JFrog Ltd Aktie

My name is Koji Ikeda. I am one of the software analysts here at Bank of America. Welcome to day 3 of our 2026 Technology Conference. Here to kick off day 3, absolutely thrilled to be hosting a fireside chat with JFrog. We have Ed Grabscheid not working yet. Ed Grabscheid, CFO of JFrog; and Jeff Schreiner, Head of IR. So thanks so much for joining us.

Thank you for having us.

Thanks for having us.

I guess maybe just to kick it off, I always like to start with just a high-level overview of JFrog for the listeners in the room that are maybe new to JFrog's story. It's in the weeds of DevOps, but DevOps is a fantastic category. And so maybe just a high-level overview of what you guys do would be?

Yes, sure. Happy to do that. And it's great to see everybody. I see a lot of familiar faces, some new faces. So I'll give a very high-level overview of JFrog and kind of what we do. And you're right, it used to be in the weeds. I think there was a lot of people that misunderstood the story of JFrog and what we did and what is a binary. Binaries were kind of something that was a machine language that nobody really talked about, nobody understood. They understood source code that you write code, you write it in English, German, Spanish, but that converts into a machine language, which is known as a binary, and that's what JFrog does from the day 1, it always focused on binaries.

We have built a trusted secure and automated platform that delivers software through the software supply chain as a binary to distribution. And in the era of AI, as much -- many, many new binaries are being created, we have built a very scalable and holistic platform that allows for this tsunami of binaries to come through into production and through distribution.

So when you're out there talking with customers and in this category of binaries and the security of binaries and frankly, the whole supply chain of software, when you're winning deals out there, what are maybe the 1, 2 or 3 top priorities and differentiation that JFrog offers that they're telling you, this is the reason why we chose you guys.

Yes. So as I stated when I opened about JFrog, what are the critical kind of technical moats that we win? And one is the automation. So the automation process, it is a holistic universal solution, which means that more than 30 languages connect with JFrog. It's agnostic to any languages, any cloud provider, any foundational lab and AI model. So that's number one.

Number two is it is a control plane. So everything comes through JFrog. And you have full visibility, a holistic view of all of those applications that are coming into JFrog or into your organization and through JFrog. And so when you look at other solutions and when we go up against other maybe smaller competitors, the technical win here and the moat that really separates JFrog from those others is the holistic view and the scalability and the fact that it's not just a good enough solution in the era of AI, it is the best solution that supports all of the languages that are available to any of the agents.

One of the questions I often get from investors, and I try my best to answer it, but would love to hear how you guys would answer it is. Hyperscalers and other big platform vendors might have some sort of binary solution. If you quickly Google, artifact management, things come up. And so why are you winning a lot against some of these bigger platform vendors that offer a lot of solutions out there?

Yes. Why don't I start? Yes, I'll start, and then you can jump in, Jeff, with maybe a little bit more of a technical view on this. But it goes back again to the universal approach and the fact that we support many languages and the scalability. So maybe a year or 2 ago, you might have heard where the hyperscalers had a container registry offering and maybe smaller organizations were going to the container register offering from the hyperscalers.

And there was a competitive advantage from a price perspective and a good enough solution to support 1, 2, or 3 languages. Today, it's very different. If you're using agents, and we can say that 80% or so of developers today are using an agent to build code in some capacity, the agent itself is going to go out and pull what is the most efficient way to write the code. And that may not be with npm. It may not be with pip. It may not be with Conda Nadianne. Its own language that it creates on its own. And JFrog can support all of these languages. It's a scalability.

So as a new language comes into play with the development process, that language can be added to JFrog very quickly, and it doesn't break the model or the platform. And therefore, a good enough solution like a container registry coming from the hyperscalers is not going to be good enough in the era of AI. And this is why nowadays, JFrog has become mission-critical to all organizations.

We're focused on the enterprise, but we see an opportunity for that to expand to all organizations, especially as we move forward to an automated fully autonomous process in AI.

When we look at the infrastructure software category and frankly, the entire software universe, I think investors out there are trying to figure out AI winners, maybe AI neutral and AI, we don't know. And JFrog definitely, I think, falls in the AI beneficiary category facing a lot of tailwinds here or benefiting from a lot of tailwinds here. And so maybe help us explain or help me understand why JFrog is a beneficiary of AI. I know the simple, simple, simple thesis is AI equals more binaries, equal more JFrog. But why is that?

Well, it goes beyond that, to be honest with you, Koji. It's not just the ability to take the amount of new code that is converting to a production level binary into the organization and handle that level of the surge. And clearly, those numbers that you see in our cloud growth, 50% cloud growth now representing 51% of our total revenues, a great outcome, and this is strategically what we've been building towards is moving as many workloads from self-hosted to cloud, landing new customers in the cloud, building new capabilities in the cloud, and that's being represented in the numbers, but it's also the governance aspect.

And so where the AI winner, maybe they're looking at JFrog as an AI winner, it's around the governance. So once you allow the machines to be autonomous or working in collaboration with the machines, you have to have the governance. And this is where JFrog quickly is becoming recognized as a beneficiary to AI because the model itself is a control plane or the platform is a control plane, allowing the governance.

So when the machines start creating binaries, you have to have the governance layer on top of this. This is why it's critical with AppTrust, a product that we introduced during swampUp in September of last year, and it's starting to gain traction. Customers see the value of the governance as more and more code is being created by machines and not by humans that are overseeing the process, the governance layer will become critically important in the era of AI.

