the mall is actually fairly different.

Yes. Well, LLMs don't understand ordinality. So like you need a model that has that as a structure of it for this to work in different ways. We can obviously go much deeper on some of these things. But before we run out of time, I think it's also like important to perhaps reflect on some of the deployment model flexibility that you've been providing customer I mean it's almost like exhausting when you think of like the level of innovation you guys have been able to put out. I think that you could probably help us reflect on like from a buyer standpoint, 1 of their frustrations when they're forced to have a single deployment model that may not be ideal, whether or not it's for data residency or credits they already have. So maybe help us understand some of what you're doing with bring your own cloud and some of these models.

Yes. I mean -- and just to zoom out, customers in general, they face just an explosion of the amount of data and applications and everything, right? So -- and so there's -- in absorbability, there's a perpetual quest for efficiency, and making sure that would you spend on measuring auditing, rating your application remains predictable and within range of what you were expecting. For that, there are 3 core mechanisms. One is you need to have a feedback loop that lets you understand what you're consuming and what's valuable and what's not and what you need to do more or less of. And that's something we built big part of the product is making sure that the right people have the feedback loop at the people who care and who capture their information, see exactly how much value they're getting out of it and how much is costing them. So that's 1 part.

Another part is to be to get ever smarter on how much of a sample you need to understand the data. And that goes back to all the smart building the models, et cetera. etcetera. And the third part is to keep innovating on how you can, how efficient you can be at storing and acquiring and managing the data and all the various topologies you can employ to make that work for various customers. So on that last one, One thing we started doing last year, and we see quite a bit of demand for and we're accelerating on is, we are doing some we call bring you on cloud options for our customers to deploy our product. meaning that they can store the data on infrastructure that they manage themselves even though we still run and update the application for them.

And the reasons for that are -- so we see cost at large scale is 1 of them actually have petabytes of data, that's something that you probably won't. Other reasons would be data resiliency lows, and we see the world in general is fracturing into smaller zones as opposed to one big go. So we expect a lot more of that in the future. Even though today, it's still a native for most customers, we expect that to be more front of mind, maybe 1, 2, 3, 5 years. It's hard to tell when exactly. And the last reason might be that customers have -- maybe they have a large data center, they already built and they want to use. Maybe they have a large commitment with a specific cloud provider, and they want to use the what they've committed already for that. Like there's all sorts of reasons why customers might want to control more of the deployment there.

Historically, we had focused on being purely SaaS and there were a few reasons for that. One is, I mentioned earlier that we thrive on being product-led and product adoption, meaning it's very important for us to be able to iterate very quickly. and also to understand what customers are using and not using and what's valuable and valuable. .

So we never offered an on-prem version of our software early on for that reason. Like we thought it would be a huge tax on innovation. And also it would make us a lot more a lot number in terms of who we build product. And we've seen that play out. Like most companies that start having a product that has some success in the enterprise, get offers, they can't refuse early on of having buildings on-prem versions of their product. And that's a great idea in the short term and a horrible idea in the long term because those companies end up slowing down a -- the way we built this technology now is bring you on cloud, allows us to have the best of both worlds because we still manage and run and update those applications and meter those applications. But at the same time, we can run a data plan fully within control of our customers and on their infrastructure.

And between this and some of the stuff you guys have done with FedRAMP perhaps this even leads to some of the opportunity you could see the scale around because like if I look at many of the other software companies I cover, somewhere between 7% and 15% of their revenue come from federal. And I think you guys are -- if David's comments earlier, we're accurate closer to like 1%. So you could see even just on that 1 customer category, the type of revenue opportunity that it opens. We only have a couple of minutes left. You talk to a lot of investors, be talking a lot of these different events. What do you think the most underappreciated aspect of Datadog's forward-looking opportunity is?

I mean, there will be 2 halves to it. 1 is the world is exploding where they are right now. And this means so much more stuff so much more complexity, so much more infrastructure. And we see that play out with the large dialytic today, but that's going to come to the rest of the market. And so that's a huge opportunity for us. If there's only 1 category that remains, and I know everybody here has been debating the software ever going to be worth anything ever again. Well, I mean the answer Genies yes. But the -- if there's only 1 category that is left in the world, it's observability because the 1 job that remains for us humans is to control the machine and understand what it's doing, understand whether it's delivering the right outcomes, how much it's costing and whether the intent of the -- of what we wanted to do is actually what would get performing in the end.

So I think that's a huge opportunity for us. There's a lot to be there. It's a challenging market because it's changing really fast. So there's a -- that muscle we built on product innovation, I think, is going to be brought to bear, but that's -- we see that as a huge opportunity. The other half, the more boring half is that all of that is driving the need for standard observability. Those large AI companies are consuming large amounts of infrastructure and applications in the same way the other companies are -- and as I said earlier, like we have less than 14% of that market, and that market is growing and is going to grow faster, I think, now as the -- as you see this huge CapEx investments in infrastructure and general and it's not use GPUs anymore. It used CPUs and everything else as these agents need to use tools.

And so I think we -- that part of the business, the boring the existing part of the pan, that's not new form factors, your plication or anything. That part has another 5 or 10 x into it, and that's very exciting.

Well, you heard it here first. The only business that will remain after the robots take over everything else is observability in Datadog. So make sure that you're ready to be an SRE in your future. Oli, thank you so much. I really appreciate you taking the time here and sharing with the audience your experience and kind of where Datadog is going.

All right. Thank you.

Welcome, everybody. Thank you for joining. It's a great pleasure to be here with Olivier Pomel, CEO and Co-Founder of Datadog. Olivier, thank you for joining us.

Great to be here.

I think we'll start off on a great note. This time last year, you were on stage with us. We highlighted that you were one of only 4 enterprise software companies growing mid-20s plus. We said something very rare was happening at Datadog a year later and looking back, I think maybe we understated that a little bit. You're now growing over 30% at a $4 billion scale. So can you help us in layman terms understand the problem that you're solving out there? Why it's so critical for customers and what do you think the phenomena that are fueling your growth at scale?

Yes. So I mean, look, we do observability and security. So we sell to engineers and product teams and our customers. We help them understand whether the software, the services that they're shipping are actually working. They're working as appropriate for their customers, if they are fast enough and if those products deliver the right business value for them and for their customers. So that's fundamentally what we do. We serve every type of company from the 10-year startup, the newest AI company all the way to the largest and the oldest enterprises and with pretty much everything in between.

The reason why we see demand, I think there's 2 aspects. One of them is we're in still fairly early in a super cycle of digitalization and cloud migration. So we started the company 15 years ago. It was right at the beginning of the cloud migration. That's still ongoing. And that's still a big driver of our business and still something that's going to keep going for many, many years. And then the second aspect is, I'm sure it's not lost on anybody in this room that everything is changing with AI. There's a lot more that is being built that is being shipped in software. There's a lot more interactions that are being automated. And all of that creates new kinds of complexities and new kinds of surfaces that and also, frankly, quite a bit more infrastructure. And all of that needs to be managed, monitored, observed, secured, and that's what we do.

Yes. Yes. I was preparing for this. I want to say you guys made it easy to come with some of these questions because you had such a good Q1. Yes, I think one of the best prints we've had this earnings season, 32% revenue growth rate, as we mentioned, that's $4 billion scale, accelerating for the fourth quarter in a row. Largest sequential add in a while, a very, very healthy raise, really just a lot of strength across the board. I'd love to kind of decompose that a little bit. Where are you seeing those pockets of strength? I think it's more than just AI, right? Maybe what surprised you? And what do you see kind of persisting going forward?

Yes. So look, the -- as we said on the call, I think on the earnings call, we saw acceleration across every single part of our business. We saw acceleration with the brand-new AI-native companies, whether they're small or very big. We saw also acceleration with the rest of our business, which is even more interesting. So the non-AI part, all the companies that were around before AI started taking over. I think the drivers are a little bit different there. I mean, on the AI side, obviously, we see that part of the ecosystem blowing up, like the AI is getting into production, some use cases just coding are very real and scaling very fast.

And this is feeding a number of large model companies but also all sorts of application companies that are being built around that. So I think the drivers are fairly clear there. What's even more interesting for us is that the bulk of our business, the rest of our business that is -- that existed before AI became a thing. That business is accelerating as well. And part of that is those companies have to -- they understand they need to modernize and modernize faster so they can be ready for AI. Part of it is just that we're still very early in cloud migration. There's a number we like quoting, which is that I think Gartner has this report every year on the market share in ITOM and a number of other fields. And we're the leader in observability, we're the #1 there, but we still have only 13.6% of the market according to them.

So this tells you how early it is in the market for us, how much opportunity there is ahead of us, even factoring out all of the new developments with AI, all of the new explosion of demand we can see there. And so we think we're still early in a super cycle there.

Yes. Yes. And definitely, some threads will have to pull on in a little bit. But one topic that's been out there for the entire time we've covered you all is this concept of customers doing it themselves, using open source now or building it. And it's been persistent, even though none of the data points really supported that much. I think your retention is stellar. But this last quarter, you guys started talking about getting the hyperscalers as customers, right? And if there was anybody who could do it, they do have it, right? They have these tools within their own ecosystem. So if we think about it that way, like if even the hyperscalers concluded, they need to buy from you rather than use what they have. I mean, was that surprising to you at all? And what do you think that says about the value proposition that you have and even the moat that you've built?

So I mean I wouldn't call it surprising, but it's definitely a very interesting proof point because the way we see it is the reasons to build it yourself are usually mostly cultural. That's because you want to build yourself, you want to have your teams do it or somebody on the team wants it so bad and nobody else around them wants to prevent them from doing that. So that's cultural. Typically, you don't get nearly as good of a result when you do that. For one thing, it takes a lot of time and a lot of focus that you should spend somewhere else. So typically, when you try and build it yourself, you end up with a solution to -- in 2 years from the -- to the problem you had last year. And that's typically not great.

You also end up -- justify what you might think when you go into it, much worse economics, and that's not a fantastic solution long term. That's something that many of the hyperscalers, again, are happy to live with because for cultural reasons, they want to in-source everything. Now what we see happening now is there's an extremely competitive situation right now around the development of AI and I think it really focuses the mind for many companies. When they realize, "Hey, wait, actually, instead of waiting 2 years for this, we can have it now, and it's going to run better and cheaper in the end." So what are we doing? What should we be doing there? So again, our business model in the end is not to serve the hyperscalers. There's not enough of them. They are too fickle also for cultural reasons, as I mentioned earlier.

But for us, it's a great proof point. Even the companies that have virtually unlimited access to top talent and a strong cultural bias for building use our product. It means that it really doesn't make sense for anybody else to build their own.

Yes. Yes. I think that it's an excellent ironman argument for what you guys have been differentiated. And I just had a conversation with a partner where he made a very interesting point that I hadn't thought about, which we're asking about why coding all that. And their comment was AI has accelerated the pace of code being generated, and it's increased the competition because everybody can do it out there. And so the effect that it has is the opposite is now if you don't use something that's really trustable and powerful like Datadog, you might actually be hindering your development process, and that's your bread and butter. So in fact, the effect is that it makes people more reliant on something like Datadog. Do you see that dynamic?

Yes, definitely. And look, most of the -- actually, pretty much all of the top 8 or 10 coding companies or -- coding companies, whether you're talking about the models people use in production or the large companies or the products that are built more for the consumers, all of those companies use Datadog behind the scenes. And so that tells you about the very specific need there is there. Look, at the end of the day, we see there's so much more stuff that is being built, it's being produced so much faster that, by definition, the folks who produce or that have no idea how it actually works.

That's the least understood thing about productivity is that the more you increase productivity, the more complexity you create because folks manipulate way more things in way less time and as a result, doesn't go through their brains and they don't understand what's going on so they need help to actually understand what happened, to actually make sure it works properly, to make sure it delivers what it's supposed to deliver for the business in the end to make sure it keeps working, when everything keeps changing around it to make sure it's secure, and that's what we do.

Yes. I think a great point to kind of get into the AI-native cohort that you guys have 22 AI native spending more than $1 million and 5 spending more than $10 million. And a lot of the focus tends to go on some of the larger ones. But I think the impressive part is how diverse even that has got foundation models, codegen, Vertical AI. So when you look across your book, where in the value chain do you kind of see the strongest opportunity? And what is the differentiation you have for those customers?

So I mean, those customers have like a high pressure to deliver a lot and deliver very fast. And look, the core of what we do is what they all start with. So we cover everything from end to end from the bits that go through the CPU, the network, the GPUs all the way up to the end users, how they're using a product, whether they are coming back, whether they're completing what they're supposed to be doing, how much business value they generate for you in the end. And we cover absolutely everything in between in a way that's fully integrated. If you go to any of our competitors' website, they all say they do it, the reality is it's actually really, really hard and really differentiated to do it well, and that's why those companies all use us.

This is -- when you think of the investments they are making, whether it's on their engineering teams, their research teams, their GPU fleet, all of that goes to waste if you do not build -- understand how to build the right thing and then ship an experience that works for your end users.

Yes. Yes. The need to perform is the highest for them. I think one of the most striking comments that you had on the Q1 call was the change in posture with regards to training. I think the quote was last year, we said training was not really a market for us yet, and we're actually starting to see it become a market. So that's -- it's a big step function change, I think, in people's minds. When did you start to see this happen? Was it -- maybe you don't like the word surprise, but was is it surprising to you guys? And what do you kind of see about that opportunity looking forward?

Yes. So this one was actually a bit of a surprise, yes. So it made intuitive sense to us before that training might become a market and because we saw -- look, a while ago, training was mostly pre-training. It only made sense for 5 to 10 companies in the world to do it, it was large scale. It was completely homegrown not a great fit for building a product typically. But we saw that the technology changed quite a bit. So models went from being mostly pre-trained to being largely post-trained. The post-training was becoming increasingly specialized to different types of verticals. The stacks that are used for post-training also are becoming richer and richer. So now when you post-train a model, you're going to run all of those different environments, basically, you're going to run all the applications you can run, simulate behavior in those applications, capture what comes out.

So you end up running way more complex tasks for doing that. And then we saw -- instead of having 5 to 10 companies doing that now there were 50 to 100. So intuitively, it made sense to us that something might be more interesting there. We were surprised, though, to see a number of different customers, including multiple hyperscalers come to us for training at about the same time and which tells us there's something happening and there's potentially the emergence of a new market there. Again, too early to call it because the technology is changing fast as the markets are changing fast. But if we get into a situation where we go from 50 to 100 of these companies to 500 to 1,000 to maybe more, then that becomes a really interesting market and there's a real problem to solve, and we can do it.

The other thing that's interesting, if you look at the evolution of the technology is we're probably going towards models that can learn on an ongoing basis. So instead of training a model, whether that's pre-training, post-training, if you keep improving your model with online evaluations and online improvement, then that becomes really very much ongoing production concern and something that you can repeat in many, many, many different companies over there. So we are hopeful, interested in that -- it's not the core -- it's still small compared to the inference business compared to everything else that happens in the stack that our customers are running. But it's an interesting new green shoot, I would say.

Yes. Can you -- do you have a sense you said they kind of all came to you at the same time. Do you have a sense of what the trigger was from their perspective? Was it a lot of this stuff going to post training? Is that just something that makes more sense for you guys that don't have solutions for it like, what was the catalyst?

I think the -- a lot of these companies are on a bit of a similar clock. So everybody woke up to ChatGPT at some point and then there were a number of internal efforts started in a number of different places. Some of these efforts worked, some of these don't. Yet you have these cycles that companies live through and when something didn't work exactly the way you want, initially, then you try and reset and do things differently. I think part of it is that part of it is companies being somewhat aligned on that on the same because they all compete in the same market.

Yes, yes. Okay. So let's think about the fact that blossoming out of AI, right? Where we're seeing this move, like you said, beyond just a handful of companies. I think there's been a lot of indications from people across the stack in the software space talking about that hinting at it. I think maybe that's underappreciated by a lot of people. Like I said, a lot of focus just goes to those kind of very well-known LLM companies. But could you provide your thought on that? You said it might democratize quite a bit more, right? So both on the training and inferencing side, whichever one you going to dig into, like how are you seeing that democratization happen? Are we literally in the very first inning of that? How do you see that curve growing?

I think we're still super early. I think the focus for most companies or most of users of AI today is to make sure it works. So everybody is getting from -- step one is, let me make it work once and then step 2 is, okay, so now let me deploy that at scale across my company across similar use cases and things like that. I think we're still very much in that phase. We are not at all in the phase of, okay. So now what else can I build on that or how do I rationalize it? Or how do I optimize, that's for later.

That later might come sooner if we see the explosion of the AI company's revenue continue at the rate it is right now, that money is coming from somewhere. And it's -- there's probably going to be a push for rationalization sooner rather than later. But the mode right now is still very much, let me get it to work. Our mental model for what the market looks like in the end is that it's similar to what you see in the overall cloud infrastructure or even the database market. Like database market you have a number of options up there. You can have -- you can buy a closed-source database that you're going to run yourself. You can buy closed-source cloud databases that you don't run yourself that are completely black boxes or turnkey and if somebody else is running them for you.

You can use open source one that you run on infrastructure somebody else provides for you. You can run the open source on your own infrastructure. You can build your own databases and all of those actually coexist like there's reasons for all of those to exist and customers typically are going to mix and match a number of those. They're going to have different cloud providers they're going to have different database providers, you're going to have a bit of everything in there. My guess is we're looking at a market that looks like that for AI inference.

Yes. I want to get to non-AI cohort accelerating, which I think is, again, a big part of the story, but one last one a couple of years ago, we kind of gave you this analogy, at least that was our thinking, which training is like a bottle rocket and inference can be like a steady compounder to get the time you kind of agreed with that framing. We touched on it a little bit, but do you think that maybe that framework has changed a little bit where the training might be a little bit more of a steady compounder on its own as well?

So it's inching closer to being an ongoing recurring thing but it's still a little bit one-off, like what we see still you have these large training runs or these smaller training runs, but there's still easily run like you do a run, then you do another one and then maybe nothing happens for a few days, then you do another one. And these also tend to be still somewhat like custom coded, like you don't have like a standard way of doing it that every single company out there is using. So it's a big improvement from where it was a couple of years ago where again, only a handful of companies were doing it. It was extremely hand-coded, absolutely not production minded in general like it was kept up basically by people babysitting those training jobs night and day.

And being like very large pretraining runs and then nothing after that. But we're still not yet at the point where it's an ongoing, always on every day of the week, live with customer data kind of operation. I think we still don't know whether the market is going there. We think it might be, but that's for the future.

Yes. Got it. So moving to the non-AI cohort. So the core customers they've been accelerating, I think, every quarter for a little while now. As you've said, as we've said many times, your story is about a lot of customers and a lot of things going in the right direction. But it has been accelerating. So I'm sure the pieces are kind of similar, but what's changed? I mean, a year ago to now, what is going on in this core drivers that's pushing that a little bit higher?

Yes. I mean, look, so we do have some proof points that those customers are starting to adopt some AI in production. It's still small relative to the size of those customers and still a small driver of our growth there. Like we see traffic, for example, in our MCPs and traffic also to our LLM observability product. We see a very strong growth. I think we released some numbers in the earnings call on that. But that's still a small amount, but just give us an idea that these customers are moving there. But the bulk of it for these customers is that they are still modernizing, moving to the cloud. We get into more of them because we have successfully built up sales capacity.

We go to market in more regions and more segments. We get great return on investment on that. And then we've been expanding the product -- set of products. We can sell in the categories we sell into. And we have enough of those products that are reaching product market fit, getting to inflection points in their growth and where we basically see great adoption and consolidation from customers in our product. So I would call it the boring side of the business. It's not AI revolutionizing everything. But it's a predictably high return on investment, very buildable part of the business where we keep building those products, it makes sense to our customers, and we keep investing in the sales capacity because we are still early in what is a very large market opportunity.

Yes. Yes. I wanted to touch on that bit, SRE. Even before you guys have launched that, I talked to a few people and they're pretty excited about it, saw a lot of value in it. I think you guys have a 100,000 investigations since launch, 1,000 customers. Really going well. I'd love to hear what you see on that front in terms of customer utilization. And one thing we do here is that the more autonomous, the technology, the more powerful it is, customers do have a bit of a challenge of how do you integrate that into the workflow, because you kind of have to change how you run your business. So is that something you see as a real hurdle? Or are there customers out there who are kind of really pushing the edge there?

I do. So we see a ton of pull for it, and it gives us a lot of areas to develop in the product. So one thing we keep hearing from customers is I want it to go faster into auto resolution. So initially, we were worried that if it does too much, it creates a trust issue and it's hard for customers to control. But with AI taking hold pretty much in every single part of our customer's businesses, I think that train is leaving the station now and folks are getting comfortable with automation. And so they are pushing for more and more end-to-end automation with it. So maybe -- okay, it's fine. You told me what was the issue? Why it was there? Who introduced it, how to fix it. Now give me a button to fix it or maybe even better, fix it for me and then tell me about it. So we're working on that.

A second area of pull we hear from customers is, okay, this is great. This works really, really well. But I wanted to work across my other systems. So it should work services within Datadog, but also I'm using some other logging system. I'm using some other security tools, I'm using all these different things. I have a lot of open source, can you work across all of that. So we're also pushing so that Datadog actually connects to all those bits and investigate across all of those different systems. Another area we're investing in is getting more proactive and predictive. So we don't have to wait until you run into an incident to investigate, maybe solve it or prevent it even and so there's quite a bit of R&D that goes into that.

So we're developing our own models for that. We've released actually 10 days ago or the second version of our time series foundation model called TOTO and you can look it up. There's a blog post and some results that we've published there. This model is open-weights, but we're also being adopted by the community as well. And what's really exciting there is -- so it's a time series model. It's a general purpose, even though it was trained almost exclusively on observability data, it performs extremely well on everything else. Actually, it's state-of-the-art across all time series use cases, not just observability. And it is a competitive field. Every single large company has time series models. So it's very exciting for that reason. It's also very exciting because it's a first time series model that shows scaling, meaning we can train it with more data and train larger models for longer and we get better performance.

It used to be that it didn't work for time series model. Like we all know that what started the current AI revolution we're living through is the fact that we saw large language models scale, like we saw, and I think it was started with GPT-2, we saw that you could throw more GPUs and more data and you get better results. I think we're getting to that point with time series models. So it's very exciting for us because that's the path for us to get from, okay, we run an investigation after you had an incident to we're going to be predictive. We're going to understand what's going to happen next in your systems, and we can run AI directly within our data plane without having to get outside of Datadog, which is very exciting.

So that change in kind of being able to feed more data that could really kind of bend the curve on the capabilities.

That's what we're working on. Yes. That's exciting. And again, it's research. So -- but if research is good enough that we can open with it, we can open it up, and we see a large amount of adoption from it.

And what do you think that would look like once you start being able to do that, I think, is it charging for like issue that you've proactively resolved? Or how would you think about that?

I mean right now, we charge for investigations for it. Long term, we don't actually know what the model is going to be in part because the broader market is still figuring out how to package intelligence. And so it's not clear yet what customers will relate to the most there. And so I guess we'll see. The short of it though is, for us, it doesn't really matter because we have a usage-based business model in general.

And so it doesn't matter whether we have a new dimension in usage that relates to in this particular type of investigations or whether that gets attached to other parts of usage we have, whether it's on the data volume we process, the number of events, the footprint our customers have in the cloud, like there's a number of different ways to look at it. So for us, it's not a big deal either way.

Yes. Let's talk a little bit about your R&D, your headcount and kind of how you're handling that. You guys are obviously kind of very forward-leaning. We had a discussion last night at dinner about how you guys are kind of thinking about putting productivity and R&D through people and then through code, tokens, however you want to think about it. Can you just frame that discussion and how you guys have set it up?

Yes. So I mean, the short of it is we are currently -- like if you think of what drives or limits our growth. We're limited and driven on 2 sides. One is the sales capacity we have and we're still early there compared to the number of markets and segments and customers, we need to be talking to. So we still have to grow that capacity. That one is still largely human-driven like the buyers are humans maybe at least for the foreseeable future, the humans will be buying. We'll see if that changes at some point, but that means the sales capacity is largely driven by humans.

And we're still growing that as fast as we can. We get great return on investment on that. We're also driven and limited by the number of products we have in relevant categories with the right amount of functionality and quality. And that is driven by our investments in engineering and R&D. Historically, that's been mostly about human labor and how many people are in the team. Now maybe there's a bit of a mix shift maybe you will have the same overall R&D investment, which right now is around 30% of the top line, but some more of it will go to tokens or GPUs and less of it will go to labor, I think it's unclear, but we know we'll keep investing. And we also know that right now, we're also -- we keep hiring. We think we can still scale. We need to still need to scale, and there's still a lot more we need to be for our customers and a lot more demand out there.

When you say that you're growing as fast as you can in terms of the sales and marketing side, is the limiting factor, just your ability to hire people?

Well, when you talk about sales and marketing and sales capacity in general, it's -- you can't just think of it in terms of a big pool of labor. You need to think of it in terms of the right people with the right territories in the right segments, everywhere around the world. And so it's way more of a bottom-up kind of a thing than a top down. And so -- and that's actually really challenging to do well and at scale.

Yes. Zooming out a bit at the end here. Everybody's kind of seen the -- all the models. There's an LLM du jour, it seems like ChatGPT, Gemini, Claude, so on and so forth. Do you see that leapfrogging effect? Do you expect that to continue. What is your view on how that's evolving?

So it's a super competitive market right now, which is fascinating because these are products that, on one hand, look like commodities because you can hotswap them pretty easily and from a far, it's hard to tell which one is better from the consumer. But at the same time, it's an incredibly competitive market where improvement is very rapid. Similarly to what we've seen with the cloud like 10 years ago, like there was a fear at some point that Amazon was going to run with everything, and they would compete with everyone in every single field. We've had a little bit of the same and it didn't turn out that way, right? I mean it's a very healthy multi-vendor, very innovative marketplace where you have these large-scale vendors that provide things that kind of are largely commodities with some differentiation, but not all that much.

And it's a very dynamic marketplace. My guess is we're going to see the same happen with the AI models. I think we're well on our way to that already, there is multiple vendors, they compete. There's a lot of innovation. There's a range of frontier paid versions. There's a range of open source versions that are lagging behind, but not crazily behind. And so we -- my guess is that we're not going to be in that situation either. What's interesting, though, is that for customers and users of those models, it's really hard to understand how well they work, how they compare, whether they still work the same way they used to, whether something has changed, whether something would be more appropriate for them.

And we think it's a big opportunity for us and for observability in general, like our job among many other things, is to tell them, "Hey, that thing you're using, it's actually not performing as you thought it was or it changed or something else changing is better or this is where you can mix and match them and I think it creates a long-term opportunity for us." The same way we've had an opportunity with the various kinds of clouds being mixed and matched by our customers.

Yes. Another vector of growth. I think at our 2024 TMC conference, you had this comment where you said, it doesn't take a lot of imagination to see how we can get to 5 to 10x the size we are today. I think you were about $2.5 billion run rate then and now you're at $4 billion and the comment back then were like, listen, the ITOM market is growing at a healthy rate, and we're winners in that market taking share, so we're going to grow even faster. We've talked about AI in a bunch of different ways today. Do you think that's transforming that TAM that you have? I mean does that change this concept of how big you can get?

So I mean, on the first part the part about the lack of imagination, I can -- we can still grow the same way without imagination. We're still in a similar position. I think I remember that time -- I remember the Gartner number we quoted at that time was around -- had around 10% of the market. Now we have around 13.6%. And so the math that gets you to the 5x is pretty much the same as it was at the time. And that's without entering massive new categories or without huge tailwinds in AI.

Now the thing that's new now is this explosion of AI and we think it opens up a number of other opportunities. In particular, we can do so much more and we should do so much more in automation, which allows us to deliver a lot more value to our customers. And we think observability was only ever one half of a solution so yes, okay, you've observed, but now what would you do? And I think there's a lot more we can do to automate. Yes. So that's the exciting part.

Yes. We're coming down to the last few minutes here. We'd love to ask this question, which is when you're hopefully back here in 12 months, what do you think that the audience is going to know or see come to fruition that you're kind of seeing right now? Like what's going to surprise that you're kind of already seen in the pipeline and not discussed as much as it should be.

I mean, look, the -- I don't know if it's going to be truly surprising to anybody here, but the amount of stuff being built and the way the development cycles are collapsing is creating so much opportunity for us. Just on one hand because of the sheer volume of stuff that's coming up, but also on the other hand, because the value is moving from the act of writing code to everything that comes before and after. So what should I work on? How do I know it's working? How do I know it's safe. How do I know it's actually -- my users actually find value in it. All of that creates such an amount of -- such amounts of opportunity that, again, I don't think it's always understood by the market, especially some days where [indiscernible] these trends, definitely, is understood at those times. But I think it's -- for us, it's a huge driver of short and long-term growth.

Yes. And we've had a lot of such days between now and the start of the year. We've got just a few seconds left. Anything on security, I mean, it's one of those areas that's growing very fast in the background. Where do you see that going?

So it's exciting. I think there's 2 areas in particular where we find great satisfaction over the past couple of quarters. One is code security. I mean that whole field is upended by coding model. And so there's so much more to be done there, and we find a great amount of demand for it. So that's exciting. The second area, we find a lot of success in these -- our cloud SIEM product. And with resonates really well with the market there is a combination of 2 things.

One is it's a really top of the line, event management, storage, log management system that lets you stream data from anywhere to anywhere, storage in all sorts of extremely efficient way, curate very well. And that's because it comes from our observability product. So it's already way above what a pure-play security company would do, for example. And we combine that with the surprisingly good best AI security assistant. I say surprisingly because we were surprised by it, and we find that it constantly rewards customers when we show them that. And it's fairly unique. You don't see anybody in the market that has this combination of the super good data back end with the super flexible with the AI on top of that. Most companies try and focus either on one or the 2. And so that resonates very well in the market.

Yes, full solution. Thank you so much for your time and your insights. Really appreciate it.

All right. Thank you very much.

Good day, and thank you for standing by. Welcome to the First Quarter 2026 Datadog Earnings Conference Call. [Operator Instructions] Please be advised that today's conference is being recorded.

I would now like to turn the conference over to Yuka Broderick, Senior Vice President of Investor Relations. Please go ahead.

Thank you, Lisa. Good morning, and thank you all for joining us to review Datadog's first quarter 2026 financial results, which we announced in our press release issued this morning. Joining me on the call today are Olivier Pomel, Datadog's Co-Founder and CEO; and David Obstler, Datadog's CFO.

During this call, we will make forward-looking statements, including statements related to our future financial performance, our outlook for the second quarter and the fiscal year 2026 and related notes and assumptions, our product capabilities and our ability to capitalize on market opportunities. The words anticipate, believe, continue, estimate, expect, intend, will and similar expressions are intended to identify forward-looking statements and similar indications of future expectations. These statements reflect our views today and are subject to a variety of risks and uncertainties that could cause actual results to differ materially.

For a discussion of the material risks and other important factors that could affect our actual results, please refer to our Form 10-K for the year ended December 31, 2025. Additional information will be made available in our upcoming Form 10-Q for the fiscal quarter ending March 31, 2026, and other filings with the SEC. This information is also available on the Investor Relations section of our website, along with a replay of this call. We will discuss non-GAAP financial measures, which are reconciled to their most directly comparable GAAP financial measures in the tables in our earnings release, which is available at investors.datadoghq.com.

With that, I'd like to turn the call over to Olivier.

Thanks, Yuka, and thank you all for joining us to go over a very strong start to 2026. Let me begin with this quarter's business drivers. I'm very pleased to say that our teams executed very well and delivered revenue growth of 32% year-over-year, accelerating from 29% last quarter and 25% in the year ago quarter. We showed broad-based acceleration of revenue growth across cohorts, including both our AI and non-AI customers. Our AI native customers cohort continue to grow and diversify rapidly, both in the number of customers we serve and the scale of those customers. In this quarter, including new land deals with 2 of the world's biggest AI research teams, helping them improve and optimize their training workflows. I'll talk more about that in a bit.

Even more impressive was the growth in our non-AI customers. Non-AI customer revenue growth accelerated again this quarter to mid-20s percent year-over-year up from 23% last quarter and 19% in the year ago quarter. We think this is a sign of strong continued cloud migration, greater adoption of our products, and customers of all kinds accelerating their use of AI. Finally, churn has remained low with gross revenue retention stable in the mid- to high 90s, highlighting the mission-critical nature of our platform for our customers.

Regarding our Q1 financial performance and key metrics. Revenue was $1.01 billion, an increase of 32% year-over-year and above the high end of our guidance range. We ended Q1 with about 33,200 customers, up from about 30,500 a year ago. We also ended with about 4,550 customers with an ARR of $100,000 or more, up from about 3,770 a year ago. These customers generated about 90% of our ARR. And we generated free cash flow of $289 million with a free cash flow margin of 29%.

Turning to product adoption. Our platform strategy continues to resonate in the market. For example, 56% of our customers now use 4 or more products, up from 51% a year ago. 35% of our customers use 6 or more products, up from 28% a year ago and 20% of our customers use 8 or more products, up from 13% a year ago. So we're landing more customers and delivering value across more products. And our business continues to grow. Our total ARR now exceeds $4 billion, and our quarterly revenue exceeded $1 billion for the first time in Q1.

This is a big achievement for all of us at Datadog and is a product of years of investment in building and innovating for our customers. But we are still just getting started. Of our 26 products, 5 are over $100 million in ARR, another 3 are between $50 million and $100 million ARR. We're working hard to build and deliver further growth in those products. And this leaves 18 other products, which are earlier in their life cycles. We believe each has the potential to grow to more than $100 million over time.

Moving on to R&D. Our engineers enabled with the latest AI coding tools are building rapidly to help our customers confidently and securely deploy their applications. So let me speak to a few of our product launches this quarter. Let's start with AI. As a reminder, we're talking about our AI efforts in 2 buckets: AI for Datadog and Datadog for AI.

So first, AI for Datadog. These are AI products and capabilities that make the Datadog platform better and more useful for our customers. In March, we launched our MCP server for general availability. With MCP server, developers access live production data to debug their applications directly in their AI coding agent or IDE.

We delivered Bits AI Security Agent, which autonomously triages Datadog Cloud SIEM signals, conduct in-depth investigations of potential threats and delivers actionable recommendations. We've seen Bits AI Security Agent reduce investigations that could take hours to as little as 30 seconds. We also shipped Bits Assistant now in preview, which helps customers search and act across Datadog using natural language prompts.

Moving on to Datadog for AI. This includes Datadog capabilities that deliver end-to-end observability and security across the AI stack. We launched GPU monitoring, enabling teams to understand GPU fleet utilization, workload efficiency, thermal and power behavior and interconnect performance. This drives higher GPU ROI and operational reliability.

Our customers continue to move forward with their AI activities, and we can see that in their usage of the Datadog platform. We now have over 6,500 customers sending data for one or more of our AI integrations. Though this is only 20% of total customers, they represent about 80% of our ARR.

And our customers' usage of AI within Datadog platform continues to grow rapidly. Bits AI SRE agent investigations have more than doubled from December to March. The number of spans sent to our LLM observability product nearly tripled quarter-over-quarter. The number of Datadog MCP server tool calls quadrupled quarter-over-quarter and the number of Bits Assistant messages increased by a factor of 12 in that period.

While we are aggressively building with and for AI, we also continue to expand the Datadog platform to deliver against our customers' increasingly complex needs. To speak to a few of these efforts, last month, we launched experiments for general availability. Experiments work hand-in-hand with our feature flagging product and combine best-in-class statistical methods with real-time observability guardrails so companies can test for impact, choose among alternatives quickly and ship with confidence.

In addition, our customers now benefit from APM recommendations. By analyzing telemetry data from application performance monitoring, real user monitoring, profiler and database monitoring, APM recommendations automatically identify performance and reliability issues and most importantly, explain how to fix them.

And we announced our plans to launch our next data center in the U.K. We see a large opportunity to serve our British customers as cloud adoption accelerates in regulated industries. Last but not least, we are pleased to have received FedRAMP High certification from the U.S. federal government. With this certification, we can now move forward with federal agency customers that require FedRAMP High to handle sensitive workloads.

Meanwhile, we continue to expand our product offerings, go-to-market teams and channel partnerships for public sector customers, both in the U.S. and internationally. So our teams were hard at work again, and we're looking forward to sharing many new products and future announcements at our DASH user conference on June 9 and 10 in New York City.

Now let's move on to sales and marketing and highlight some of the deals we closed this quarter. First, we landed 2 large deals, a 7-figure and an 8-figure annualized deals with the AI research divisions at 2 of the world's largest technology companies. These organizations are building and training the most advanced AI models in the world. It is critical for them to reduce engineering friction and increase training velocity, but fragmented internal and open source tooling made it harder to identify and solve issues and reduce engineering and research productivity.

By using Datadog, both companies are accelerating their pace of innovation on their hyperscale AI training workloads. And this includes optimizing their workflows using GPU monitoring on large parallel GPU grids.