I think in the new world of AI and the current world of AI, everyone is trying to understand, number one, where technologies sit in the stack -- and then also where does the technology sit within the ecosystem. So I want to ask you the ecosystem question. And when you look back at the transcripts in the prepared remarks, there's a lot of commentary of how you guys are playing partnerships here and there. And so maybe help sum that up for us, where does JFrog sit in the ecosystem?

Well, I think in the ecosystem, I think you heard Ed speak about it earlier, Koji, and that's the holistic view that JFrog has. And I think maybe that's one thing that might be underappreciated is with the universality that we have with so many tools connecting or utilizing JFrog or metadata from the binaries, we have a whole picture view. A model may have the view of the model. I mean many models, if you try to move model to model, tell you, no, that model sucks, I'm better than that model work with me.

So they have a very centralized view of what they're trying to do as do others, whereas the holistic view of JFrog, i.e., the Switzerland of binary, so to speak, gives us a unique positioning in that ecosystem. And I think that's also a competitive moat for us in a way because why? Because why would a larger player want to allow that universality, right? A larger hyperscaler may want to enforce their tools, their cloud in that critical area where a lot of the enterprises need that universality so that they can choose the best solutions to plug in and work with JFrog Artifactory and with the software supply chain managed by JFrog.

Got it. When I look at the business model, I think one of the key growth drivers is cloud. So cloud just grew 50%. It grew 42% the quarter before, 50% the quarter before that, all great, right? But how do we -- what's the best way to understand the durability of cloud and maybe get a little bit more comfortable with 50% going to 42%, going back to 50%? And what does that look like? Or how should we think about that?

So I think you're alluding to the seasonality. And that's a question that's been asked quite a bit as we've been on the investor circuit. What is the seasonality going forward? What should we be thinking about for JFrog? And how does workloads now -- how does that play out with the rest of the year, especially in the era of AI. Previously, you had holidays to take into consideration. Q4 historically has always been a slower quarter. We always knew that Q3 was kind of a capture quarter in terms of commitments.

But everything has changed in this environment because machines are running in the background. It doesn't take a break. It doesn't go to the restroom. It doesn't go on a holiday. It's the million-dollar question, to be honest with you. We don't know what the seasonality is. And what we do know is we have a commitment, and we're different than a pure consumption play SaaS company.

We have a commitment, an annual commitment to JFrog. It's contractually obligated. It's a legal binding agreement, and we guide on that. So whatever we're guiding, we know it will not fall below that. That is the floor. We increased our guide heading into Q2. We raised it from a guide of 30% to 32% to 34% to 36%. And we also believe that we have a better confidence level of a sustained usage in the cloud above minimum commits today for a couple of reasons, and I'll explain.

Number one, we've had now 4 or 5 consecutive quarters where we've seen strong growth, even though it's not included in our guidance, we feel more confident about that. Number two is our customers are telling us, budgets have been loosened. The purse strings are loose, and we're not being pushed back to slow down the innovation that's coming from the Board being asked by the Board or by management. And we're not even being pushed back by the office of the CFO or procurement to go do a re-up on the agreement. We're going to spend over the minimum commitments for now.

Now once you get to an annual commitment at the end of the contract, you're going to have to renew. And we still see customers renewing at a higher annual commitment. It makes a lot of sense. We saw that. We captured some of that in Q1, and we hope to do that going forward. But at the end of the day, we start to see customers spending over minimum commitments, and they're continuing to spend over the minimum commitments until they have the clarity. And once they have the clarity, they may re-up to a higher annual contract.

Today, we feel like the seasonality behind with that is unknown. And the intent right now is to try and capture a higher annual commitment. So therefore, we have the visibility and we can guide clearly to the group here and the investors and the Street.

So not only do you have cloud, you have a very robust on-prem, self-managed solution too. And so maybe talk a little bit about why a customer would go cloud versus on-prem and vice versa?

Yes. I'll start, and then maybe you can talk about the technical side. Yes. So what -- there's 2 ways to look at it. There's organizations that first come to JFrog and many of those today as a new organization landing at JFrog will land in the cloud. They see economic benefit because they don't have to manage their network. They don't have to manage deployment across a global topology. Everybody is up to date from a version perspective. We also have customers today that are on self-hosted that are still migrating. Even though we talk about a slowdown in migration, there is still benefit for those customers to move workloads from self-hosted to cloud.

Now what's happening with the very large organizations that have millions and millions of artifacts that are 20,000 developer shops or so that had intention to migrate from self-hosted to cloud. These are very advanced developer organizations that are also using AI. They don't have the clarity yet on the cost structure, and those have been paused. It's not a matter of if, it's a matter of when, and they also see the value of moving to cloud and putting the workloads into the cloud. But those are being paused at the moment right now.

Yes. And I would just add, I think, Koji, the decision to be on cloud or self-hosted, that's a customer decision that we leave to the customer. And I think what you're going to see over time is there will be a cohort of the self-hosted group that will be in regulated industries or be somewhat forced by some type of regulated policy to have a self-hosted instance. And those will be the customers that will continue to use that.

However, we are starting to see things emerge with the use of hybrid in the era of AI because we all go through these periods of costs of cloud when new technologies or the use of technology and shifting from self-hosted to cloud earlier in the few years back that we had seen causes some type of realignment of the vision, but inevitably, cloud goes on. But I think right now in the era of AI, certainly, some are looking to say, is there a hybrid approach in which I may choose to put certain workloads into self-hosted and then move certain workloads into the cloud.