Next, we signed a 7-figure annualized expansion for an 8-figure annualized deal with a leading online recruiting platform. This customer is centralizing on Datadog to reduce complexity, drive developer velocity and improve efficiency. With this expansion, they will replace a stand-alone tool with Datadog LLM observability to correlate LLM signals with APM and user experience data. This customer will grow to 16 Datadog products, including Datadog MCP server.

Next, we signed a 7-figure annualized expansion for an 8-figure annualized deal with a Fortune 500 bank. With this expansion, this customer will migrate their remaining log data into Datadog, fully replacing their legacy log vendor. Most notably, our Flex Logs give them granular control over costs while meeting strict compliance requirements. This customer uses 10 Datadog products, including Bits AI Security Agent to accelerate incident response with AI.

Next, we signed a 7-figure annualized expansion with a leading global hedge fund. This customer operates thousands of on-prem hosts and network devices. At that scale, their open source monitoring stack has become operationally unsustainable, impacting portfolio managers and investment analysts.

With this expansion, they will replace their entire on-prem observability layer with Datadog infrastructure monitoring and network device monitoring and will have unified visibility across their cloud and on-prem environments. This customer will expand to 11 Datadog products.

Next, we landed a 6-figure annualized deal with a Fortune 500 insurance company. This company's fragmented observability stack led to long outages with incidents spotted first by their customers instead of their tooling. By using Datadog and consolidating 3 legacy APM tools, they expect to move from reactive responses to proactive incident detection. They will adopt 10 Datadog products to start, including all 3 pillars and LLM observability.

Next, we signed a 7-figure annualized expansion with one of the world's largest travel groups in APAC. This customer was using Datadog on one business unit, but in 2 others, they were juggling multiple tools and lacked actionable insights. By consolidating 6 legacy open source and cloud monitoring tools, the customer saved money and improve platform resiliency and performance. This multiyear commitment positions Datadog as their strategic observability provider.

And finally, we landed a 6-figure annualized deal with a leading Latin American fintech company. This customer serves tens of millions of users across critical financial flows. Their rapid growth outpaced their fragmented front-end monitoring setup and outages expose them to financial, operational and reputational risks.

By adopting our digital experience monitoring suite, including RUM, Synthetics and product analytics, they now have full visibility of user activity with the cost control they also previously lacked. This customer will start with 5 Datadog products. And that's it for our wins. Congratulations again to our entire go-to-market organization for a great Q1.

Before I turn it over to David for a financial review, I want to say a few words on our longer-term outlook. We are pleased with the way we started 2026 as we support our customers' inflection in AI usage and application development and as they lean into our AI innovations, including Bits AI Security Agent, Bits AI Security Analyst, Bits Assistant, Datadog MCP server, GPU monitoring and many more.

There is no change to our overall view that digital transformation and cloud migration are long-term secular growth drivers for our business. But we now have an additional secular growth driver with AI as we help our customers deliver more value with this transformative new technology. Now more than ever, we feel ideally positioned to help customers of every size and every industry as well as all types of users, whether humans or AI agents, so they can transform, innovate and drive value through AI and cloud adoption.

And with that, I will turn it over to our CFO, David.

Thanks, Olivier. This was a very strong quarter for Datadog. Our Q1 revenue was $1.01 billion, up 32% year-over-year. Our 6% quarter-over-quarter revenue growth is the highest for Q1 since 2022, and our $53 million quarter-over-quarter revenue added is the highest ever for Q1. That included the strongest quarter of sequential usage growth from existing customers since the first quarter of 2022.

We also delivered an all-time record for sequential ARR added to the quarter. ARR growth accelerated in each month of Q1, and we see a continuation of these healthy growth trends in April. We also achieved strong new logo bookings. New logo annualized bookings set a new all-time record by a significant margin and more than doubled versus a year ago quarter. These included wins in observability and included some of our newer products like security, data observability and Flex Logs. And our new logo average land size also set a record and more than doubled year-over-year as we continue to land larger deals.

Revenue growth accelerated with our broad base of customers, excluding the AI natives, to mid-20s percent year-over-year, up from 23% last quarter and 19% in the year ago quarter. We saw robust growth across our customer base with broad-based strength across customer size, spending bands and industries.

Meanwhile, our AI native customer growth continues to significantly outpace the rest of the business. This group continues to diversify and grow, including 22 customers spending more than $1 million annually and 5 spending more than $10 million annually. This group includes the leading companies in foundational models, code-gen tools and vertical-specific AI solutions.

Next, regarding our retention metrics. Our trailing 12-month net revenue retention percentage was in the low 120%, up from about 120% last quarter. And our trailing 12-month gross retention percentage remains in the mid- to high 90s.

Now moving on to our financial results. Billings were $1.03 billion, up 37% year-over-year. And remaining performance obligations, or RPO, was $3.48 billion, up 51% year-over-year, with current RPO growing in the mid-40s percent year-over-year. RPO duration increased year-over-year as the mix of multiyear deals increased in Q1. As a reminder, we continue to believe revenue is a better indicator of our business trends than billings and RPO given their variability.

Now let's review some of the key income statement results. Unless otherwise noted, all metrics are non-GAAP, and we have provided a reconciliation of GAAP to non-GAAP financials in our earnings release. First, Q1 gross profit was $807 million with a gross margin of 80.2%. This compares to a gross margin of 81.4% last quarter and 80.3% in the year ago quarter. As we've discussed in the past, our gross margin varies from quarter-to-quarter with investments into innovations for our customers, offset by efficiency efforts.

Our Q1 OpEx grew 31% year-over-year versus 29% last quarter and 29% in the year ago quarter. As a reminder, we continue to grow our investments to pursue our long-term growth opportunities, and this OpEx growth is an indication of our execution of our hiring plans. Q1 operating income was $223 million for a 22% operating margin compared to 24% last quarter and 22% in the year ago quarter.

Turning to the balance sheet and cash flow statements. We ended the quarter with $4.8 billion in cash, cash equivalents and marketable securities. Our cash flow from operations was $335 million in the quarter. After taking into consideration capital expenditures and capitalized software, free cash flow was $289 million and free cash flow margin was 29%.

And now for our outlook for the second quarter and for the fiscal year 2026. First, our guidance philosophy overall remains unchanged. As a reminder, we base our guidance on trends observed in recent months and apply conservatism on these growth trends. In addition, as with last quarter, we are applying a higher degree of conservatism to our largest customer.

So for the second quarter, we expect revenues to be in the range of $1.07 billion to $1.08 billion, which represents a 29% to 31% year-over-year growth. This guidance implies sequential revenue growth of $64 million to $74 million or 6% to 7% due to the strong growth of revenue in Q1 and into April. Non-GAAP operating income is expected to be in the range of $225 million to $235 million, which implies an operating margin of 21% to 22%.

As a reminder, in Q2, we will be holding our DASH user conference, which we estimate to cost about $15 million and which we have reflected in our operating income guidance. Non-GAAP net income per share is expected to be $0.57 to $0.59 per share based on approximately 369 million weighted average diluted shares outstanding.

And for fiscal 2026, we expect revenues to be in the range of $4.3 billion to $4.34 billion, which represents 25% to 27% year-over-year growth. Non-GAAP operating income is expected to be in the range of $940 million to $980 million, which implies an operating margin of 22% to 23%. And non-GAAP net income per share is expected to be in the range of $2.36 to $2.44 per share based on approximately 372 million weighted average diluted shares outstanding.

Finally, some additional notes on the guidance. We expect net interest and other income for fiscal 2026 to be approximately $170 million. We expect cash taxes for 2026 to be approximately $30 million to $40 million. We continue to apply a 21% non-GAAP tax rate for 2026 and going forward. And we expect capital expenditures and capitalized software together to be 4% to 5% of revenue in fiscal 2026.

To summarize, we are very pleased with our execution in Q1. We are well positioned to help our existing and prospective customers with their cloud migration, digital transformation and AI adoption journeys. And I want to thank Datadogs worldwide for their efforts.

With that, we'll open the call for questions. Operator, let's begin the Q&A. Thanks.

[Operator Instructions] Our first question today is coming from the line of Mark Murphy of JPMorgan.

Congratulations on an amazing performance. Olivier, is there any way to conceptualize the growth in the sheer raw volume of code that's being produced in the world today due to adoption of code generators such as Claude Code and Codex and Cursor because they seem to be developing the capability to take on full projects. And some of the charts are showing these capabilities are just exponentially exploding upward in a straight line. I'm wondering how much of that code is going into production and therefore, driving activity for Datadog?

Well, we definitely think and see that there's many more applications being created. There's going to be way more complexity in production. We see some of that happening already today. Some of those new applications are getting into production. They're finding users. We see some signs of that at every layer of our platform. We quoted a few stats on the increasing data volumes we see in our AI products. That's definitely a reflection of that. So we see an inflection point there in consumption from customers. We see a move to production that is very real, and we see that across both AI native and non-AI companies.

Okay. And just a quick related follow-up. If we click down one layer, and I'm wondering how you might view the increasing heterogeneity of the environment at the silicon level, because the -- when you look across Amazon, with Trainium and Graviton, Google with TPUs, Microsoft has launched the Maia Silicon. It looks like that is starting to explode. And our understanding is that trying to monitor the mixed environment is a lot more difficult than if you just have a uniform fleet of Intel and AMD chips. And we keep hearing all the traditional monitoring tools, they really fail on the custom silicon and Datadog handles it well. The -- and then all this new telemetry, including high-bandwidth memory and that type of thing. Can you speak to whether that trend is giving you some tailwind?

Yes. I mean, look, the broader market that's interesting here is training, the training used to be something only 2 or 3 companies were doing or maybe 4 or 5 at a large scale. And it looks like training actually might democratize quite a bit more, and many companies will train models on a regular basis. So it becomes more of a viable category for service providers like us basically.

I think the heterogeneity of the silicon is definitely a trend that plays in our favor there. The more heterogeneous, the more you need someone else to make sense of everything for you and tie it all together and also tie it all with the non-GPU aspects and the rest of the infrastructure and the applications and the users and the developers like basically everything we do for living.

There's only -- when you think of who is actually -- who actually has heterogeneous environments today, that is still a very small number of companies, Google, barely just started selling their TPUs to the outside. So I think it's still a small number of companies that are there, but we see a growing opportunity there.

Interestingly, last year, when we reported earnings, we said we're mostly interested in inference workloads and training is not really a market for us yet. Now we actually see training becoming a market. We started landing customers that are actually hyperscalers that have a whole host of homegrown technologies and that are using us specifically in their super intelligence labs to help monitor their workloads, accelerate the training runs, monitor the GPUs also. So we see that as a point of validation that there's going to be a great market for us.

That's amazing. I think there's a whole new dimension if you can move from inferencing into the training side. And I caught the reference in the prepared remarks of how you landed a couple of those very large labs. So congrats on everything.

And our next question will be coming from the line of Sanjit Singh of Morgan Stanley.

I want to start it off with David. And this guide to start the year is probably the best we've seen in several years, David, and you laid out the underlying assumptions quite well. I just wanted to do a sanity gut check just on the sort of overall backdrop -- macro backdrop. We do have some geopolitical tensions and those types of things when we think about. Your Middle East business and any impact from like in your e-commerce or retail business where there may be some consumer discretionary impacts. I just want to get like how you're thinking about those parts of the business? And then I had a follow-up for Oli.

Yes. We had a very strong quarter across the board. We had multi-industry, multi-geography type of quarter and SMB was very strong. And that -- the source of our guidance and our raises are at the core, that type of performance. We haven't seen any particular effect in the consumer businesses or e-commerce businesses yet. We basically have a continuation of trends in those businesses, travels and things like that, that are very similar to the other industries. So we haven't seen it yet. We obviously watch it and look at analytics, but we haven't seen it. In terms of our overall guidance, the trends that we have in organic, we discount across the board, and I think we mentioned our particular treatment of our largest customer.

That's very clear. And then, Olivier, for you, I think when we talk to investors about the debate in this category longer term, is just what does this -- what does the category look like when agents are doing the triaging, investigating versus human engineers and human SREs. And so what is your sort of vision of that -- how that evolves for Datadog, both from a product standpoint and an experience standpoint from a UI perspective. But also like is there going to be new modalities in terms of pricing when agents are consuming the Datadog platform to a higher degree than engineers do today?

Yes. Look, I think one thing I'd say is it's hard to tell where we're going to be in 4, 5 years. If you had told me 2 years ago that most engineers would go back to coding in the console, I wouldn't have believed you. And yet, that's one of the winning modalities today.

Look, as far as we're concerned, we don't care whether most of the usage is humans, most of the usage is agents. Our business model lends itself to it pretty well, like we're usage-based, and it doesn't really matter where the usage is coming from, from that perspective.

The way we see trends up right now is we see both a stratospheric increase of agent usage. So we have a ton of usage on our MCP server. We see customers trying to automate a lot with their own agents, using our agents, using a combination of those. But we also see an increase of usage of the web interfaces by humans. So right now, the two work hand-in-hand, and we keep developing and pushing on both fronts.

Next question is coming from the line of Raimo Lenschow of Barclays.

Congrats from me as well. One for Olivier and one for David. Olivier, if I listen to the -- to you in your prepared remarks, there's a lot of like consolidation that people try to do open source tooling and then realize they kind of needed to come to you and come back. On the other hand, in the industry, we still have a lot of like noise around that level. How do you see it in real life? To me, it seems a little bit like observability is just very hot and then there's different categories where you use certain vendors and some open source tooling. Can you speak what you see in real life there?

I mean in real life, most companies have open source in some capacity somewhere. But when it comes to having a platform that unifies everything, takes care of everything, does more of the problem solving for you, that's typically what customers use us. And the motion we see pretty much everywhere is customers have 4, 6, 7, 15, 25 different things and different pockets in the organizations and different business units, and it's a huge mess, and they come to us so they can unify all that.

They get better results because all of the data is in one place. The workflows can be automated from end-to-end. You can get end-to-end visibility, you don't have blind spots. And also they save money because they don't have all these pockets of inefficiency everywhere. So it's a win for everyone.

The thing that's also interesting, in particular this quarter is that we also landed some large parts of hyperscalers. And hyperscalers typically have a culture of building everything themselves, and they certainly have the balance sheet and the human capital to support some of that build-out. Like if there was ever a set of companies for whom it makes sense to do it themselves, that would be those companies. And yet, we see that they have the same issues. When it comes to going as fast as they can and being as efficient as they can with their resources, like they come to us to replace some of the things that we were using before.

Two things, two metrics to look at that to make the points, Oli, you're making. If you look at our platform adoption and you see both the growth of the different categories and the extension of the categories out to lots of products that shows you that the consolidation on the Datadog platform has continued and is a very strong trend. And part of that is the movement of solutions, as Oli mentioned, that are both open source, but also the competitive point solutions onto the platform. That's been a significant driver of the revenue growth for some time now, and that continued certainly in Q1.

Okay. Perfect. And then, David, for you, the last year, and -- we did a lot of investments around go-to-market, especially on sales capacity. If you think about now the non-AI category doing better, how much of that is like people doing like the cloud migrations again, so that's like an industry trend? And how much of that is like you guys actually being broader positioned?

Yes. Well it's a number of things, including one is the expansion of the platform, the consolidation the successful ramping of sales capacity, which is -- while not jeopardizing productivity, which has resulted in increasing ARR and a good environment as well. And I think that's what we said last time. There are a number of factors. And certainly, what we're proving out here is the investments we've made in go-to-market and are continuing are paying off and were the right decision. Oli, anything there?

Yes. And look, we -- at the end of the day, there's clearly some market tailwinds with the adoption of AI. And -- but also, we are outperforming all of our competitors at scale, and we're taking share. And that relates to the structure of our platform, the way we expand with new products, the way these products are maturing and starting to win in their respective categories and the way we've successfully grown the sales capacity.

Certainly, the AI investment trend has helped, but what we're trying to do is separate that. So -- and AI investment is probably helping the overall as well. But when you really take that out, you still -- you see a very pronounced acceleration here, and that has to do with the factors that I mentioned and Oli talked about.

Our next question is coming from the line of Gabriela Borges of Goldman Sachs.

Olivier, I find your comments on training versus inference so interesting. Maybe just crystallize for us, why do you think the training opportunity is happening now or inflecting now? And then either for yourself or David, how do we think about the attach rate on training versus inference of observability? If there is a way to benchmark observability spend as a percentage of inference spend, does that number change given the new data that you're seeing on the training side as well?

So on the training side, training was very new a couple of years ago. It was something that was only done by very few companies, and it was, in a way, very artisanal. Like, it was not a production workload. It was something that researchers were building and that was very one-off and homegrown in ways. And now it's turning into production. It's turning into something that many more companies are doing. It's scaling by orders of magnitude.

And it's becoming something that has to be on all the time, reliable and every minute you lose is -- or rather every failure you have in your training runs is a week you give away to the competition. And so as a result, it becomes way more interesting as a market for us. And we see some signs of that. Again, we didn't have a lot of it. We didn't see a lot of it last year. Now all of a sudden, we're starting to see quite a bit of activity there and demand, and we have success landing with large customers with those products.

Yes. I think going back to the metrics that Oli talked about in terms of attach, we said that 6,500 customers are using our integrations and that's 20% of the customers and 80% of the ARR. So there is attach. I think it's earlier days for the training. That looks like it will be a contributor. But I think we -- that's early, and I would sort of look at the larger attachment at this point as the evidence of inference, but also some training.

Our next question is coming from the line of Karl Keirstead of UBS.

Okay. Great. I wanted to start, Olivier and David and Yuka in congratulating all of you and the team on reaching that $1 billion milestone, well done. David, maybe the question is for you and to hone in specifically on the 2Q guide. Even if you put up a modest beat on that guide, it's going to be by order of magnitude, the largest sequential dollar add, I think, in the company's history. And I just wanted to unpack what's giving you that confidence. And in particular, is there anything interesting to call out, David, in terms of the ramp of a couple of the larger research labs, one of which renewed with you guys in the fourth quarter, another one just landed. I presume they're ramping nicely in 2Q, but would love any color.

Yes. Let me unpack this in a couple of ways. As you know, we're a recurring revenue model. So the biggest indication of in the near term of the next quarter is the ARR growth in the previous quarter, and we said we had a record. So essentially, at the bedrock of this is sort of the run forward of ARR that we've already signed. The ARR add was very broad-based and was not very concentrated. So -- whereas we pointed out some very significant adds I would say that the first quarter and that ARR add was really diversified and from lots of different places.

So the -- and I think Oli will come in here. But the confidence that we have is, you're right, we essentially take what we already have. We discount the growth trends that we've seen. And that produces what you exactly said, which is whatever your assumptions are on beat, a very impressive sequential really due to what happened in Q1 and the rate of business accumulation by Datadog. Oli, do you want to add?