And as you utilize JFrog, you're not experiencing any difference in how that's interfaced. The UI is completely the same to you. And again, I think that's a differentiated opportunity versus not only the large players, but other smaller players beyond just scalability that each is kind of restricted to their area in their sandbox, one in self-hosted and one cloud native.

In the software landscape, pricing models has become a pretty topical conversation. you guys have been a very consistent pricing model. And so maybe spend a minute or 2 and remind us exactly how you price on cloud and on your self-hosted solution?

Again, another question that's asked quite a bit. I think there's concerns around where is pricing going to go in the era of AI, especially with the seats and the change in seats going from humans to agents and how do you price that? And the good news is JFrog has never been a seat-based pricing model. On the cloud side, we always monetize based off of data consumption and storage. And on the self-hosted, it's on number of servers. So we've avoided the seat discussion.

On security, we do monetize based on a number of contributing developers. That's a common currency, and we went to the market with our security solution based on that. We thought about maybe does it make sense to do something on scans or artifacts, but it was confusing for buyers. And as we were stepping into security, we wanted to make sure that there was a common currency for CISOs to understand the value of JFrog.

Now as you move forward and you think about tokens in this token environment and as tokens start to increase, there's a lot of discussion about now as more and more use or more and more development will go through tokens, what does that mean from pricing? We're not going to be the pioneers behind the pricing. We'll certainly follow the trends, but there could be some kind of bundle of pricing or a shift in the way that things are being priced, and we're keeping a close eye on that. We don't have any updates yet on JFrog. There's no intention to change anything yet, but we're keeping an eye on the market and see where it goes.

So you mentioned security there in your answer, and we haven't touched upon it. We've gone 18 minutes without talking about one of the most exciting growth drivers, I think, for JFrog. And so clearly, you guys have built an incredible franchise with your Artifactory product. I mean, I don't use best-in-class all that much in software, but we hear it all the time with your partners and customers out there. So congratulations on what you did with Artifactory. But let's talk about security. What is security to JFrog? What is security to your customers out there? And how are they thinking about it?

So security, there's 2 ways to look at it. There's a security, the basic security that's added to our subscription called Xray. You have Enterprise Plus, Enterprise X and Pro X. That is a Tier 0 scanning capability once you bring artifacts into your organization. We also have an add-on security. This is really what's driving the growth in the ASP, which is driving our net dollar retention rate, which is driving the expansion that we're seeing and very healthy KPI and metrics that we shared at the end of 2025. That is curation and advanced security.

Curation is the firewall that sits outside of the organization that you curate what you can bring into the organization. And then once you bring those packages and artifacts into the organization, you use advanced security. Advanced Security is a platform that replaces 7 disparate scanning tools into one platform, replacing those tools.

Now security has gained a lot of momentum over the past year and in particular, since Q3 of last year, when you started to see a lot of these open source vulnerabilities, NPM attacks, the Shai-Hulud attack that created a lot of awareness for curation and protecting your organization outside of the firewall or outside of the organization as a firewall. And we saw great momentum, especially in Q4, where we had a lot of security curation purchases that were done really within the quarter in a short period of time, and that momentum carried into Q1, and we're building a very strong pipeline.

It's no longer a situation where there's an incident creates awareness, there's maybe fear buying. This has become a pipeline builder. The awareness is clear. It's really a no-brainer to protect your organization, especially in the era of AI when packages are moving at the speed of light into the organization. This is the best way to control that. It's the best way to keep developers productive and curation is becoming a very good product for JFrog and a nice pipeline builder.

Can you break down exactly what curation does from a very simplistic view. We've done a fair amount of work on it. We hear good things, but always a good reminder of what exactly does it do?

Yes. Thanks, Koji. As I described it, we try to paint it for investors to understand is kind of the wall around the castle to keep out the barbarians at the gate, okay? And this has become something that was actually developed, and I want to highlight this, that what JFrog has been able to do well and continues to do well, and I think will benefit us as we continue to work with these frontier labs is create products that solve very important pain points for our customers.

So there was a large financial services firm who came to us at the time because at that point in time, a few years back, prior to curation, what an organization did, at least this organization did, allowed the developer to build, but you were not allowed to be -- have the package approved, and you had to hope that the package was approved by the time you built because they brought nothing in and you had to get that one package approved to be then built.

Others at that time were just bringing everything in and scanning with Xray to try to secure in that sense. And so this customer said, well, wouldn't it be great if we could have a centralized policy in which I could state and as I stated, curate, that's the name curation, the repositories that I, in fact, want to work with and want to allow my organization to use.

And so I will tell curation to point to repository A, B and C. And within that repository, it will be curating and managing the packages, and it will manage the packages within that repository that I want to use and keep an eye on it what is going on in this repository. Has anything been changing with the packages that I'm assigned to be in charge of. Thus, it creates that firewall, which in essence has become critical in the era of AI because now the hackers have AI and the patch and the remediation of all these CVEs we're seeing almost by the hour. Where are they happening? They're happening at the binary level.

And so when I now want to start moving towards more autonomous work as I start to shift at some point to allowing agent 101 to build and go into production. What I want to have is a sandbox that I've approved centrally to say, do as you will, be creative, build, build, build, baby, but do so with what tools, what packages I've allowed into the organization.

So in the world of AI, that's very critical to enterprises. And I think it's -- that's why it's becoming somewhat almost of a no-brainer solution to have because it sits outside your IDE, sits outside source model where the data is created. So that will eventually as well, another thing I want to add as I wrap up the answer as it relates to curation is that our CTO kind of snips at Ed and I when we're having conversations because he's like, you guys talk about curation a lot of security. I'm like, yes, well, Vdoo built it for security, it's governance and you wait. They'll learn that it's governance. And so it's going to be a critical aspect to the governance aspect that Ed talked about and tying into AI catalog.

So curation has become a very critical product. And the last thing I'll add there is that -- it's become so critical that there's been some Johnny-come-latelies that have tried to fill a void. And they tried to fill a void on the typical way that Ed had described earlier in his first answer in which I come with a good enough solution to tell you that you don't need everything JFrog offers and this is going to be just good for you.

Well, some people chose that route, and they've been hacked. And I'll say that curation has never been penetrated. So I think that there's a lot of opportunity out there for us still as it relates to curation.

Is there a benefit to all with what you learn within curation. Oftentimes with software, you hear, hey, we take all the data and what's happening, and we don't share everything with everybody, but the learnings that we do within certain products do help everybody. Does that have aspect?

I think it's different in the way maybe you're talking, right? We're obviously part of the community, and that's another thing we do that I think many don't want to do in terms of really working with 36-plus languages, that's all the repositories that come with that, supporting the community and doing community-led things that will always lead to monetization. That's some of the things we've done.

What you may be talking about or what I would probably answer the question with Koji to how you presented it is back to what I said with solving pain points is now being very tied into some of these AI frontier labs that you are all very familiar with and understanding that when they came to us, we have very, very smart people. But they had a light go off when they were shown what an AI company views JFrog and a system of record as in the world of AI. And now we are able to start to work with them and try to build to their pain points as AI scales and grows at their individual companies.

So the security story sounds great. And so what are you guys seeing that is giving you confidence in the durability of growth for security? Maybe things that you're seeing from the customers or usage or conversations, whatever it may be, that helps support this durability of security?

So as I mentioned in the beginning, we have 2 products. One right now is probably weighted a little bit heavier in terms of criticality, which is curation because it's -- there is no alternative solution at this point and organizations didn't have it unless they tried to build it on their own, which they weren't able to do. So there's an immediate caffeine high, so to speak, to bring curation into your organization, implement that and see value immediately. And as I mentioned previously, this is no longer an incident-driven purchase or a fear buy. This is becoming obviously critical every week, maybe even daily, you're starting to see these open source attacks and vulnerabilities that are happening.

So if you're an enterprise and you're using AI capabilities and you're concerned about security, you're obviously going to take curation. So there's durability there based off what we're seeing in the pipeline. The second piece is advanced security, which, again, is critical. It's critical to consolidate all of these point solutions that many of these are not publicly traded companies. They had a value proposition maybe 3 or 4 years ago. Their value has come down, consolidate it to one platform with the provider that is securing your asset, the most critical asset, the primary asset, the binary and managing the binary.

So it makes a lot of sense to consolidate those point solutions to one platform. So there's a lot of value in that as well, and we continue to see strong demand and interest in advanced security as well.

And I would say the advanced security, just to add quickly, Koji, it's going to have a much long tail type effect because you can think that some of our large deals that we signed a few years back, the largest deals in history, they'll be coming up over time for renewal. And let's see if in the first multiyear, 3-year agreements that they've signed, do they, in fact, replace all those 7 tools because that's the difference here. As Ed said, I'm putting as many contributing developers behind the firewall.

Now the JAS solution, advanced security is more for the good actors gone bad inside the castle, so to speak, right? And I may replace one tool in year 1 because I have to run in parallel for 9 to 12 months. And then obviously, I'm going to take a different step-up in seats by year 2 because I'm running one, and I may now be testing 1 or 2 to replace as well. Now each has a built-in step-up over this multiyear period, but I think the step-up is more weighted for advanced security as you build in the tool adoption over time. And so that's still very early innings in terms of the consolidation.

I got you. Running up on time here. So I do want to ask you the growth and profitability question. And one thing that I've always loved about JFrog is your responsible growth. You've never done a reduction in workforce, always been growing, always been innovating. And so you guys are really like facing a nice opportunity here. And that always bears the question, why not invest more for growth. And so how are you thinking about balancing that growth and profitability framework?

Yes, you know us well. You've been with us for a long time, and you know the story. We're a very efficiently run company. We're very disciplined in the way that we deploy our capital and where we hire and like you said, especially now and there are so many headlines with reduction in headcount. We knock on wood have never done that, and we don't intend to do that. We continue to hire at this -- in this environment because we want to capture the innovation and make sure that we're staying very relevant in the AI era.

In addition to that, we're also implementing a lot of AI internally. So we will have efficiency gain over time like many companies, we're continuing to educate ourselves and exploring, and I think efficiency will come in 2027 around AI, and we'll have to consider what that means from a hiring perspective. But for the time being, when there's an outperformance on the top line, and you understand our guidance philosophy, we're a little bit more conservative or responsible on the top line, less so on our operating expenses from a guide perspective.

So we would expect that outperformance, a portion of that to flow to the bottom line. We had a very strong quarter in Q1, over 21% operating margin, a nice growth in operating margin on a year-over-year basis, and we'll see how the rest of the year goes, but we'll maintain that efficient and disciplined approach.

Got it. We're out of time. Ed, Jeff, thanks so much for doing this. Appreciate it.

Thank you, guys.

All right. Good morning, everyone. My name is Brian Essex, JPMorgan's large-cap/mid-cap software analyst.