Yes. I mean, I basically want to double down on what David just said. The adds were broad-based. I mean, look -- and when you look at why do we have a great Q1, we also landed great customers in Q4. We had talked about it a quarter ago. But even if you take out the customer we land in Q4 that added the most revenue in Q1, we still had a record quarter in terms of ARR adds. So this is really broad-based. And we landed a few more customers in Q1 that don't contribute any revenue yet, but we expect to be big contributors in the future. So when you put all that together, we feel very confident about Q2, hence, the numbers you've seen.

And our next question will be coming from the line of Fatima Boolani of Citi.

Oli, I wanted to double back on a question that was asked earlier with respect to telemetry volumes, essentially going parabolic and you are accessing brand-new demand vectors in the foray into training and monitoring and observing training model environments inside some of the world's largest frontier labs. And so I wanted to ask you about the structural changes to the capital intensity of the business. I mean your CapEx levels are still pretty respectable and pretty muted. So I wanted to get a better understanding of what sort of extrinsic or intrinsic engineering efforts you're undertaking to keep a very efficient CapEx envelope in spite of the fact that it seems like that would increase because of the torrent of telemetry you're seeing on the platform.

And then as a related matter, we've seen a rise of sovereign data and data residency requirements kind of ramp as AI models move into the territory of national security and things like that. So just wondering if you can kind of talk to some of the engineering horsepower internally that you're leveraging to be able to keep a really tight command on capital intensity and frankly, your gross margins?

Yes. I mean, look, so the investments we're making right now, we run most of our workloads on cloud, meaning you'll see all of that in OpEx, not in CapEx. So we have low CapEx. If it changes, we'll tell you, like if for some reason, we decide to make different kinds of investments and some of it more upfront, some of it more CapEx, we'll tell you, but that's not the case today.

We are definitely ramping up our investments, in particular, in R&D and in the scale of the models we train ourselves and things like that. Right now, there's no -- nothing that you can actually see in the numbers that move any needle, but if that changes, also we'll tell you. We don't expect any change to our model. So that's on the CapEx side. We're very different businesses in that way from the AI lab.

On the subject of data residency and sovereignty of AI and things like that. We definitely see more push for that, more demand for that in the customer base. And for us, that means investments in 2 areas. One is in deploying into more geographies and having more certifications to sell to the public sector and to the highest levels of the public sector. So we mentioned today data center in the U.K., for example, and our FedRAMP High certification. We are not stopping there. In terms of the certification, we're going after with the federal government. So that's an area of investment.

Another area of investment is our bring your own cloud products and -- where we can actually run on our customers' infrastructure. And so we announced that. We released some products there, and we have heavy investment in that area so we can support customers that want to operate in a slightly separate way from the rest of our customer base.

And our next question is coming from the line of Kirk Materne of Evercore.

Congrats on a nice start. Oli, I was wondering if you could just give some thoughts on the idea of sort of security for agents. I think one of the big issues in terms of getting agents into production is sort of the security aspect of that. And how do you see Datadog plugging into that opportunity?

And then just a quick one for David. Congrats on the FedRAMP reaching that milestone. Are your partner relationships in place to take advantage of this? I realize it will be a long-term opportunity, but just kind of curious how well established you are down there to start seeing some maybe bookings in that area.

Yes. So on the security of agents, we interface with that in 2 ways. So first, there's the agents we build ourselves because we are building a lot of automation inside of our product for our customers and agents that automatically identify but also resolve issues without you having to do anything. And there, a lot of it has to do with understanding what permissions to apply, what kind of guardrails to apply, what kind of -- how to interface with the humans and how to make that trustworthy and visible in the right way.

And so that's pretty much the whole product surface is to [indiscernible] data. The automation itself actually kind of works already. So you should expect to hear more about that at our conference. This is definitely one big area of investment for us.

On the security aspects of our engagements, look, we -- our belief in security is that you need to integrate -- you can't just have point solutions that look at one lever of the whole security posture. You need to look at everything altogether. And that's one of the areas that we are also covering with our security efforts. So that's part of the whole platform, actually.

On the FedRAMP, we've been working on both the different certifications. But at the same time, we've been investing in the go-to-market function, both in terms of reps and channel partners for a number of years. Certainly, there's more investment to be done, but we invested ahead of the certifications because in this sector, building pipeline, et cetera, takes time. And certainly, the channel partner relationships are a very important part of this, and we have been investing, but also have more investment to do.

Our next question is coming from the line of Patrick Colville of Scotiabank.

Echoing the congrats of my peers. I guess, Olivier and David, you guys are very deliberate in your messaging on the prepared remarks. And I just -- I guess I want to double check the kind of wording of one of the comments. I think, David, you said a higher degree of conservatism to the largest customers. I guess, did I hear that right? And then does the higher degree of conservatism reference versus the other customer cohorts? Or does it reference versus your guidance philosophy in prior quarters vis-a-vis this customer?

It's both. It's the same guidance we used, and we're being very explicit. For all the business, except for the largest customers, we've always taken the drivers and discounted them. We -- for this particular customer, we took a higher degree of conservatism than the other part of the customer base and discounted it more. And we were, I think, in the remarks, and you interpret it correct, very explicit and you're correct.

I wouldn't give that much weight to the very specific way. We're deliberate, but not all that deliberate. And similarly, both David and I have a raspy voice today, but there's no other meaning.

But I will remind everybody that we did not change. So it's the question also, I think you asked, is did we change or is this a different methodology of both the overall and the large customer than the guidance the last quarter or the previous? The answer is no. It's the same methodology and -- that we've had. So no change, but that has been what we've always been doing.

Okay. And Olivier, can I ask about your comments about the hyperscalers because I thought that was particularly interesting. And the reason why is I don't think you called them out previously before, and they are so prevalent in the modern tech stack. To your point, they could do this themselves. So I guess how are they using Datadog? Is it for more kind of traditional observability? Or is it for these newer areas like GPU monitoring that Datadog has performed so well of late?

Well, it's both actually. When you look in general at the large AI customers, they use Datadog the way other companies are largely with a fairly broad set of our products to cover the full surface of observability. What's new is we now have a product for GPU monitoring. It's a very new product. And we see the hyperscalers that are coming to us for training workloads in particular, being very interested in that.

So again, it's too early in the product life cycle and the customer life cycle for these specific customers to call definitive victory there, but we see that as a very encouraging sign of where the market might go in the future because we think this might be a bellwether of what the next 10, 100, 500 companies that are going to start training workloads are going to want to do. We have some signs that go beyond the customers we signed this quarter that point that way too.

And our next question is coming from the line of Peter Weed of Bernstein Research.

I'll echo others on the momentum. Great to see. One of, I think, the great successes you talked about was landing a couple of the AI labs for the hyperscalers. Although I think on the other hand, you've talked in the past around hyperscalers are typically building observability in-house. What is it really about the AI workloads that are making it more attractive for them to use Datadog? And what might give you confidence that Datadog might be more persistent with them in these types of workloads and that's kind of a signal for maybe how other customers might use Datadog around AI differentiated from things that they might be able to bring in-house other places?

The same reason all of our customers use us. It's high stakes, high complexity and not core, they have to be what's differentiated. They can't afford to be late, and it's a really hard job to do that. So that's what we build our whole business on. And it's also very true for -- at the highest level for the largest companies.

Yes. No, I was just going to say -- but I guess the point is you've emphasized that those largest customers have been able to go in-house on some other things. Is there something unique about AI that prevents them from doing that here?

Well, I think the urgency of their development efforts focuses the mind. That's what I would put it. I would say it forced you to figure out what's core and what's not core and what's the -- who you want to get to the -- what you need to do to maximize your chances of success. And again, it is the same thinking all of our customers have all the time.

I think the equation for hyperscalers has often been fairly different because they have, let's call it, unlimited access to staffing and they could sort of set their own time horizons for the developments they wanted to make. I think the situation is a little bit different with the AI race, maybe.

And our next question is coming from the line of Gregg Moskowitz of Mizuho.

And I'll add my congratulations on a terrific quarter. Just one for me. Oli, I know it's not GA yet, but curious if you have any early feedback on your new cloud prem offering. As you noted earlier, providing the ability for Datadog to run on customer infrastructure. Could this be another -- yet another, I should say, incremental growth opportunity for Datadog? What are your expectations for this?

Well, definitely, we think -- I think there was a question earlier on data residency and living in customers' environments. We definitely see a great opportunity there. There is a chance that a good portion of the market leans this way in the future. Today, it's not the largest part of the market, but we definitely see a potential for that. So we're investing heavily in that sort of our product.

We are starting to see some interesting customer traction there. So we think this can be another growth lever definitely. We also think that it can help us getting into some extremely large-scale workload where customers would not have considered a SaaS offering before where we can be in the running. So that's very exciting.

All right. And I think that was our last question. So I want to thank you all for attending the call. And I'll remind you that we have our conference in just a bit more than a month, and I hope to see many of you there. So thank you all.

This concludes today's program. You may all disconnect.

All right. Good morning. Day 2 morning session of the Morgan Stanley TMT Conference. We are super thrilled to have the Chief Financial Officer from Datadog, David Obstler. David, welcome back to the TMT conference. I think you guys have been here every single year that you've been public.

Yes. Thanks for having us back.

Awesome. Yes, we're going to -- lots of talk about. Business is doing well. We've going to talk about, of course, AI and all sorts of other topics. But before I get there for important disclosures, please see the Morgan Stanley research disclosure website at www.morganstanley.com/researchdisclosures.

So with that, let's kick off the conversation. If we look at this year, you're coming off another strong year business accelerated from a growth perspective to 28% to $3.4 billion in revenue, delivering operating margins of 22% and you're now serving 32,000-plus customers. We're at an interesting time in the market, and particularly with software, and then there's a kind of a return to first principles thinking and assessing how software providers create value. So with that context in terms of getting back to basics, David, what are the core problems that Datadog helps customers solve today? And what problems will Datadog help solve going forward?

Yes. Good question. So Datadog helps companies migrate their mission-critical applications, usually customer-facing to the cloud or manage them in the cloud. Our legacy has been -- and observability, which has really been around infrastructure, APM, logs and digital experience. And as we've expanded the product set over the years, we tend to handle more and more of the problems of clients always in a single pane of glass, heavily integrated, our customers in DevOps and SRE can come in, turn it on and see what the environment. And we've been moving the platform from observability to areas like security, towards the front end, towards digital applications and digital experience, now product analytics, to service management and workflows and injecting AI in the platform, both the platform itself and the monitoring, all to try to move from observability to recommendations and an action, which has been the core value proposition of Datadog from day 1.

Awesome. Let's talk about some of the business trends that they mentioned. One of the things that stands out for me is just the core business sort of reaccelerating. It's accelerated for 2 quarters in a row. You grew 18% in Q2, 20% in Q3, accelerated again in Q4. What are the factors driving the reacceleration? And how durable does this rejuvenation in growth feel to you in the context of the current demand environment?

Yes, there have been a number of factors. This is the business in companies that we're not calling AI natives. And first of all, we're very, very early in the migration of applications into the cloud and in the modernization of applications and infrastructure. And what we've seen is a good buying environment, meaning corporations down the SMB have turned back towards moving applications into the cloud. That is likely we said and we believe to be accelerated and complemented over time by the re-platforming that's going to be caused by increased complexity because of AI. So we have a good buying environment.

The second thing is that we've expanded our platform substantially over time. So we have a lot of new products, a lot of new products that are getting to and achieving scale. And so we have a broader value proposition to sell. I would say the third thing is that we've been winning market share and consolidating. So remember, the value proposition is single pane of glass, a platform. And we're finding that we've been able to consolidate that market. There's some really good evidence like the acceleration of the APM product. That means we're innovating, but also taking market share.

And I think the last thing in our hands is that we've expanded our go-to-market capabilities. We successfully expanded our quota capacity. We've been able to do that while maintaining productivity. That includes new geographies, governments. I think we're getting better and better about enterprise selling and the go to market motions. So that's something very much in our hands, and we've been investing for the last couple of years, and that has been paying dividends, helping to accelerate that business.

Yes. That point around consolidation is, I think, an important one because you have to have the product portfolio to do that, right? And there's a lot of players in this space, that probably only a handful that can actually do get to -- go to a customer and say, we can consolidate 10 or a dozen of these capabilities onto the Datadog platform?

I think we've showed a very interesting statistic on our Investor Day that despite the fact we've been at this for a while, only half of our customers are using all 3 pillars. And once a customer standardizes on Datadog, their spend accelerates. So we've done a lot. As you say, we have been consolidating and we do have that product set, but there's much more to go.

And just maybe stick on that topic I was going to ask this question a little bit later. But on the customers, roughly half of the business that's not on all 3 pillars, what do you think will be the unlock to get them to adopt more of the platform?

I think you have to think back to history. Datadog's first product was infrastructure. And there were APM log products, digital experience, database monitoring out there. So I think the 2 things, and we're seeing this over and over, that are unlocking, are, one is the frictionless adoption in the platform that exposes these products to clients and then time because there are installed bases, they're champions, et cetera. So the value that's being seen from having everything in a single pane of glass is quite substantial, but it takes some time to replace those other legacy products or legacy customer bases. So we're seeing that. I think we said over the years that in our larger deals, somewhere plus or minus half of those largest deals have consolidation in it.

We've been talking about the core business trends and the forces driving that growth. You guys are also doing incredibly well with the AI natives themselves. You have 70% of the top 20 AI native. You got 19 of them spending in excess of $1 million annually. And I think roughly about 650 AI-native customers overall. So what has the company been doing to penetrate this segment of the market? And how do you think this cohort performs going into fiscal year '26?

Yes, great question. This is an ideal customer base for us like cloud native. They don't have legacy infrastructure apps. They are modern cloud companies and Datadog, as you know, designed its product to be optimal and to be a product that is -- handles many of the needs of this type of customer. And these customers are growing very fast. So they're tending to land and adopt the Datadog products very quickly. They don't have something else they used before. So they're landing. So I think we have a natural product fit.

As you can say, we've been comprehensively winning. And what we're trying to do, it's similar to other fast-growth companies. We're trying to work with them in terms of both landing and expanding, expanding the product set, having very good account management and technical management with them, we've evolved over the years to have lots of different departments at Datadog that help a client understand. And what we're finding is they're deciding that it makes so much more sense given their huge investment in their own products to buy Datadog's. So I would say that we're winning. And we're also -- all the type of developer marketing and types of things we do hit very well with that constituency, and we're continuing with that.

On the last earnings call, you announced an 8-figure land with another leading model provider. There's a debate in the market that these types of customers want to build their own infrastructure and tooling. And so what were the reasons that this customer chose to go with Datadog after its observability needs?

There's a debate in the market, but the evidence is quite the opposite. In fact, when you look at how Datadog has grown and the market share it has taken. The predominant decision has been to use the Datadog platform rather than build it yourself. One, those companies have a lot to do. Two, when you look at the total cost of ownership, in terms of development and the platform and the cloud, it's efficient to use a Datadog and you get the best of breed.

So I think that's a good example of a customer, that early on, experimented and trying to do it themselves. But I think we said it's obviously one of the larger companies in the space and along with the other large companies in the space, they've decided to use Datadog. That is completely antithetical to, I would say, the concern out in the market. But when you really look at it and you look at Datadog's gross retention, which very upper 90s, and we showed this in the Investor Day, what you'll see is it's very much fringe cases that decide to do it themselves.

And the weight is really on the other side of using Datadog. We also have customers that do both at the same time, that maybe experiment with that and come back. So I think the weight of the market has been to buy the Datadog platform for lots of reasons, including efficiency, efficacy, return on investment costs, et cetera.

Awesome. Let's talk a little bit about, you guys had reported your Q4 results and guided for 2026 the other week. And so I just wanted to review some of the assumptions around there. You guided for 19% revenue growth at the midpoint, excluding your largest customer, you think the remainder of the business will grow in excess of 20%. Given that the core business grew 23% in Q4, what gives you the confidence that growth excluding your largest AI customer will prove durable at current levels?

Yes. Great question. So we really -- so there's 2 different questions. What do we guide to and what are we seeing. And I think we said and showed that the business had been accelerating. We made a comment when we did our earnings that we saw that follow into this year. And generally, the business is predictable in the shorter term when you look at the sort of the trends. So the reason why we see it is we see a very good end market, we see a great adoption of our products.

We see us landing more logos and larger logos. We can talk about all this. So all of those trends are compound on themselves and give us the confidence. Now -- then we put conservatism on it. We take those growth trends, and we discount them in order to provide the cushion in our guidance. But pretty much that type of accelerating performance translating into guidance is due to what we're seeing repeating the 3 or 4 factors that I mentioned upfront.

Yes. It's been a methodology that you've got in place for years when it comes to the guidance. So let's talk -- you brought up the point on new logo lands and the deal sizes getting better. I mean one of the things that I found interesting about the Analyst Day is that the size of the enterprise land deals really stepped up in 2025 versus prior years. Can you speak to the size of these enterprise land deals and the force that's driving the big increase in the overall land size?

Yes, definitely. I think we're -- our product suite is broader. That means we are landing and expanding faster and more products. So we have -- we're still land and expand. We're still get in there very long. We showed in the Investor Day, very long-growing cohorts. But what we're seeing is we're seeing -- and this has to do with consolidation, replacement, quite a number of lands that are larger and are more comprehensive and our service model in terms of how to deal with those customers, how to sell through channels, how to help the business owner and the CIO make that migration, has improved, which would result in more acceleration. So those are all the things that product side and execution side that have created that broader enterprise selling.

Yes. Very impressive to see. So when this year started, I had investors reach to me like Sanjit, congratulations, you don't cover seat-based models, right? You cover the part of software that's insulated from an AI risk and those types of things. I have to say in the last couple of weeks, everything is getting questioned, including the names that I cover as well as across software and across sectors. And so I want to spend a couple of minutes talking through the debate in terms of defensibility against potential AI disintermediation.

And there's a couple of different angles that I'd love to get your perspective on. One question I get is how does the value proposition of the Datadog platform change when agents are doing the investigating and triaging of incidents versus human DevOps or cyber liability engineers. Is there an observability solution that has a dashboard that is being used as interface, is relevant in an Agenetic world.

That's a big question. There are lots of things, but I want to say something on front. They're seat-based, I think, which is an important thing, but there's also the word infrastructure. So when you're infrastructure and you see that's related to seat-based, we're monetizing based on the agents or the containers or the servers, we're also monetizing based on how it's used. But what are we doing? We're monitoring infrastructure. And what we're finding, and we've seen this through time is as the technology evolves, you need, and I would argue, we would argue increased need to have visibility into the infrastructure. So that's one thing that's very important.

Another thing that's quite important is we deliver and connect in a variety of ways already. We don't care if you're coming in through a desktop, through a wireless, through open-source, through OTel. I think as you look at our Investor Day slides, you'll see that we're investing a lot of money in making sure we can both cover and come in through the information around agents. So the delivery. And then I would add that when you look at the value that's delivered, you have the access, but you also have the integrations and all of the data brought together in an increasingly complex world.

You have the organization of that and then we're calling it service management or closing the loop. You have what to do about that. And when it comes to foundational models or access to data, we're either integrating or -- and I think we demonstrated at our Investor Day, our own foundational models to make sure that we're investing so that if critical capabilities involve, what's the most cost-efficient and best foundational model using the vast data, we have it. So I think that there's lots of reasons seat-based is important. So congratulations you're covering, but there's lots of other things under the hood that you have to look at to when I think if you're evaluating defensibility and frankly, the increased value add in your products.

Yes, that's an important point. And so the other angle on sort of the risk to observability players, including Datadog is the potential for customers to combine open-source tooling and manage their metrics, traces and logs, combine that with agents from like the model providers to reach over the data and execute the incident response. What's the company's argument for why this line of thinking is off pace?

We -- I think that we would view -- when you think we had the same discussion about OTel. The key is not -- it's -- yes, having access to all the data and Datadog through its MCP server, through its LLM monitoring, through all of this, through its integrations is always going to have and is investing significantly in having access to all the data. And then it's always been the case, whether it's OTel or direct integrations, et cetera, that what -- the magic is what happens after that.

So I think -- and we talk -- we can talk about AI for Datadog and Datadog for AI because there's a whole another set of things about AI-ness of the platform. But in terms of access to the data, I think if you look at our investor presentation, we're doing it, then, okay, why come to Datadog, why continue? And we believe that platforms like Datadog has to be completely AI-native, integrating with agents, but using agents within Datadog, and that's what we're calling our SRE Bits, our Security Bits, et cetera. That means that we are investing and leading the way in the agents in the platform to provide value, do diagnostics and eventually self remediate. So yes, it's important. This is an important DNA. And I think that given how much we're investing in R&D, and looking forward, all credit to Ali and the R&D team, I think we're doing it with and to ourselves.

Let's talk a little bit about the AI for Datadog. So just to add up on the point that you made and particularly around the models that you guys are developing. How much of a competitive advantage are the AI models Datadog has built with a huge data sets that you have access to when thinking about competition versus other AI natives of the research. Maybe said another way, will Datadog release more capable agents, more quickly in this domain versus the research labs and other competition due to a data and AI model advantage.

We do believe that will be the case. We don't know for sure, but we believe that when you think about efficacy and cost that having models that are based on the huge amount of data we have about this problem and trained on these data sets will be part of the equation that delivers most value and at a good price. Because when you're talking about generalist models foundational models, they are essentially using a vast amount of data. We don't have as much data as they have, but we have data that's more on point.

And we're spending our R&D dollars training that. I guess this might be one of the reasons why all of those AI native companies are using Datadog because what they're finding is the platform and the AI nature of the platform is a better solution for observing these workloads, securing these workloads and creating action than their more foundational generalist models.

Can we talk a little bit about the momentum that you're seeing with Bits AI, SRE agents. What's been the customer feedback with regards to its accuracy, reliability and speed in terms of finding the root cause to incidents.

Yes. It's been great. Look at the analyst presentation for some truly impressive quotes. We have just put it in GA. We have 1,000-plus customers using it. We have ARR where it's being paid for. We have a pricing model on the site. So the initial reception has been very strong. And of course, like all of Datadog's products, we launch them, we get feedback, and we continue to improve that. So I think we're going to -- we're still early stages, but getting great feedback in terms of the fact that we're on the right track. Customers are finding value and we're getting the feedback to be able to handle more and more use cases.

And I'm saying we're going to do that. The other one is that -- the other Bits products around security and around development, I would say they're a little earlier in this cycle, but we plan to do the exact same thing, which is have co-development partners, figure out the uses, get them in GA and then learn. So it's part of a whole portfolio, as you mentioned, in the AI of the platform.

Let's return back to the other aspects of the AI debate when it comes to this category and Datadog. So let's just go to be sometimes expressing the view that observability is just data ingestion into a time series database, we use some event streaming, and it's a dashboard. And so I think the implication that they're getting at here is that observability can be replicated with the help of AI and coding agents. I've heard you guys in the past mentioned that 50% of your engineers work on the core platform itself. And so I'd love to -- if you could expand upon the scale and sophistication of the platform as well as the services that the core platform provides. My guess is that you guys have spent over $1 billion building the core platform. So I'd love to get your perspective on why Datadog will be hard to replicate.

Well, a lot more than that because we're spending over $1 billion a year, and we said 50%. And I think there are a number of things. One, scale, how to handle the data. When you think about -- is it -- when you think about ingesting infrastructure data and creating metrics, No, our product is so much broader when you consider that it has that aspect, but then it has front-to-back application monitoring, customer behavior, databases, data observability pipeline, logging, service management, all sorts of things, all correlated. So I think one is all of those things that are orchestrated and all the integrations.

Two is it's been a huge competitive advantage in that with the platform and how we've designed it, we can build this additional functionality. That's at the core of our speed, much faster, much more cheaply and in a more orchestrated way than competitors. And if agents are important in doing that, they'll be in our platform. So I think the platform itself has been -- it's kind of a 2-way thing, a circle. It's enabled us to have all this functionality. And it's also enabled us to integrate new functionality in it, which I think will position us very, very well as the world gets even more complex and agents start to.

I mean there's all sorts of evidence that we have now in terms of the number of calls out to MCP servers accelerating dramatically between the third and fourth quarter. The spans that are sent from LLMs 10x over the last 6 months. So all of this is proving true that the platform is a contributory factor in having us continue to be even maybe more so in the future at the center of all of this.

You guys built multiple billion-dollar-plus business, like your Logs business. I think some of those businesses were built with relatively few engineers because they're building on top of a platform.

There's no question. The return on investment and the ability to get through, I think we said infrastructure had gotten to $1.6 billion, logs over $1 billion, APM and digital experience over $1 billion. And there's lots of other examples we've been given. And I think a lot of it has to do with the basic fact that Datadog started as a core data and infrastructure company which has created this competitive advantage and the ability to develop very quickly.

Awesome. So let's talk about some of the opportunities to unlock growth further across the business. Let's get an update on the security business. Security seemed like it shifted to a higher gear in 2025. And one of the -- going back to the Investor Day, one of the data points that you guys put out there was 70% of your $1 million dollar customers are using one or more security products. But in terms of the ARR, it's still relatively modest. So what are the initiatives to improve the security adoption with your larger spending customers?

Great question. We launched security initially to work with our platform in DevSecOps and more progressive customers, more cloud-native customers and I would say, more limited functionality than we've arrived to. And so the change, the thing that's inflected here is that both with the maturity, and I'll talk about Cloud SIEM as an example, the maturity of the product, the architecture of the product to use are very entrenched and excellent enterprise logs business and expand the uses of logs outside of observability logs.

The redesign of the front end and the addition of channels and services -- service partners on top has allowed us to begin to attach Cloud SIEM in particularly, but also more broadly onto some of our sophisticated large enterprise customers. And so that's what we started to do last year, and that is starting to take up the ARR per customer and move it into, I would say, larger use cases, more traditional use cases, et cetera.

So it's a combination of product maturity, particularly in Cloud SIEM, using our very good installed base in logs and the go-to-market. We also have a good market environment with some things that have changed in some of the other competitors, where we've been able to, like we did in Cloud Logs, focused on the Cloud SIEM work functionality and start to penetrate some very significant customers.

Guys like me have been asking for several years when it comes to the security strategy? Like are you guys need to have -- are you guys going to -- Datadog going to build a specialist security sales force. It sounds like you guys finally made that decision to pull the trigger on that. So the question is like why now in terms of deploying that specialist security sales team? And what do you hope the impact will be in sort of year one of the deployment?

First of all, I think that we acknowledge, security is a different go-to-market motion. It's more channel led, it's more centralized. And so I think the reason that we didn't do it upfront is in order to win the hearts and minds of channel partners and to have the right motion. You have to have a competitive product, equal or better in functionality. It probably doesn't make much sense to have all that distribution if you're not going to be at that level.

So I think it was a step function in that we had to get the product, in the right shape and Cloud SIEM got there and et cetera. So I think it couldn't have happened. We're early on in it. I think the progress we've made already has been mainly due to our existing go-to-market motions and our cross-selling into enterprise logs customers. And we're just getting those -- both the channel and the specialist salespeople potted in. We don't know the answer. But I think it's going to help accelerate by having quota-bearing either channel partners or salespeople who do nothing else. So it's early, and we'll report on it, but we're optimistic that it will help accelerate.

I guess the signal here is that you guys are confident now, more confident in the actual product capabilities. And because you're more confident now you feel more motivated to engage the ecosystem in the channel.

To win the channel partners and to get them recommending. You have to have a number of things going on, including the product. So I think what we said is we're ready for that, and that's why we did it last year and are going to invest behind it.

We talked a little bit earlier about AI and how it's embedded within the Datadog platform. Let's go the opposite way Datadog in terms of helping customers with their AI initiatives. LLM observability is a product I think you guys had out for a little over a year, maybe a little bit longer than that. Do you feel that this product has found product market fit? And what are some of the recent patterns on usage? .

Yes. Definitely. So yes. But this is essentially dependent upon our customers having LLMs that are in their production environments. And so we are setting it up and starting to see inflection in that. I think we have the right product. And the amount of use of integrations and the span sent to us have accelerated substantially. I think I wrote down a couple of metrics in that. We have over 1,000 users and our spans sent to us in the last 6 months have expanded by 10x.

I think that's evidence of our clients starting to integrate LLMs into their own production applications. And we have the right product and fit. So I think we're early on, but we're getting a lot of good usage and the acceleration rate of the use of that -- those integrations and the data they're sending us is quite pronounced recently.

Yes, it's great insight. Another area, one that I'm particularly interested in, because we sort of live in a GPU economy. You've talked about a product for GPU monitoring that's in preview now. So why decision to enter this category? What types of customers do you think will be most interested in this offering? And what do you think pricing will look like?

Yes, we've had it. We've had the ability, but I think we're -- we understand that the GPU is larger than the CPU and required. So I think this will be as our clients, broadly speaking, not just the AI natives, but as enterprise customers or cloud native start to create their own models and integrate it into their own production environments, they'll be using more GPUs. And just like all the other product areas, we want to be there to be able to monitor it.

So right now, we have some use. What we want to do is increase the functionality, optimize the pricing, the same way we do always. And so to be there for our broad client base as they use more GPUs. Don't forget, most of the GPUs have been fairly centralized right now. And the infrastructure providers and the model. But we don't think that will be the state of play of the market down the road.

Expect some broadening. That makes a lot of sense. I wanted to wrap up the conversation just on some of the consolidation that's going on in the market. So you've seen some large security vendors acquire into observability. You've seen data platform providers acquire into observability. So the questions here is, can -- the assets have been acquired can they pull off those consolidation deals that you and maybe a handful of others can do in the market? Or are they -- what's your assessment of the breadth of that data capability.

Well, this is not brand new. This has been going on for a while. Since we've been public, we discussed it. There have been a number of companies in security or in automation, in IT automation they've tried to do this. And they basically have not succeeded. So I think it depends on what you're acquiring. So far, most of what has been required are point solutions that don't offer the observability platform. So I think, number one, in order to compete against the company that has the breadth of product and the single pane of glass.

The market has not been kind to point solutions. So they have to do that. They don't have that. And then just like our challenge in go-to-market and security, they would have a challenge in sort of the bottoms-up selling into DevOps. So I think it's a tall order. And most of the companies that have been acquired, our point solutions that may have use cases, but really haven't been competitive in the broad obseverability platform.

Maybe one last I want you to comment on, as with respect to the competitive element. What do you think will make Datadog the winner versus the security players who also have their agents deployed in the customer environment just like Datadog does.

Yes. I think it's essentially having the data and the integrations and the user interface that are optimized for those use cases that are -- we've learned, we've just talked about are different than security. And so I think it's the same thing on the other side, whether it be agents or whether it be developers or whatever the combination, these raw materials have to be integrated with the data. And so I think it's going to be a lift for those others to do it. It's certainly -- they've been there trying to do it for 20 years. It hasn't happened. Who knows about the future, but it is a difficult lift.

Well, we'll leave it there. David, you -- asked a lot of product questions, less financial questions. So kudos on that. Thank you for giving us the update on Datadog. Really appreciate it.

Thanks for having us. Appreciate it. Thank you.

Good afternoon everyone. My name is Yuka Broderick, and I lead Investor Relations here at Datadog. Welcome to all of you here in the theater and everyone joining us online for our 2026 Investor Day.

Before we dive in, just a few reminders. During this presentation, we will make forward-looking statements, including statements related to our strategy, product development, market opportunity and financial goals. These statements reflect our views today and are subject to a variety of risks and uncertainties that could cause actual results to differ materially. For a discussion of material risks, please refer to our Form 10-Q for the quarter ended September 30, 2025, and other filings that we may make with the SEC, including our Form 10-K for fiscal 2025.

We will also discuss non-GAAP financial measures, which are reconciled to their most directly comparable GAAP financial measures in the appendix to this presentation, which is available at investors.datadoghq.com.

All right. Let me briefly run through our agenda for today. In the first half, we will focus on our strategy, platform and product innovation. You will hear from our Co-Founder, CEO, Olivier Pomel; and CTO, Alexis Le-Quoc, followed by product leaders, Yrieix Garnier, Tim Knutson, Michael Whetten; and our Chief Product Officer, Yanbing Li. We will follow that with the Q&A with that group. In the second half, we will discuss our go-to-market, how we deliver value to customers and our financial performance. Our presenters will be CRO, Sean Walters; COO, Adam Blitzer; and CFO, David Obstler. That group will be joined by Olivier.

With that, let me turn it over to Olivier to kick things off.

Thank you, Yuka. All right. Hi, everyone. My name is Olivier Pomel. I'm the co-founder and CEO here at Datadog and welcome to the Investor Day. So you're going to hear from several of our leaders today, and I know they are quite excited to present. And so as we discuss what we're doing here at Datadog, our goal is to show you why even after 16 years of building the company, we believe that we're still only just getting started. We're barely scratching the surface of the opportunity. So I'll kick us off with a quick recap of the problem we solve today, how our platform has expanded over time, what we're up to with AI, and I'll give you a sense of the broad direction we're taking over the long term.

All right. All right. So let's start with cloud migration and digital transformation. So you have here the usual beautiful chart [indiscernible], which shows a sustained rate of migration over the past few years that is expected to continue for the foreseeable future. And it is worth noting that Gartner expects spend on public cloud to exceed $1 trillion by 2027. And even with that, it would still add up to only 16% of global tech spend.

So why is this happening? Because as a company, any kind of company, you absolutely have to, you have to interact with your customers online. You have to differentiate from the competition through innovation. You have to run into the cloud to get agility, short time to value and efficiency. And to be honest, you also have to lean into the best tech so you can hire and retain the great engineers. And in the end, this modernization leads to better business outcomes. All of this was true over the past decade, and we expect it to be even more pronounced in the age of AI. As you cannot adopt AI, if you are not digital and in the cloud.

So our customers want to build their applications in the cloud as fast as possible using the latest technology. And this is a slide I've been using for a while now, and that is still true. And it illustrates the explosion of complexity. We've all witnessed [indiscernible] innovation over the past 20 years. So I won't spend too much time on it because we've gone [indiscernible] before. But you can note that we added a few things in there such as on the chart on the top right, which is the scale and compute units. You see that now we've added the GPU fleets with many [indiscernible] being deployed all the time. And the way you read those charts is you multiply each of them. And so what you end up with is truly an explosion of complexity.

Now AI is a large, rapidly growing and exciting [indiscernible] spending. And I think we all noticed. But to put things in perspective, we have a few Gartner numbers. So the market opportunity really is very large. And for us, it is the next big thing that's going to bring both better business value to customers and additional leverage internally in our business. But it's also going to compound the complexity our customers face to take advantage of it. And so we added a couple more charts in there on the right side.

You'll see one on the top right that illustrates the number and also read the scale of models that are available in [indiscernible] phase. And on the bottom right, we have a chart that shows the increase of developer productivity. And we've all seen the shocking ascent of coning agents over the past few months. But really, this is only the continuation of a much broader trend that has been going on for decades.

When we started the -- way back when you had to write all the code yourself in low-level languages. And then you could use higher-level languages. So you became 10x as productive. Then the Internet came along and you could learn about everything so much faster, so you get another order of magnitude and productivity. Then we saw the assent of the open source software in the cloud. So you could use [indiscernible] components, so a full end-to-end services and then you get another order of magnitude of productivity. And today, with the rise of coding agents, I think we're starting at the next 1 or 2 other productivity right now.

So this is the prime result. To put it simply, Datadog exists to solve this enormous problem of complexity for our customers. We connect to all of their software components. We scale with all the infrastructure, compute units they deploy. We also scale with the services they create and deliver. We understand the way infrastructure and applications are changing, and we connect separate teams to each other across different functions.

Now what's interesting is that none of this problem are going away in the age of AI. In fact, the complexity is even greater because AI, all those things to be built faster. There's a lot more of everything. And the stakes are much higher as agents start to act on their own [indiscernible] humans. Our response to win this race against complexity is to invest, to invest in innovation and to invest on behalf of our customers.

Over the years, we've invested about 30% of revenues into R&D. In 2025, we invested over $1 billion in R&D and ending the year with about 4,000 engineers. We believe that we are investing several times more than our largest peers. We are also adopting AI in R&D and [indiscernible] products, which gives us unparalleled leverage and velocity in our space. That investment has led to an expansion of our platform over time and our successful entry into new categories to solve more problems for our customers. We delivered [indiscernible] on the unified platform. that break down silos among what used to be disconnected teams and data sets.

Now all those categories you saw on the previous slide, are part of the critical user flows that make up the activities of our customers' businesses. And what we see here on this slide is one way you can model on the customer side, the continuum of problem areas that take customers from tech innovation all the way to realizing business value. So you have writing code all the way to the left, and understanding where the value is and running the business all the way on the right.

So we started right in the middle in observability. Of course, we didn't have all the functionality listed here from day one. We added a lot of it over time as we kept going deeper and kept covering all the new technologies our customers were adopting. And this includes new innovations to help our customers observe their AI stack as you can see on the highlights there. More recently, we've added capabilities around the data layer bringing observability across the entire data life cycle from injection to transformation and all the way to downstream using reports, applications and AI models. And the market for data observability is picking up quite meaningfully as data is critical to developing and adopting AI. Data quality monitoring, in particular, resonates with customers.