With me today, I'm very happy to have JFrog with their CFO, Ed Grabscheid; and Jeff Schreiner from IR at the end there to keep them honest.

What I'll do is -- I've got a number of questions for the company, which I'll run through. I'll leave maybe 10 minutes or so at the end of the session for a Q&A from the audience.

But Jeff, thank you so much for joining us.

Thank you for having us.

Thanks for having us, Brian.

Maybe to kick it off, probably the most popular question that I've gotten after your earnings, which -- amazing results, 50% cloud growth, which is fantastic. How do you guys think about the sustainability of cloud growth, particularly given that a lot of it is overage related, and it doesn't seem as though enterprises are, at least in 1Q, too focused on managing the utilization of that overage? But we'll love to understand how much visibility you have into that growth trajectory. And how do you think about things when companies start to manage those costs maybe a little bit more efficiently?

Well, it might be the most popular question being asked to you. It's certainly the most popular question being asked to us. It's really around the cloud growth. People are very excited about a 50% growth in the cloud. And even more importantly, we're now starting to be at the point where we're seeing more contribution in revenue -- total revenue coming from our cloud than our self-hosted. This is strategically aligned to what we're trying to do in moving as many customer decisions from self-hosted to the cloud or landing in the cloud. And we have proof points in that now as 51% of our revenue is coming from the cloud. So that's great news.

The second piece is, how sustainable is this? Is this going to continue? We don't know. We really truly don't know. We're happy with the way things are going. We're built for scale. So when you have this surge of usage and the number of binaries that are being created and coming into Artifactory, we can manage that. We can certainly capture that value, and we also guide on the minimum commit. So this way, we're protecting the shareholder. We're being very responsible in our guide.

We increased the floor of our guidance. We were at 30% to 32% on our cloud. We increased that to 33% to 35%. And if this continues at these levels of usage, you'll certainly see that, that benefit captured in our revenues, and we'll increase the guide as we start to capture the commitment. But more importantly is that, that is -- that benefit in what we're seeing, as you mentioned, right now, the appetite to spend and to adopt AI in organizations is continuing. And as long as that continues, you'll see that in JFrog revenues.

Great. And what is the feedback from customers as they get the bills for that overage? Are you seeing any pushback on pricing? And maybe as a part B of that question, when a customer is evaluating committing to a more significant amount of utilization versus maintaining the current plan where they're seeing these overages, what does the cost trade-off look like?

Yes. Yes. So you would think that as a customer starts to use over their minimum commitments, and not only are they using over their minimum commitments, there is a rack rate or a rate above their committed. They go back to a rate that's higher than the committed rate, and you would think that the customer would maybe be incentivized to move to a higher annual commitment. Right now, there's not a lot of clarity. So clarity is the key here. They don't know what the usage trends are going to be because AI and this agentic economy is creating a lot of uncertainty around the clarity. So there is no pushback at the moment.

They're willing to spend at higher rates. They're willing to pay the additional premiums, as they go over their minimum commitments. We worked very closely with the customers. It's not a matter of JFrog not trying to move the customer to a larger annual commitment. There's economic value. There's benefit for both sides. It's a commitment to JFrog. It's committed revenue. It shows up in our RPOs. And it's an economic benefit for the customer, as they get better unit economics. However, the customer is telling us, "We're okay to spend more". We're okay. We're not being told by our office of the CFO to stop spending at this point. The purse strings have loosened a bit, budgets are available, and they'll continue to spend, and we'll continue to capture that revenue.

We're confident that we can move that to a higher commit at some point. When that's going to happen, that's still uncertain. And that's, again, I'm going to go back to the guidance philosophy of only guiding on those commitments because tomorrow, that could stop, and we want to make sure that we're protecting ourselves with a responsible guidance philosophy.

And what is it -- I mean, based on conversations you might have with your customers, what is it that they evaluate? What does their purview look like when they're trying to figure out whether or not they commit to a higher level of spend? Is it migrating from an experimental phase to one where they have a greater degree of certainty in terms of how they're going to utilize agents or AI or utilize your platform across the SDLC? How are they looking at that, just taking a step back and trying to evaluate where enterprises are in the experimental versus production?

Yes. The keyword was experimental. We're still in this experimental phase. We would love to think that we're in this production phase. We're far from that. There's still a lot of experimentation going on. They started with the experimentation with the code assist tools, and you saw the increase in the amount of code that's being created. That now is cleaner code and going into a binary, but we're still very much at that experimental phase. We're certainly not at this autonomous building code moving into a binary into a production level quality binary. It's still a lot of experimentation. And I think this is where the decisions are being paused around taking a larger commitment or a multiyear commitment because of the uncertainty and clarity that they need once they get into production.

Got it. That's super helpful. I wanted to touch on Mythos because -- or Mythos, depending on where you're from. Just because it's had such a profound impact on the rest of our space, particularly on the security names that I cover. And one interesting conversation I had was with a large GSI on Friday. And he said, basically, since Mythos emerged, CIOs, CISOs have been freaking out and trying to figure out like how do they respond to this. And particularly given that when that came out, your stock took a pretty meaningful nosedive. Where do you feel as though initial -- I don't know, fears is the best word, but initial concerns were misplaced? And what have you seen across your platform since Mythos and GPT-5.5 kind of emerged as significant models?