Looking to the right, we have been building a great business in digital experience. We are now expanding further into user and report analytics, where we are seeing rapid customer adoption. And we are improving value for customers with our [ Synthetics and RUM ] products that can scale very cost-effectively to extremely large consumer user bases. And what we found when getting all of this into the head of customers, is that by broadening our scope and removing painful integration points, we see the value of our platform go up dramatically in surrounding areas.

Now looking left on the developer side. The developer landscape is evolving very quickly with [indiscernible] tools and rapid deployments. And so the market for tools that have developers build, deploy, debug and iterate, is expanding very meaningfully. So we've been bringing more value to developers with capabilities such as feature flags, the Datadog MCP server and the [indiscernible] agent. And we have much more to do in the software delivery space. You will hear later about this topic from Michael.

Security is a concern that spans from end-to-end across a spectrum of development, production and user interaction. And our suite of product delivers against that need. This is includes helping find venerabilities in development, identifying and eliminating threats in production, and securing sensitive data around live user activity. And we are building AI capabilities to move faster and [indiscernible] security problems. You will hear more from Tim on this area.

We've been building our cloud service management products as well. What this involves is going beyond helping our customers understand and secure their systems and into helping them coordinate people and teams to manage, communicate, organize, take action and more and more automate response. So we are building on our momentum in this space, including the successful launches in 2025 of our own core product and [indiscernible] SRE agent. And Yanbing will share more about that.

So if we look at the platform altogether, we are delivering an end-to-end suite of capabilities that help our customers build faster, deploy confidently, fix problem rapidly and deliver better business outcomes. And we are breaking down silos across operations and DevOps teams, data engineers, product designers, developers, security teams, incident responders, FinOps teams and business users. We have much more to do in each of these areas, but we've made meaningful progress over the years and are seeing broad customer adoption, which shows that we are delivering value.

As proof of the value, we're delivering across multiple categories, as I discussed on the earnings call a couple of days ago, we now have $1.6 billion of ARR in infrastructure monitoring another $1 billion of ARR each in both log management and the end-to-end APM and [indiscernible] suite. The fact that we have real balance across the 3 pillars of observability as well as meaningful scale in each one shows that Datadog is unique within the industry in establishing true platform value for customers. Even though you've heard a lot of competitors say they have 3 pillars capabilities, their business typically remains driven by just one of these pillars, which dramatically reduces the value, they're able to deliver against customers explosion of complexity.

All right. So I want to talk a little bit now about our AI build-out and break it down into 2 buckets. First, we're building AI for Datadog. We're embedding AI across Datado who -- every type of engineer can move faster. Our AI agents now surface context, identify problems and recommend fixes faster than any human can. They can be proactive and get in front of issues and users can interact with them in plain English.

The second category is to cover the AI applications or agents our customers are building or running themselves. And we call [indiscernible] Datadog for AI. If you're putting LLMs or agents into production applications, those systems do need observability. They need to be monitored like any other critical app. In fact, as I mentioned earlier, this takes [indiscernible] hire as AI agents can now take action. So we're building a full stack of products, so users can understand and improve their AI solutions.

So here's that AI for Datadog bucket, but laid out across our platform. So as you can see, there's capabilities across every single layer of our platform now, and we're building more. And here are the capabilities in the Datadog for AI bucket. Same thing here. AI-specific instrumentation [indiscernible] every single layer of our platform, and we believe we are uniquely positioned to win market share in each of these areas.

The last time we had an Investor Day, 2 years ago, we told you this. We wanted to make our customers [indiscernible] more productive for every single step going from code to business value from end-to-end. We call this closing the loop. We are well on our way to deliver this vision on behalf of our customers. We do this across 30,000 customers from the most tech-forward native companies to the largest Fortune 500 companies in every industry, in every geography around the world.

Here's the loop that we've been helping our customers close. For the production systems, it's an end left cycle of making or incurring changes, figuring out how to understand the affected systems and fixing problems as they arise. It's complicated, it's frustrating and it's expensive. And today, we save our customers time and money by breaking down silos across teams and data sets, and we close these loops tighter, faster and cheaper.

But there is a second key workflow that is becoming increasingly relevant, thanks to the rise of AI coding. And that's the loop that moves software from development to production. The time spent and value created have to this point, been heavily concentrated on the build part. But with the rise of AI coding, software engineers will work much faster. Already today, we have at Datadog experienced engineers who are building features 10x or 100x faster than they could before. And I think I've seen some of them in the room today. So please don't hire them away anywhere else.

So the value is instead shifting quite rapidly from coding to being able to evaluate changes and deliver business value. This is the space where code meets production environments, and that is exactly what bringing 100x more code to production will not be easy. The tough part is making it work in the real world. It will come into contact with all the other coding components in the environment. It's going to need to be reliable and scalable while being cost efficient. It will need to maintain the security of the business and user data. And in the end, it will have to deliver great business outcomes.

As you hear today, we think our investment in innovation, deep domain expertise, large and diverse customer base and massive [indiscernible] of data are all factors that will help us be a critically important part of the solution to this problem.

Our place in the world is right where code meets production environments, other applications, other agents, end users, and broadly speaking, the real world. And that is what we think is the most impactful place in AI development Datadog. At Datadog, this is what we've been building towards for the past 16 years. We've been around long enough to be part of the transition from monitoring to observability and to drive the adoption of DevOps at the time when cloud was just emerging.

With code development accelerating potentially by orders of magnitude, the problem we solve is expanding to be an even more important pervasive and valuable one going forward. And solving this problem will take us from observability to the edge of autonomy. Enabling autonomy will mean that we validate our customer systems, apps and agents. And that we do so as they are increasingly AI coded or rely on probabilistic AI models that are, by definition, harder to test or predict. It will mean that we help maintain their security and safety that we keep our customers' agents aligned with their intent and constraints. That we give our customers the right control mechanism and automate their feedback loops, and that we verify that every change generates the expected business outcomes. So that's where we're headed.

We think we have a unique opportunity to enable autonomy for our customers across development, operations and security to support our customers in their goal to rapidly deliver business outcomes accelerated by AI, to bring everything together end-to-end and to give our customers the ability to harness complex new technologies and deploy with confidence. This is the latest and by far, the biggest area of opportunity for us. As you'll hear later, our market just observability is very large and growing quickly. And even though we've successfully grown our business over time, our market share is still only in the mid-teens. And as I showed you, we have made significant headway in building capabilities in other areas, and those do significantly expand our addressable market.

Furthermore, we are building both AI for Datadog and Datadog for AI. And we're embedding both broadly and deeply into the Datadog platform across every single product or use of AI. And we're already delivering AI capabilities that our customers [indiscernible]. Meanwhile, our investments will deliver a quickly accelerating pace of innovation as our engineers themselves built with AI and [indiscernible] technology. And we are very focused on deploying those investments to achieve this vision of enabling autonomy for our customers. If we can get this right, the sky's the limit. And this is why we feel that we are still just getting started and barely scratching the surface.

Now we'll turn it over to Alexis Le, who will talk about our data-driven advantage and how we apply that to NextGen AI.

Thank you. Thanks, Olivier. Hi, everyone. My name is Alexis Le-Quoc. I'm the CTO and one of co-founders.. So I think Olivier gave you a clear picture of the [indiscernible] complexity that our customers operate at and presented very clearly the opportunity that's ahead of us. Given the platform we have and the platform we've been building for 16 years almost.

What I want to do now is explain further why we at Datadog are uniquely positioned to use AI to deliver on this vision. So data is unique in how much data we get, how much data we have, but also how much we know about infrastructure applications and systems running out there. We ingest data at a significant scale, trillions of data points, billions of traces, exabytes of logs. We also have a diversity of data that users are selling to us about their system. This could be besides metrics, traces, logs, user sessions, data jobs, lineage, LLM and agent traces, team structure, service names and many, many other pieces of information. And they come from our SDKs, from our agents and from our crawlers, from integrations. That data is, of course, what powers the Datadog of today. That gives the current observability to our customers.

But it is also the foundation for the AI needed to deliver fully autonomous operations. A few years ago, we started an AI research lab because we were convinced that given the amount of data we have and given our R&D capabilities, we could be leaders in building AI specifically for observability and security.

How did we start? So we wanted to prove that having lots of data, lots of domain-specific data give us an edge and thus was born our first foundational time series model in Toto. Now if you look for the largest public data set of time series data, you get to about 300 billion data points. And it covers domains that you'd expect; finance, health care, energy, transportation, some web traffic and so on.

But we train Toto on 3x that amount. And the vast majority of the data we have, we use for training, for pre-training actually, is completely unique to Datadog. And it's all related to applications, infrastructure, software systems. So as a result, when we compare Toto to other time series model and its ability to forecast, it performed far better than other models out there. So we reach state-of-the-art. And from there, what we did is, we released it as an open [indiscernible] model and hugging phase that -- I think that was last May. And we've seen a significant uptake since about 9 million downloads.

Now you may wonder why release it as open [indiscernible]? What's the point? Well, for 3 reasons. One, and maybe importantly, we want to contribute to the field. This is a nascent field that I think can use all the help it can get. Number two, it was important for us to establish our credibility as an AI lab. And number three, because it's actually a way to understand how important these models are by just seeing how many downloads they get. But I'd say the most important difference that we notice is the difference of cost compared to AI models you use every day. We spent about [ $750,000 ] to train this model in 2025, which was, at the time, 3 to 4 orders of magnitude cheaper than a frontier model of the same vintage.

Now sure, frontier model can do a lot more than what Toto can do, right? It can speak hundreds of human languages. It can review and amend legal contracts or analyze medical imagery. But none of that matters, right, in the context of observability. And with Toto, we can show that -- we get -- we showed that we can get good results with lots of proprietary data and small models. But I'll get back to that.

Let me talk a little bit about training. So really, when you have a -- when you're creating a model from scratch, you spend a lot of time and money on the pre-training stage, if you will, training is an important piece. And it's both pre-training and training are essential to produce models and agents that are useful in real-life situations. So let us see how we're training the Bits AI [ SRE ] agent. And if you're not familiar with Bits AI, sorry, it's what's in the name, it's effectively a site reliability engineer. It is tasked with finding and building a plausible causal chain starting from a symptom on a software system somewhere and sort of building the chain from there.

So when an alert signals an issue in an application, for instance, Bits AI SRE, develop hypotheses about weather problem, where the problem is coming from and [indiscernible] is all the data available. And the goal is to identify the root cause that can then be addressed. And so that the symptom goes away. And as you can imagine, and you'll hear more about Bits AI SRE, it's very popular because there are always problems out there on software stacks around the world.

Now in order to train an agent to correctly identify issues in production, we need a solid baseline of past incidents and correct root causes. And here's how we've done it. Like any complex system, our own platform is constantly evolving and constantly being maintained. Think of it like the [indiscernible] around us. The streets need to be plowed, [indiscernible] needs constant attention and effort to fight entropy, so that life can keep going.

So in the course of any day, our engineers investigate in fixed issues. And what they do to is the record their finding as well as the entire set of ability that was needed to reach a conclusion. And they turn that package into an evaluation or an eval. Because that eval comes from human analysis of an expert, and it was used successfully to troubleshoot and fix an issue, we know we can trust it. And every time we make a change to a model or a model instruction we run through the growing body of evals and see if the model gets better or worse.

So this is a chart behind me showing a number of evals that our engineers have recorded over time. This is not something that can outsource to nonexperts out there. And it is important that you cover as many use cases as possible. And for that, you need a large infrastructure to do it, and we have the large infrastructure.

And again, going back to the city analogy, to keep New York City up and running, it takes a lot more variety and scale of effort than if you only have to keep up a small village, for instance. And so as we recorded more evals, we've seen the accuracy of Bits AI SRE it has gone up, not as a straight trajectory, it's up and down, but we can see generally the trend. And so that's how we've improved the quality of Bits AI SRE. And normally, we're going beyond this sort of human curated set of evals in building. So [indiscernible] is to generate synthetic data, and that's really to reach an even larger scale. And we think it's necessary.

And as our customers have begun to use Bits AI SRE, they've actually -- the -- sellers feedback, right? This is useful, not useful. This has worked, not worked, and here's why it was useful and and why it was not useful. And we, of course, use that as evals. And that's great because it continues to enrich the diversity of environments and problems that our agent faces. The reality is there's no shortcut in training -- pre-training and training to get this kind of high-quality result. You just need to have lots of data and expertise. And that we feel is a strong differentiator for us.

So let me maybe sum up our advantage. One, we have access -- continuous access to lots of clean and rich data that we can use for pre-training, for training and -- of financial models and agents. Number two, we are building our own models. Number three, it's not only the shared amount of data matters, it's also the diversity. It has to come from a broad context. That's very helpful. And before, we bring to bear our domain expertise to improve our model and agent performance. So we think that we're unique in our ability to deliver the best in observability powered by AI [indiscernible].

Now you may wonder, okay, that's great. Why not just take the data and throw away out of the frontier model and see how it does. And thus, you would replace data [indiscernible] with just a bunch of data in frontier model. And maybe you have the data to do that. But first of all, is it enough? And what you'll get is you'll get -- so we've done that internally because our customers probably do that. What you can do is you can -- the Frontier model is going to be great at summarizing data. You have to give it a bunch of context. You're going to get some good results. The main problem is extremely expensive.

So here's the idea. Here, I'm plotting costs and accuracy, if you will. And read -- first of all, training a frontier model, it's starts at $1 billion. And it's unclear how high or how the accuracy slope trends as you go orders of magnitude past $1 billion. Whereas the research is telling us a different approach works and this we think is better. We've built much smaller models on the exabyte of data we have. And the small models, what we've proven is that they have orders of magnitude better accuracy per dollar because simply, we're not paying for things that frontier models have, but we know won't be useful in the context of observability and security. And even I'd say if you hire lots of engineers to fine-tune and do RHLF and so on, it's not going to be able to match the accuracy at the same cost. Because really, when you use the frontier model, observability is you're paying for the amortized cost, the pretraining, training and on top of that, all the -- so the vast quantitative hardware that's needed to run in front of the Frontier model.

So in summary, one, we believe that the autonomous operation needs very good models at a low cost. And number 2 is the approach with small dedicated models, lots of [indiscernible] training data and ongoing evals, we think, is the way to go. So I've shown you the reason why we build our models and also how we can improve them based on real life observation.

Now you may wonder how do we tie all this together to get to fully autonomous operations because autonomy requires validation, safety, security, alignment and control. In other words, we have to go beyond only observing a system. We have to understand its behavior and maintain control through verified changes. So in our lab, we've built, as I mentioned, a bunch of models, but we've basically understood that what we need really for this domain is a world model.

So how are we going to do that? We take all the data we have, and it needs to encode and represent the state of distributed system. Now we are already in a position to observe these systems, right, because we plugged in everywhere. We get that -- these exabytes, trillions and so on of data every single day. So what we're working on right now is an optimal representation of code, system structure, system behavior. And we think it is essential to predict the future behavior of software with high enough accuracy. For that, you need to understand past behavior and have knowledge on how such [indiscernible]. Without it, it's not going to work. Without -- and finally, without high accuracy, they won't be autonomy. So we have to get to a high accuracy.

And when we think about steps towards autonomy, we think in terms of stages, here represented on the right-hand side. Started with customized and adaptive observability than proactive alerting, automated remediation. And then you get to the predictive and proactive and preemptive sorry, -- and finally, you get to autonomous operation.

And with bit SRE, we've tackled the first stage, automating away sort of the more manual steps of observability. So the need, for instance, to build dashboards by hand and so on is gradually going away. And we've built proactive learning and automated remediation with the goal to take care of an increasing number of cases when currently people are seeing the loop to fix things. Now there, we think our trajectory is going to be like that of coding agents, limited in scope in the early days, more narrow, but increasing as time passes as we refine our models. And as that happens, as accuracy and coverage continue to increase, we'll be able to further predict and then preempt and prevent issues before they occur. When we reach that, we have a self-healing, self-managing, self-optimizing system, able to operate customers' infrastructure and application with fairly limited human intervention. And that is our goal, as Olivier said.

Now there's obviously a healthy amount of work to get there. But we think we're uniquely positioned to make it happen. That is where we're going because that is the only worthwhile goal. But for now, let us here about what we're delivering for customers today.

And for that, I'll hand it over to Yrieix.

Thanks, Alexis, and hi, everyone. So my name is Yrieix Garnier. I'm super excited to be here today for my second Investor Day. So Olivier and Alexis talked about how we're moving forward towards autonomous operations. In these sections, I want to take it back to its foundation, the Datadog platform. Platform is what Datadog started. And over the years, it really helped us break down silos amongst teams and data. So to do that, we've built a robust set of elements, thousand of integrations, common UI, data services. And we have about half of the 4,000 engineers working on the platform. And the investment allows us to be evolved very quickly and to seamlessly integrate AI to better serve our customers.

Bits AI, this is how we're calling our AI capabilities is really present throughout the platform. Let me give you a few examples. So it can analyze and correlate all the different data types that we're getting from our customers, to detect, investigate and remediate code fixes, or interact with Datadog through natural language. And we're just getting started with more AI capabilities to come.

But the platform is also critical for a rapid pace of innovations. By leveraging the platform and all these building blocks or engineers, team can really stay very lean but also move really quick to deliver either new products or enhance existing features. This is really the flywheel between platform and product. For instance, it helped a handful of engineers to deliver companies autoscaling end-to-end solutions. Or after the Metaplane acquisitions, we were able to take to market data observability in a record time. So today, we have dozens of products, and we continue to advance our platform to accelerate this flywheel effect and build even more product faster.

But let's look at the data injection side. We've also added a number of data sources including like valet, LLM inferences, and we can ingest extremely large amounts of data to provide analysis and correlations in the context of our products. But our end goal is really to deliver value to our diverse users, but also to the uprising number of agents. They all need the data in one place. We need to provide the best outcome.

As Olivier said, our customers are facing a rising level of complexity. And our job is to stay ahead of that. And if we do all that right, we provide a single set of truth to break down silos across all users and agents. So the platform actually come a pretty long way over the years. Let me reflect on that.

In 2015, we started with few integrations. [indiscernible] millions of events per hour and only had one product in from monitoring. But our customer needs and demand grew really exponentially. And today, we have over 30,000 customers, including some very, very large ones. We have 25 products and we can handle trillions of events per hour. As you can see, our platform reached unprecedented scale. And it's one of the main reason our customers keep choosing Datadog as we can store, process and move data with always improved performance or cost efficiency.

But let's look at this from a customer angle. This chart is a very fast-growing AI company. The customer demand grew very quickly. But where the end user were experimenting bottlenecks, they needed a unified [indiscernible] solutions to get to the root cause as fast as they could. And to do that in about one year, they adopted 16 Datadog to products, and that was really key to their business success. So that's a platform.

Now let me speak of the way we're scaling and our customers just retain and analyze the growing data sets. So to expand scale, let me take the example of the log management. We launched it back in 2018, and it had multiple scaling phases. The first phase was called logging without limit. And it's all about ingesting all your logs without limit and only processing the relevant ones with correlations with infra monitoring and APM. So bringing it logs in a valuable but still economical way is really the foundation that helped grow that business 7x to $1 billion ARR. But really, it's the foundations of our customers value. And it worked great for observability logs.

We also knew that we are only capturing a fraction of our customer logs. Of the use cases like transaction logs or audit logs typically involve log volume that were other magnitude larger and which needed to store for a much longer period of time.

So as the second stage of growth, we've extended the capabilities to support those use cases with Flex Logs, Frozen and [indiscernible] Search. By doing so, we've unlocked new market opportunities and delivered more and better outcomes to our customers. And it actually worked. We saw very strong adoption of Flex Logs from the start. Now our customers are storing tens of trillions of events and Flex Logs is approaching $100 million ARR and growing very rapidly.

But to keep pushing on this log journey, we've leveraged the platform to build a new product, Cloud SIEM. Cloud SIEM is actually a natural extension for text logs and it requires logs to investigate security issues over a long period of time. And Flex Logs really unlocked the security revenue. With 18x growth in 5 years, we are still seeing an acceleration in Cloud SIEM adoption. And if we look at this, we are still actually at the beginning of that opportunity phase here.

Sorry. So as you can see on the chart here, Flex Logs really unlocked that security revenue. It grew like 18x in 5 years. And we do see that acceleration. Yes, I just already said that. So Tim is going to talk more about this as we actually do see that the beginning of that opportunity.

So let's now look at the impact of the platform data scale and actually a whole customer journey. These e-commerce customers started using infra. Then in 2022, they adopted APM, Synthetics and Logs. And in Q4 2023, they become a Flex Logs and SIEM customer. As you can see over time, they consolidate more tools within Datadog, which enabled them to address an increasing number of use cases, cover more environment, larger data sets, and most importantly, enhance their customer experience. And that's how Datadog and customers really scale together.

So to finish, I want to talk about enterprise coverage. This is where the platform effort help our largest, our most sophisticated customers to drive tool consolidation. Beyond monitoring everything in the cloud, a number of our customers asking us to cover more of the environment by combining cloud and on-prem. With our extensive integrations, this is something which is out of the box for Datadog. We're already supporting on-prem servers and network. And now we also monitor wireless access point, end user devices, like laptops and desktops or edge devices. So with that, customers can really see the entire physical footprint in one place and also combined with the cloud environment. And for us, it means a larger footprint and a bigger market opportunity.

We also continue to expand our coverage across our customer tech stack. Historically

[Audio Gap] for cloud security, we provide posture scanning, run-time vulnerability detection, attack analysis and combine that with observability enriched prioritization to tell you what cloud risk matter the most. With AI and data security, Datadog automatically secures sensitive data and offers comprehensive run time protections for AI agents to enable safe AI transformation.

And finally, for developers and DevOps teams, we have code security, helps to identify vulnerabilities and first-party code, open source libraries, infrastructure code, all before they move to production and therefore, dramatically reducing alert fatigue. And with Bits AI, code security generates bulk remediations, so developers can still deliver secure code while getting to spend more time on building and innovating.

Now plus our security products, because they're built on the Datadog platform, they can all take advantage of Datadog's shared services to accelerate remediation. For example, integrated case creation and incident response to security, SRE and DevOps can rapidly collaborate with one shared view, or even one step further, agent builder can be used to build custom automated remediation workflows. So we have growing proof that the data log advantage that I just outlined of unifying security and observability is delivering value to customers.

Today, Daydog has over 8,500 customers using our security products. This includes one in 4 of the Fortune 500, and we've now surpassed $100 million in ARR. And you know who else is using our security products, we are. The Datadog security team secures the Datadog platform for our 30,000-plus customers using Datadog. We put our reputation and our business on the line using our own products as an indication of our confidence as a security vendor. So we're off to a good start. If you wanted to pause [indiscernible] okay.

So we're off to a good start. We believe we have a big opportunity ahead of us. Today, 70% of our $1 million customers use one or more Datadog security products. But the spend security represents, it's together is only 2% of their Datadog spend. So as a result, we see potential for much more wallet share as we deliver more security products and capitalize on that Datadog advantage.

Let me give you an example. Here is a long-time Datadog customer in the media market, who over time has adopted a number of security products from Datadog. They wanted a unified observability and security platform, and they chose to consolidate on Datadog over market alternatives like Palo Alto, CrowdStrike, Wiz, Google, Microsoft, just to name a few. As a result, today, 20% of their Datadog spend is on our security products. And that could just be the start.

For another one of our $1 million customers, 38% of their business with us is security. So we see a clear opportunity with all of our current and future customers to go deeper and broader with security. And how we get there is the value of the Datadog advantage. It combines and correlates security and observability in a truly unified platform.

So that's it for me. Let me hand it over to Michael to talk about what we're doing for developers.

Thank you, Tim. All right. I love seeing everybody taking a ton of note. Anybody using AI right now in this very moment? I knew it. There are some people out there -- advantage, right? So I'm Michael Whetten.

So we live in this era of speed right now, right? I don't know if you all feel it, but my customers are under intense pressure to compete and innovate and build and ship product that works to customers and try to scale out as fast as possible. And it's been like that I've been in Datadog almost 10 years now. It's been like that since the beginning with the cloud, this competitive advantage that technology can bring but it feels like it's accelerating, right?

But as Olivier said, the complexity is a drag on their ability to innovate, right? The big companies have a lot of complexity, and they're feeling the drag in different ways than small companies who are having trouble getting to scale, right? And they have drag in different ways and different types of complexity.

And if there's one thing that you take away from today, it's that -- I know some of you have traditionally written about Datadog as kind of an insurance company, utility company or some -- must have for companies that scale. But more and more, my customers are telling me or they're adopting even the fastest-growing companies in the world right now who are in the most hypercompetitive landscapes, they're buying Datadog because they need to move fast and Datadog enables them to do so.

So let me walk you through a few examples of how I see this working. So one type of complexity that we see a lot is fragmented visibility, observability or monitoring or product analytics or whichever point solution they have here. So here's a simplified diagram of an application, a single application at a company. The fragmentation, each of these tends to represent a different team at a company, right? And even though it's a single application, and this user is interacting with your application, making requests, getting response traversing through whatever your application does, the fragmented nature of the organizations that are required to serve this user, they don't always provide the best user experience so that when an issue happens, right, if I go to a lot of my customers right now and I say -- or potential customers, usually, I say what's -- my favorite thing is kind of a [indiscernible] question.

But my favorite thing to ask them is where -- how do you know that things are broken now? What triggers an incident? And most of the time, those cheapestly look at the floor and say, well, customers, like a support team where somebody tweets something is down and then we call an incident and immediately start reacting. And some of you are nodding because you must hear this from people as well, right?

And the problem is, is that all of these different teams are using different tools. So we have a bit of a tower battle problem, right? And who knows that these signals that are going off even point to the root cause of the problem. So to use a real-world example, there's a major global bank in which they have 5,000 engineers. So that was one application.

Now at this bank, there's hundreds of business units, and each one has many applications. So 1,500 applications spread across the organization, right? And the problem that they have is these -- the way that the organization is set up doesn't match the user experience, somebody might be trying to make a withdrawal or deposit at an ATM or on their phone. They don't know how the company is set up and whose problem is who. So they just know they can't do something.

But when something goes wrong, these are all interdependent technologies, and troubleshooting this is a nightmare, right? So they consolidate everything on to Datadog. They went to all those point solutions and brought it all in. And the beauty is that, not only is it just one single pane of glass, the real advantage is that from the time of collection till the time it lands in Datadog, we're automatically summarizing and correlating all that data. We're making sense of that data. So that when it lands here, most of the work is done. And when a signal goes off, it's typically in the right place, notifying the right team and they can [indiscernible] responding much faster so that you have global impact within the organization just from consolidating onto one platform, right? So millions of dollars a day in avoidance, better customer sentiment.

But the real advantage, the real value isn't actually displayed here. The real value is all those people that were responding to those incidents that were taking hours and now take minutes. Those are sometimes the highest paid or smartest engineers at your company. And rather than building value for the company and living the bank forward into what is the next generation of bank software, they're spending time and incidence. And here now, they can spend time adding value for the customers.

So ripping and replacing load-bearing technologies that have been there for 10 years across 5,000 engineers, it's not super easy for a lot of organizations, right? Either it's a mandate as it was at that bank or a trend that we see happening now to address some of this complexity and scale is a new movement that we're seeing where people are calling it user-first monitoring. So much complexity on the back end, so many teams to coordinate, so much politics to try to coordinate. Can we just start with the user experience and start making our way backwards, right? What are the most load-bearing critical user journeys that we need into in visibility and start traversing the organization that way?

And so the crux of your business really is your application front-end stack. That is what your developers are typically making, right? And then they scale it out with infrastructure network. So we have a suite of products for this solution and the space is called Digital Experience Monitoring. But really, it is what is the user experience, how are they impacted by the changes my engineers are introducing into production? Are they making things better or worse? And so in some ways, this is accountability software. Are my engineers who I'm paying who are there to bring value to the customers, are they making things better or worse for the end user?

So we see a big push towards enabling engineers to have direct visibility into the business impact of their changes. And so this is one of our fastest-growing areas. And we see that -- when the APM, which is the back-end instrumentation and the front-end user experience are stitched together, it's a story that works and it spreads through the company much faster and it spreads through organizations much faster, and we see that when these things are together, we have -- we provide a lot more value, which turns in more revenue for the business.

And this isn't unique to digital experience and APM bundling, as customers grow with us, they do find more value. The land and expand does work and more products equals more value, spreads in more parts of the company until they have that consolidated approach. Why? Because this is better for the human responders, right, bringing more people into the conversation to troubleshoot faster together is meaningful and does increase the value.

But an advantage of this single platform and that automatically correlated and summarized data that we do is it's better for our AI agents, too. We found that when the AI -- when we have the full context of the entire stack and is already correlated for humans, AI also operates much better on that. It can act faster, right? So what might take a team of people before many hours to solve, but then they adopt Datadog and they can do it in maybe one hour. The AI can come to the same conclusion in minutes, and we'll hear more about that soon.

But it can traverse. This is, again, one application. It can reverse all the applications. It's working 24/7, right? And so the advantage here is when it has -- you can't separate the AI from the underlying data formats and context. There's a lot of work you can do to make the AI more efficient, faster and more important, more accurate.

So Datadog does bring equilibrium into that DevOps life cycle. So the developers can develop as fast as they can. What are they doing? They're introducing change into production. That's their job is to introduce new things, but production can adapt to that quickly and make sure that it's done safely and securely. At Datadog, we benefit from this. The reason that we're able to ship so fast is because we use Datadog. And so we are the embodiment of that DevOps life cycle, and we do have that equilibrium. This last year, we released -- we announced and released hundreds of features into production, right? These aren't ideas, they are actually out there in customers' hands, iterating with customers, proving value.

Now as has been said already, there's a sea change, right? When I walk around Datadog office, I see a lot more engineers coding on their phones, having multiple windows open, directing lots of agents and writing code 24/7, they say it's addictive, right? And I'd say it's only accelerating. We've got one guy who's coding with his glasses, right? He's sending prompts and he's saying, looks good, now change this, right? And so this is a real movement in the industry right now.

But what does it do? We don't hire developers just to write code. So 100x more code might not actually translate to 100x more value. Are you really going to release 100x more features? Are you going to release one feature that's super sophisticated and can do a lot more? How do we verify that this 100x code is good that it works and that is a good product? There's a difference between code and product. So this is -- the bottleneck is no longer coding as Olivier was talking about, but bringing value to customers maybe at 100x. And this is where Datadog has a unique opportunity to help the AI movement forward. So developers love to make changes and bring things into production, but they break things.

The number one cause of incidents is faulty code, right? And that's when humans are reviewing it and taking painstakingly efforts to make sure it's good. But Datadog has automated integrated code testing. We also have the best understanding of how production works, how that code is deployed, how it's connected to other parts of the application, what data is flowing through that code at all times, that we can actually inform the entire life cycle here, whether it's humans or agents with AI coding, we can tell the code that you're writing right now, hey, this code that you're -- this idea that you have that you're going to propose into production, it's going to increase latency. It's going to make things slower because look at all these requests coming in right now and feed that into the agent. So the agent can make better decisions or so that the person whose hand coding can make better decisions and write better, more efficient, more secure code that we already know is going to work in production.

Now as you're sending it out, right, there's a risk mitigation strategy that already exists. You write your code, you write unit test, you run those tests locally. There's Git hooks in there that's going to run all your test locally, then you push your change in some centralized pool of [indiscernible] called CICD and it runs a bunch of tests and those tests, sometimes they pass, sometimes they fail and you have to kick it and it goes again. Then it will land in a staging environment where people kind of vibe check it, type of few things, looks okay, let's push it into production. If you're more sophisticated, you got feature flags, which means that you're going to slowly roll it out region by region, 5% hit it, 10% hit it, 15% hit it, what are you doing? You've got Datadog at least you've got monitors there to tell you if things are going off. If you don't -- usually, you're just praying and waiting to see if requests are going to come in, we can actually do much better than that.

We have automated testing and we just launched our feature flags product, which is unique in the domain in which we can actually contain or sandbox that change. When it goes to production, we can attach metrics to it automatically. And as it starts to roll out, we can statistically tell you this is going to be better. This is going to be worse. Roll it back or go straight to 100%, right? And so that's feature flags and experiments.

So but the gist here, and I can talk about any of these things, but the gist is we can automate much of this, just as AI is pushing more code or AI and the manual tooling that we're enabling for customers as well, which the AI will use, can accelerate this path into production. So rather than becoming a bottleneck or a super high risk and just breaking things all the time, we can reestablish that equilibrium even at the pace of AI.

So that's -- my throat is getting dry, but I am almost through it. So the AI -- or sorry, the code gen that most people are doing right now is to just build existing applications. And there's some exciting stuff. Technology does empower the creative individual, right? So I had -- somebody last night told me that from the time they got a white paper for a new database -- this person works for a database company. So research paper, they went [indiscernible] the idea into a working prototype and they thought they were so smart and novel and they were the third one on GitHub to have put it up. And this is like in hours after the white paper was published, right?

And so this is what I mean by speed. Well, everybody is trying to take advantage of -- they're trying to build more intelligent applications and they're trying to also see if they're going to be safe in the environment. So it means that us as Datadog, we have to continue to build new types of products, new types of observability, new types of security to be ahead of the curve when people need to start bringing these things into production.

So one of those things -- when I do talk to CEOs, CTOs, CIOs, one of the big concerns right now is the proliferation of agents in their company. What are these agents, right? Who made them? Why are they there? How much do they cost? How are they permissioned? What are they supposed to be doing? Are they doing a good job at that, right? Am I getting value? I see 5 of these agents that are supposed to be doing the same thing, should we choose one of them, right?

So the AI agent [indiscernible] gives leadership a way to actually track and teams who are building these things a way to track all of these questions about the agents that they're bringing into their market. Now those who are building these agents are trying to enable their product with more intelligence, they're also stuck unless you're at a leading research lab or Frontier model lab, you probably don't know where to get started. And so there's a lot of information out there.

The nice thing about AI observability product is that it comes with out-of-the-box framework. So just by using the product, you start to understand more about if you're already an expert and you'll be familiar about these tools. If not, it's almost an on-ramp in how to think about nondeterministic applications. I need to evaluate these things for basic competencies before I send them out into the world. Once they're in the world, I can experiment to see what is good behavior, what is bad behavior or trending which direction, I can try different models, et cetera. I can sandbox and I have playgrounds, all the stuff that these researchers and AI engineers need.

So the momentum on this is growing. There's been a lot of investment in learning over the last 24 months or so, but we're starting to see these things try to get into production and who helps people get things into production is Datadog. So that's why these products are growing with us now.

And the last thing I'll say here is we're not only getting started, right? This movement and helping this movement to figure out what it is and what it will become. Datadog is perfectly situated to enable that and to talk more about the exciting things we're doing with AI is our Chief Product Officer, Yanbing Li.

Thank you, Michael. How's everyone doing? This -- you are getting to the last stretch of our first half. So my name is Yanbing Li. I joined Datadog as the Chief Product Officer about 18 months ago. Before Datadog, I spent time at Google, responsible for the observability function powering Google's [indiscernible], infrastructure and services. So when I went to speak to a senior SRE leader to get some candidate feedback, this is what they tell me. They simply said, go look at Datadog. And this was back in 2019 before Datadog was even a public company.

After Google, I got to lead engineering and product at autonomous trucking company or innovation, actually, the very first company to operate commercial driverless truck on the U.S. public roads. At Aurora, get to learn firsthand what it meant to ship safety-critical autonomy product into production and at scale. So this is why I'm excited to be here at Datadog to help our customers ship faster without breaking things and operate reliably and safely, all while navigating the increased complexity of AI.

So let me circle back to this DevOps loop that Oli showed earlier. This is the reality of what our customers DevOps team leaves through every day. They need to detect issues as they emerge they try to investigate and find a root cause and the next step of action, and they take action to remediate back to health. And because systems are always changing, with new code, new traffic, new dependencies, this loop just doesn't stop.

So what happens when a major accident -- when a major incident happens to a production system? Our largest customers often tell us, they need to mobilize tens or even hundreds of engineers because they bring different knowledge, different data, different tools and also different system boundaries. And also, most of those teams, not only they have a partial view, they're motivated by proving it's not their problem rather than finding the problem.

So the area under this curve represents the time and times resources that's part of this operating expense. And certainly, incidents, we all know is very expensive to business outcomes with lost revenue, lost customer trust and reputation of risk. So this is a structural efficiency we are trying to solve at Datadog.

So Datadog is in the business of keeping this DevOp loop healthy and running for our customers. So when there is production stress, we detect issues, we help coordinate the customer's team, get the right team involved to investigate and take action to remediate the system back to health. And the previous speakers have already talked about Datadog's unified end-to-end observability platform can shorten the incident response with fewer people, less time and closing this loop faster. So the result can look something like this with faster detection with the right team involved with the right information, they're solving the incidents much faster.