Yes. Well, we described February 20 as Anthropic Day inside JFrog. I think it was a day in which investors were coming to some sort of conclusion, maybe more simplistic conclusion, and I understand that, that the models were eventually going to start moving in and doing security, which is a key growth driver for JFrog. And thus, now the terminal value of JFrog has somewhat been impacted in a negative way. I hope we've done a good job since that date in explaining to people that in the world we're moving towards with AI and its intelligence getting better, that's going to be the creator, i.e., a lawyer in a courtroom. But no courtroom is going to be sufficient if it's just the lawyers trying the case. You're going to have to have a judge or a governor. And that's where JFrog comes in and where we think we have the right to win today.

And what I tell people is that even if we don't win, someone else will be that governor. Even when we get to the point that machines -- I mean, let's be realistic, they're only using human language because they have to talk to us. When they don't need us anymore, they quite frankly, may just go to what they need, which is producing the binary, right? And even in that instance, when they're producing binaries to try to go straight to production, you will still need to separate the creators and the governors. And I think that was the fundamental misunderstanding.

I also think, as it relates to the security aspect you brought up, is that there's maybe a misunderstanding that if they do some code scanning on their end, and yes, they're picking up vulnerabilities and finding things, that's good for the community. They're not yet remediating a lot of those. I think you understand that with some of your companies that you cover. But I think that with the security aspect, I think that the better code that is created is going to likely end up into a production level code that can lead to a production binary.

And so I just think that there was some misunderstanding of that and what they do. And I think the word vulnerability, whenever it gets published, somehow is tied to JFrog. And if somebody says vulnerability, then that means JFrog is at risk. And I don't see that as the case. And I think that I try to remind investors as well that remember, curation is outside the IDE. So it's security that's happening before a model has even been engaged.

I'm going to add one thing. Jeff is exactly right on everything that he said. But the other thing to remember, JFrog is a system of record. So you have these foundational labs and these large models that are coming in. It is not a system of record. You need a system of record, what Jeff alluded to, and this is where JFrog really separates itself in a control plane because a model sees one aspect, JFrog is a control plane. It sees a holistic 360 view of every open source repository that comes into your organization, and this is really where the moat is being built. And this is a misunderstanding.

You saw an aspect that the model was creating, I think, people overreacted clearly, and there was an over rotation, and it impacted our share price, but there is also now proof points that the critical nature of JFrog and the need for JFrog in the enterprises and to have a system of record and control plane is critically important.

That's helpful context. And any visibility from your customers? I think you guys have commented that the volume of binaries driven from LLMs is substantially larger than your typical application. Any visibility in terms of the magnitude that maybe your customers might have visibility? I mean, you may not have it. I don't know if you do. You may not have visibility, but do they have visibility? And can you just help us gauge the scope of expansion due to the usage of LLM?

Yes. Yes. So a couple of things. First off, we talk about 3 of the 5 foundational AI companies that we currently have as customers. None of those customers today are a cloud customer. They're self-hosted. So we're not necessarily seeing the cloud growth and the overusage in cloud coming from those foundational lab companies. That's number one.

Number two is what we're seeing in terms of the trend and the over usage, which could be driven by AI workloads, it's a combination of conventional package type and certainly AI workloads that are increasing across all package types. It's diversified across all of our customers' portfolio. So we have several thousand customers that are on cloud. All industries, all geographies, we saw an increase during Q1. So it's not concentrated by one customer in one enterprise, one specific group of customers, it was very diversified, which is good news. It's not concentrated. It's happening across the board. And again, we continue to capture that value in our cloud and the scalability. As more and more binaries are coming into the organization, we're able to capture that and turn that into meaningful revenue for shareholders.

Great. And I wanted to ask, I mean we're seeing some dynamics in -- across other stocks that cover meaningful -- notably identity where you have nonhuman identities just proliferating throughout the environment. And some of these identities are more sophisticated and some are less sophisticated than human identities depending on what they might have access to and how they're provisioned and governed. Do you have a similar dynamic on your platform with the introduction of LLMs and new features and functionality like MCP server? How are customers? And how are you thinking about pricing with the adoption of those new platforms relative to how things have been priced before?

I'll start with the pricing piece, and if you want to talk maybe a little bit about where the market is going with MCP and skills. From a pricing perspective, we recognize also that nonhumans are now starting to make their way into the organization. We monetize in the cloud on data consumption and usage. And for security, we monetize based on number of developer -- contributing developers. That could be a machine in the future. We're not at the stage yet, and we certainly don't want to be pioneers in the pricing world, but we're evaluating that. We realize that the gorillas, the big companies, that the hyperscalers and infrastructure companies, they recognize they've historically priced off of seats, and they recognize that there's an opportunity to price off of maybe some type of hybrid model of seats and data consumption. We're following those trends very closely, and we'll see how that transpires, and we'll adapt. But today, we're continuing with the data consumption in the cloud seats for security. It's a common currency. And we'll see where the market evolves, and we'll change those practices as they evolve.

Yes. And Brian, as it relates to -- I think you were talking about MCP, and I'll add skills as well. I think this adds to the scalability, the word you're going to hear me use a lot during this quarter and going forward. I think scalability is a twofold word for us. It's a moat that we have in how we do things, but it's also scalability in this aspect that we added new opportunities in Artifactory with new technology being created.

The creation of MCP, which is a connection outside of your organization, to use tools within your model, like Atlassian's Jira, like JFrog's Artifactory, other tools, those are exposures to your organization. And as you said, it could be spoofed. So those are things that you're going to want to manage and have a system of record for to say, when was this last accessed? Was it accessed properly? Where did it access? Was it doing something different?