And we all know when you have an incident, time is money. Let me take you through this with a concrete customer example. So this is a major U.S. insurance company who's been a customer for 5 years. Before Datadog, they experienced thousands of severe incidents every year. And with Datadog, after they standardize on our core pillars of product, they begin detecting and fixing those issues proactively, preemptively before they became real production escalations. And as they have seen a 10x reduction in their severe incident count.

And certainly, when they have fewer incidents, when they solve them faster, there is a significant boost to their engineering productivity, and they're saving about equivalent to 70 employee years and translate to $11 million every year.

So what does this mean from their business point of view? Again, with fewer incidents with faster resolution they are actually seeing a whopping 20x reduction in customer impact that's caused by this incident. So this is the kind of value we've been providing to our customer with our unified platform. Now by applying AI, we're taking that to the next level. We've launched a fleet of Bits AI agent. Yes, in Bits with those futuristic sunglasses. And so -- so we're helping our customers autonomously detect, decide and taking action so that we're closing this loop even faster.

So let me give you a few examples of the Bits AI agents and starting with the SRE agent. So you've heard about this several times throughout this presentation. So why SRE? By now, the world has recognized that the future of coding is going to be AI coding. But still, a lot of our customers are struggling to really measure and establish the real ROI. And we pick SRE as our first agentic effort because not only our primary user personas are SREs, but also when an incident happens, it's often acute, high stake. And better yet, the verifiable results of what Bits AI SRE can do is very obvious to our customers. Actually, many of our customers to test Bits AI ASR, they simply play back all of their previous serious incidents and see if Bits AI can get it right.

Obviously, the business outcome is also very tangible when you can reduce incidents. So because of the verifiable nature, our customers are really excited about what Bits AI SRE can do for them.

So how does it work? Don't worry, this is not an eye test. Let's focus on the left-hand side. When alert triggers, Bits AI autonomously investigate the issue. It first gather all the necessary data and relevant context, it then can reason like a group of engineers in different parts of the system, try to establish multiple hypotheses of what happened and then investigate all these hypothesis in parallel.

The right-hand side is intended to show you how that parallel works and is very visually explained to our users. It then identify the root cause and even can propose the next step of action based [indiscernible] and better yet, Bits AI can learn and it's getting better with every investigation.

So the superpower of Bits AI SRE also comes from this holistic understanding of our customers' entire environment, spending systems and applications and users and teams and even business processes. So it doesn't just leverage the rich real-time observability contacts and telemetry inside the Datadog platform is broadly integrated with third-party knowledge stores and also third-party telemetry. So our customers can really get that full picture of what's happening in their system.

Even though we're still in the early days with Bits AI SRE, we are already getting a lot of positive feedback from our customers. Here, you see 2 examples that customers are telling us how Bits AI SRE can accelerate their incident resolution and how it's acting like experienced engineers to help them understand their complex system. And when they matter the most, our customers are also telling us Bits AI SRE get the job done.

So if you remember, the major AWS outage last October, we've got many customers reach out to us saying, when that outage happened, Bits AI SRE was able to autonomously root cause to the outage before being notified by AWS themselves. So even though Bits AI is AI agent, it actually gets a lot of love notes from our customers. As you can see on the screen here, and not just because this is a Valentine's week, we hear this from customers all the time how pleasantly surprised at how smart Bits AI is, how it gets to the root cause, how it's saving them time, how it's boosting their productivity.

And the best indication of that is the actual usage. For new product, we look at the usage metric very, very closely. So since the Bits AI SRE launch, we've had our customers run well over for 100,000 investigations and since our GA last December, this rate is increasing and accelerating. And in January alone, we have more than 2,000 customers run investigation with Bits AI.

Okay. So let me switch gear to talk about how we're using AI for some of our other use cases and products and starting with this security example. We have our Bits AI security analysts in preview. And so this agent can autonomously investigate Datadog's Cloud SIEM signals and conduct in-depth investigation for potential threats and also enable users to remediate those threats all in the Datadog user interface.

I think a better way to explain this is a real example. So a major financial services company was testing Bits AI security analysts for the first time. And actually correctly identified a live, serious security threat. So the situation is a compromised automation system in their environment change their cloud firewall setting such that it's open to the entire Internet. Some of the sensitive management ports are open and exposed. Does this sounds quite serious?

Yes. So without Bits AI, so the investigation would happen that they may receive some security signals and that goes into a queue and human investigate them one by one, and it could take hours for them to come to this realization versus was this security analyst agent could do investigating in parallel. And within minutes, we were able to service this severe threat to the customer. Obviously, the customer did become true believers of this technology afterwards. And this is just an example how Bits AI security analysts can truly transform how security teams can investigate and resolve security incidents.

Let me give another example because Bits AI SRE and security analysts, they're doing investigation or trying to understand what's happening in the environment. What about autonomous remediation. So this is why we introduced the Dev agent. So the Dev agent can automatically analyze the telemetry and code when there's an error happened in our customers' system. It can explain the root cause in plain human language and even map it directly to the relevant code, files and function. And then proceed to generate a context of where fixed and [indiscernible] is important because this fix would be generated based on the real production contact that Datadog uniquely brings to this problem. We can then proceed to test the fix in isolated sandbox so that you have high confidence that this fix is ready to push to production. And all of this can happen without a developer even logs in. And of course, the tool can interact with the developer. When they do log in, they can review the code, they can ask questions and they can help also merge the PR.

So you may be wondering, is this yet another AI coding agent? Now there are already so many on the market. The answer is no. Because Bits AI Dev agent is deeply integrated within our DevOps loop to truly [indiscernible] bringing that full production context, so to help create a better PR. It's also very proactive. A lot of our customers are really pleasantly surprised that they received a Slack message from Bits informing them, there is a high severity error, and Bits already fixed it for them with a PR that's ready to be merged.

So I shared 3 Bits AI agent example. And the important thing is with AI, how our users and engineers are interacting with their tool is also radically changing. So the good news is our customers can use Bits AI from anywhere that's fitting into their workflow, whether it's in their Datadog UI or through collaboration tools or from their favorite IDE or getting involved by another AI agents. And all of these interfaces are also enabled by the Datadog MCP server.

So the Datadog MCP server enables our customers and the AI agent to access Datadog's docs AI-driven observability [indiscernible] from -- directly from their existing workflow. And as you can see on this chart, since the launch of our MCP, we've seen also exponential adoption and growth. And many of the customers are integrating this to their existing workflow. They're also building custom AI agents so that they can build those agentic workflow to help them with incident investigation performance optimization and many other use cases.

So with Bits AI and MCP, now incident resolution can look like this. We can help our customers narrow to the root cause, take action within minutes and with very few people involved as opposed to the tens and hundreds of engineers working tirelessly over hours and days. And better yet, the Bits AI agent can easily work alongside human SREs and human security analysts and developers to make them far more productive. So we're closing the loop even faster. That's the value we're bringing to the customer.

So I just spent the past 10 minutes also taking you through how Datadog is solving the structural inefficiency in the DevOps loop by providing an end-to-end unified observability platform and now turbocharged with AI that we are helping our customers closing this loop very, very rapidly. And as Michael and Olivier alluded to earlier on, we've also been shifting to this loop on the left-hand side to the preproduction environment to help our customer ship better software into production.

So our long-term vision, as Olivier outlined, is to achieve autonomy across Dev, Ops and Security. So that will require us to help our customers validate their system, their application, their AI agents, and in addition to helping them shipping production ready [indiscernible] hold faster and preventing incidents from occurring at all, we have to help them maintain their safety and security to achieve true alignment of their AI application toward their intent and business outcomes. And in the meantime, give them control and feedback to help them improve.

So I am personally very excited about this vision. It is special having built autonomy for trucks now building autonomy for development, ops and security. So at Datadog, we are very excited of this vision that our customers need us to bring this together now more than ever.

With that, thank you, everyone. And I will hand it over to Yuka for Q&A. Thank you.

Okay. Thank you, Yanbing Li. We are going to start a Q&A session. Now joining me on stage are all the print centers you just heard from. Their names are up on the screen. We are -- we're going to be taking questions from the in-person audience. I have 2 of my colleagues, Megan, on your right; Eric, on your left with mics. And we'll alternate between them. So please, yes, raise your hand and wait for them to get to you so we can all hear your question.

We're going to start on Megan's side.

It's Kirk Materne with Evercore ISI. Tim, I was wondering if you could talk a little bit about the silo tax that you brought up. I mean one of the reasons there's always been silos is buying silos, meaning you've had SecOps, DevOps, IT apps have different budgets. I was curious as we head into a world where you need more talent [indiscernible] across all those areas. Are you seeing those budgets collapse into one? And if not, how do you make sure that you're talking to the right person to get more sort of visibility on the security [indiscernible].

Great question. So I think we're seeing 2 things right now. Number one is the recognition that we can make security easier, better with the unification and consolidating on a single platform. And it's more just the [indiscernible] or no longer the need to have those separate budgets for those items because they can get it all through the single platform. So that's one item.

The second point is, I think there's also a broader recognition that, in fact, we use this ourselves at Datadog that having organizationally security and SRE, the same word has a lot of benefits, particularly when it comes across the board for both reliability as well as security. The team is working together. Of course, in our case, using our own platform and tool sets it will work much more effectively. So I think that's something that will also emerge over time is probably a growing trend across organizations as they see the need to be more effective.

Just to add up to that, it's not completely about the [indiscernible] between buyers. But if you think of the impact of coding agents, there's going to be much less separation between the roles like I think as we build a lot more, a lot faster roles that used to be [indiscernible] as how you're building it separate from whether you're building the right thing, which is separate from whether you or not [indiscernible]. [indiscernible], which is separate from who you operate it. I think all of that gets merged together quite a bit as the coding itself and how you build it is left to the agent.

Great. Thank you. Eric's side.

Alex Zukin from Wolfe Research. Thank you so much for the presentation today. I particularly wanted to ask about Bits SRE agent. And given kind of the increasingly heterogeneous environment that is increasing the magnitude, scale and complexity. When you think about the ability to read from other kind of data sources outside of the Datadog perimeter and compiling kind of complete tasks across that information. Can you talk a little bit about kind of where your tool, how you differentiate in that context versus other folks building there? And then maybe some pricing and even a customer example kind of who's at the frontier showcasing.

Great question. So when we started Bits AI SRE, we actually focus on the data and telemetry that's within Datadog because the most important thing for AI agent is to show that they can get it right. It shows value to the customer. And when we are within the platform, we have real-time rich data that allow us to really showcase that value to our customers. Obviously, a lot of our customers do have those heterogeneous environment. So now we're expanding our telemetry to cover those outside data sources. So yes, so -- and we also see there's a lot of other start-ups and company out there that they probably take an approach that's looking from the outside. And so the way we are uniquely able to bring the power for Datadog but also integrate with those external data sources have shown that we can simply generate better outcome and results.

We'll be more right with the data we have and we -- for which we have more -- or I would say more higher resolution, more reach, et cetera, et cetera. If we took out the whole stack, we can be more right, which is also why that's where we started. But I will say, when you look at where the market is today, the market is very active today. The market right now is you have an issue and you [indiscernible] and to be honest, you can get a pretty good result if you ask [indiscernible] to do that. It can ask a number of different systems. So for that particular thing, I think it's okay.

For where the market is going, which is you want to be [indiscernible] proactive and you want to prevent issues that just doesn't work at all because the data doesn't flow through [indiscernible] different systems. [indiscernible], back to autonomy in self-driving. So after [indiscernible] as today, you can send the pictures to ChatGPT and it will help you tell you who was right or wrong, but the crash happened. But ChatGPT is not going to drive the car. You have a separate [indiscernible], you have a separate everything. And I think the same happens is going to happen to us with observability. As we get all of the data as we control the data plan for that as we can run -- develop and run models live on all that data, we'll be in a position to get in front of the issues and prevent them.

Yes. And in terms of the customer adoption, as I mentioned, we have 2,000 customers. So the product is still fairly new on general availability. So we're in the process of getting -- make sure customers can let us use their names. But what I can share that our 2,000 customers is widely represented in all kinds of [indiscernible] and verticals and geo locations from the largest Fortune 100 companies to the most innovative AI start-up. So there is a fairly broad non-discriminative adoption of this technology. And this is why I'm personally excited about SRE is really a very strong well-fitted use case for AI because of the result is in [indiscernible] and verifiable. So this is why we're seeing such a rapid increase in adoption.

Then finally I just mentioned that our pricing is completely transparent. You can go to our corporate website, and I believe that Bits AI SRE $500 per 20 investigation, right, Kai? So but you can check out all of that stuff by yourself whenever you want.

All right. Megan? Megan's side

It's Keith Bachman from BMO. Tim, I also wanted to direct this to you is how do you think about the boundaries within security about where you want to be in terms of portfolio expansion and talk about some areas of interest or where you don't want to be?

And the reason I bring it up is to your slide deck, you're still a relatively small part of generating ARR for Datadog. And some of your competitors have a much broader portfolio [indiscernible] a consolidation game. So in some measure of success begets success. And so I'm just wondering about how you think about portfolio expansion to try to get deeper penetration within your existing customers?

Yes, something I think about daily, hourly, minute by minute. One of the things that leads our thinking obviously is where there's existing mature spend. And we believe, as you saw in my overview of the portfolio that we are well positioned to go after areas of established budget, established spend. They also have a lot of established competitors. But then again, as you probably heard me say during my -- repeatedly in my piece, I think we do have an advantage because of the fact we have this unification in the platform, which really pays off when it comes to incident response or is being more proactive with security overall.

So right now, in the markets we're in right now, we see a lot of runway to go after that. We see the advantage that we can bring for differentiation. And we also see a strong ability to pivot off of our existing relationships to get into those security conversations. Now even within those areas, obviously, with AI and agentic, there's going to be new areas, new services that we'll be looking at because it's a new set of problems for security teams and for the enterprise as a whole. Outside of that, obviously, we'll look at where it makes sense to expand where we bring that advantage we have with our platform.

Great. Eric side?

Ittai Kidron from Oppenheimer. And my question is to you, Alexis. Your presentation was quite interesting, and thanks for making the case that frontier models. I guess, can't scale or do what you do without going outrageously cost or ineffective giving the results that deliver accuracy. I guess when I look at your business, you talked about the future massive feature set that you have, the domain expertise and your ability to take a small model, right, and make a much better use of that in your business. I guess if we try to flip it on the upside down a little bit, the question is if you take $750,000 to build a small model that delivers much better accuracy, how do you think about the barrier of entry from third parties into your business. How do you think about the risks? I mean, the opportunities are clear. The revenue is clear. What, in your view, are the risks of AI to your business?

I think the -- this is where the data advantage, I think, is clear. And it's -- I think on the -- so in the case of our [indiscernible] model, it's just -- we just have a volume of data like legitimate real data that's just not publicly available. So that's one edge. I think in the case of the sort of the training of the agents, it is the -- both the volume and the quality of evals we can build, and we have that I think differentiates us. So I think there are obviously other companies trying to build small models. They don't -- there's no -- I don't think there's a clear financial barrier of entry, but the quality of the data you get, I think that's the real moat.

The -- one of the issues, and we see is for instance, generating synthetic data in our domain, it's not terribly easy. It's not like you can basically sample what's out there, text to image and then you remix and create something plausible. It is much more if you will, the relationship between the way software is built and the observed behavior, it's much more intricate. And so you can't just go to one of the providers and say, okay, I want synthetic data. I want tons of it, so I can train my sort of small models. So that's where we retain advantage.

I think the other piece is what Olivier alluded to, which is, look, we just sit that data, we get it for free as anywhere for training purposes, right? Because it's used for by our customers in the day-to-day. And that's where someone else just isn't sitting in that flow. They have to require that flow somehow. They don't have the scale to do that. So it really -- I see it as a positive flywheel that we get data, we get more and more data. We can get the SRE agent to generate more and more evaluations, which makes, hopefully, Datadog more and more valuable. So it's really accretive in that sense.

And remember, we -- so we spent $1 billion in R&D last year, maybe between $700 million and $800 million the year before and then another $500 million the year before that. So spending all this money is why it costs us $750,000 to train the model.

Great. Megan side.

Sanjit Singh from Morgan Stanley. I wanted to get the team's view on how fast are we racing towards this vision of autonomous operations. Like what does this look like a year from now? What does this look like 3 years from now? And in terms of executing is that vision, are there other pieces of the stack that Datadog needs to own? Do you need to own the software delivery pipeline to really execute on this and that gets into a build versus buy question. So just on the -- in terms of the race to autonomous operations, what does that look like 12 months from now and over the next couple of years?

Yes. I think it's really hard to tell. If there's one thing we could tell with AI is that the rate of progress is very surprising. So you get these big jumps of capability. And then looks like things are sort of stagnating a little bit. And that when you're ready to write it off and it starts jumping again. We've had a jump recently just 2 months ago in the quality of coding agents, for example, that made a very notable difference, at least in our usage internally and what we can see from our customers.

So I think it's smart to tell whether we get there in a year or in 3 years. But what's pretty clear is that we're going to get there. Like the problems are getting solved one by one, the technical approaches, look like they're working. So we're going to get there.

In terms of the moving pieces, we had identified a few areas that -- where we thought we needed to move faster. So one of them is future flagging and experimentation. We were not all that interested in future flagging and experimentation a few years ago because we thought future flagging was a bit of a commodity itself, and we thought that experimentation was more on the surface, more something that related to AB testing, button colors and things like that, which is -- it's interesting, but that was not our core business.

I think we've changed our minds quite a bit on the topic as we understood that this would be a key part of automatically shipping and iterating on software. So that it could really make use of the productivity gains you get from the -- on the coding side from the AI agent.

Another example is data observability. We thought also this was an interesting market, but this was a little ancillary to what we're doing. Now that this becomes one of the key limiting factors or the data quality and timeliness is one of the key limiting factors for building and deploying AI models is also something that is -- that has boiled up to the top of the list for us. There's a few other areas we're thinking about, but I'm not going to tell you about it.

Can I add some more comments that -- what I've learned about autonomy for Dev, Ops and Security is not a 0 to 1 game. This is very different from putting trucks on the highway without the driver is absolutely 0 to 1. There is nothing in between. What we're seeing is the adoption of this closing this autonomous DevOps loop is the evolutions of technology and human behavior. So we start partly with the investigation piece. Initially, a lot of the customers don't trust the result of the investigation, and they verify that. But now as they use it more, they start to gain trust. Same thing was pushing a code fix. Most of the people still are very comfortable absolutely requiring people in the loop but as more confidence get built and the technology to get more mature, we would expect customers to get more comfortable.