And I think as it relates to skills, I think the unique aspect there is -- you heard Shlomi talk about that was built in 2 weeks ahead of GTC. Now, you can think about 10 million, 20 million, I don't know, maybe I'm being too low, of the amount of skills that will be built into agents inside an enterprise. And again, is that agent going to hallucinate? Is that agent doing what it was supposed to do? Is it acting differently? Those are things that will now likely be managed in Artifactory in our new scalable revenue opportunities that have been built and added on in the world of AI.

How we'll monetize that? I think it's part of a discussion of the AI governance bundle will come up with, but those could be things within AI catalog that you say, "I'll take AI catalog, and I'd also like skills registry and MCP registry as add-on features". And just to let you guys know, our technical team sees skills as the next real vulnerability injection point as risk to the enterprise.

And how are most of your customers managing access from an identity perspective to make sure that the right people, the right agents are accessing your platform? Are you just partnering with some of the identity vendors? Or are you assuming some of that security?

We're assuming not the identity side of probably matching the identity. We're assuming more the management of the MCP since it's built as kind of an open source architecture anyway. So it can be easily pulled into Artifactory and registered so that you know what it's doing. And you can be working in conjunction with like, let's say, an ID type security company.

Our security more is what's being moved in and out of the organization? Was it allowed? Was there something accessed that should not have been accessed? Those types of things are going to be tracked once you're in, right? And the idea, I think, was going to be to the spoofing of creating some type of spoof saying, I'm Jira, and I'm not really the Jira connection, and then, that becomes an exposure to the enterprise.

Got it. And maybe on the point that you just put -- you recently mentioned on scalability, maybe help -- for those that maybe aren't too familiar with the story, help us understand where the scalability enables you to efficiently grow compared to maybe competitors that are trying to edge in our space? I mean, you've got Sonatype that focuses on open source binaries. You have others like GitLab that have tried to elbow their way into your space. But you're able to scale much more efficiently, particularly into incremental languages. So maybe that as well as maybe part B, the difference between code versus binary repositories and...

Well, I'll start on the first. I think we talked about this in the follow-up from earnings, Brian, and I'll give you the same type of example I gave then. Scalability is becoming very critical. I think in the past, in the last 5, 8 years, you could be a company that said, "Well, I program in the 3 largest open source packages, so I don't really need all of the platform offerings from JFrog in the 35 languages it supports". In AI, that's not going to work anymore. That's a very much changing dynamic that those alternatives that were good enough will not cover you since AI is using all of the languages and your exposure will be to all of the languages.

As it relates to the scalability aspect, when I joined JFrog, I wanted to know, hey, guys, what's our difference? What's our moat here, right? Tell it to a dummy, a guy that's not maybe very technical. And how would I go out and explain that? And the way they talk to me about it is to think about it as a frequency. And so here's a frequency, and we've got, let's say, 10 languages being supported, okay? I add a new language, it's going to drop, and the frequency is out of balance. Now, JFrog can bring the frequency back to balance immediately. Others need time to bring that up.

And then, if you add another language, you fall back further, and you're constantly playing catch-up to get that additional language built in to keep that frequency of speed and security in terms of how you're developing. And that's really the differential for JFrog. And I think that where that can also play out into the future, as we talked about, I think that as the machines start talking to each other and they're doing more, and even with maybe human developer assistance, we could see new packages that become important being created from AI and you're going to have to scale and be able to support those to scale with AI.

And what is it about your platform that enables you to bring your frequency back into sync from maybe a technical perspective that is hard for others to replicate?

There's a checksum feature that goes beyond my software for dummies kind of knowledge. There are some aspects built in that we built inherently from the ground up. The things that my guys would tell me is that the competitor you referenced, Sonatype, they were a great competitor and have always still been someone that maybe you go to in that case I gave you, where maybe I started Sonatype because it was -- I didn't need everything JFrog had. But then, when I want to go to the cloud, then I have to go to JFrog.

Well, Sonatype's problem was that they got stuck at 15 languages, and then, the inherent scalability from the way they built the platform, that's where they're stuck. And that you would have seen them bought, I think, inherently come after JFrog by someone else if they didn't have this problem. And my guys -- and nothing against any company out there trying to do what they do, but that you would have to basically rip it to the ground and rebuild it up to break that scalability problem.

And where does this come in, in importance of the world we're in today, Brian, it comes in the importance that one of the AI lab companies who came to JFrog and is one of the leaders and has possibly a moonshot with us, unbeknown to us, had tried to build a JFrog replication, failed because of scalability and then sought out JFrog. So I think that's very much the differentiator for us.

As it relates to source code and binaries, just quickly to wrap up on that, really source code is where you're building up and you're creating the language of what you would like to go into production, but you have to hit compile. And when you hit compile, 10% to 15% of that source code is combined with the open source software, and it makes a software package that you're going to deploy, i.e., when we do our updates on our iPhone, that update is a binary. So there's a difference. And also, source code at least today is written very much in language that you and I can decipher. Binaries are machine language, 0s and 1s, and there's a very much different skill set needed to manage each of those, and that's why they've been inherently different.

Got it. Maybe on that other point, too, is I think Elon Musk recently tweeted that you'll see models writing directly to binaries. I don't think you guys have seen that yet. But if something like that were to become a thing, how might that change your model if we saw a tweet or something that all of a sudden this was happening?

You're the CFO. I don't think we've tried to model that in yet.

No, we have not modeled that in, and there's no benefit there. The only benefit that you would see is if you're starting to create code, or I should say, binaries that are going into production, that's going to be a benefit to JFrog. So skip the process of writing the code and converting that to a binary. If you're writing directly as a production-ready binary, that's going to be great news for JFrog. And we're going to be prepared to capture that in terms of revenue and the scale that, that may provide in the future.

Because the production binary, Brian, is always more valuable than one that's spun up and busted because I have to do updates to that binary in the field. I have to secure that binary. I have to have the metadata of that binary, something that goes sideways. So to Ed's point, you would probably see some additional value creation for JFrog because I'm going to start accelerating the number of production binaries that are going to go.

How would you even govern that? Like how do you review that to make sure that it is what you think it is? Is that something where you would step in on the maybe the security side or the governance side?

I think governance is the word you just used. And I think that's the next thing that you should be looking at in the world of AI for a company like JFrog that as we move to more autonomous world, governance is going to be critical, whether it be the governance of gating, all the steps were, in fact, done and are recorded, whether it's the governance of understanding things like AI packages and what packages were used in the last 5 builds, like the Wikipedia of binaries through AI catalog. I think governance is the next thing. We've had cloud -- we've got cloud and security today as growth drivers. And I think as we move to a more autonomous world, your governance is going to be critical.

Got it. Maybe on the security side. Obviously, you guys saw there's another Mini Shai-Hulud yesterday. And which just seems like every day, you open up Bloomberg or you open up your inbox and you hear about another supply chain attack, how does that manifest itself? Is this just a building rate of demand and urgency on behalf of your customers? Or do you get a bump from an event like that if it's a major one?

Well, I think the bump happened in Q3, the awareness in the first Shai-Hulud event, and then, you had events weekly thereafter. So the awareness is there, and the decisions around bringing curation, which is critically important, a firewall around the organization, has become very apparent. I can only comment on what we see in the pipeline, and it's a very strong pipeline regarding curation. So I think the awareness is there at the enterprise, and we're actively working to convert those pipeline opportunities into actual, durable revenue growth for JFrog. We've done a good job so far.

Those decisions are being made at a very quick pace compared to, let's say, advanced security, where you're replacing disparate point solutions, curation. There's no competitive alternative today that you're replacing. You're bringing that into your organization. It's a caffeine-high. It's immediate. It happens, you adopt and you put it into the organization, we see benefit in security revenues. We see benefit in the cloud because that's deployed in the cloud.

And right now, it's been a nice growth driver, and that momentum continues because of these events, unlike, let's say, SolarWinds or a Log4j that happened once every other year, exactly you mentioned, you open the laptop and every time you open the laptop, you hear about a malicious attack in an open source and at the binary level, and so the awareness is there.

Got it. Super helpful. I'll have 1 more, and then, I'll open it up to the audience for questions. But I wanted to ask you on the financial side, Ed. Amazing quarter, again, great top line growth that kind of flowed through to better profitability. Relative to kind of where you guided, which would assume, I guess, imply a little bit lower margins. How do you think about what you let flow through? And how we think about the amount of natural operating leverage in the business as we kind of progress through the year?

Yes, it's a very good question. First off, when you look at the practices of JFrog and the discipline that we have around our spend and the balance of growth versus profitability, we've done an exceptional job of continuing to expand and having durable, meaningful expansion in our operating margin. Now, the guidance philosophy, which I've talked about a few times, is obviously a responsible top line guidance philosophy. And what we saw during Q1 was the beat, 100% of that, flowed to the bottom line. Will that continue at that same level throughout this year? That's uncertain. But I'm less conservative on my operating expenses. I can't be. I already know what's committed on the operating expenses.

So as we have, and hope to have, outperformance in future quarters, I would expect a portion of that would flow to the bottom line. And that's something that we've done. We did that effectively during 2025. I would think we would continue to do that. The focus is really around what you mentioned is the balance, balancing that spend and making sure that we're capturing it into growth opportunities and that we continue to manage the operating expenses effectively and so that when we do have an outperformance that would flow.

Got it. Super helpful. With that, I wanted to check to see if there are any questions from the audience. If you can maybe wait for the mic right behind you. Perfect.

Yes. Just on that same topic, given the magnitude of the upside driven by overage at rack rate in cloud, did you guys quantify how much of the upside was from the overage?

We didn't quantify it, but I think it's pretty clear that when you have a $7 million beat, and you're carrying over $4.5 million to the full-year guide, you can assume that the difference is going to be your usage over a minimum committed, at least in the first quarter.

Any others from the audience? All right. I have more. I wanted to ask on the sales and marketing side and your go-to-market motion. How do you think about -- given where the company is today from a growth trajectory, how are the hiring trends? And what is the outlook for ramping reps and scaling reps to address the growing market that you have?

Yes. Specifically, as it relates to our sales organization, we're built off of a capacity model. Now, one of the things that we've done really effectively over the last 3 years or so is shifting the go-to-market practices to the enterprise, and we did a significant investment in that arena around bringing strategic reps going after top quality salespeople from large enterprise sales organizations and bringing those to JFrog. And you see that in the results of our $1 million customers, our $100,000 customer and the growth that we've seen there. And we've been very pleased with that.

We made significant investments as well during 2025, and we're starting to see the fruits of that labor. The usage over the minimum commitments is driving the outperformance, a lot of that in Q1, but we feel very comfortable right now that the sales organization has built the capacity to win in 2026. But, of course, if we continue to have an outperformance, we're going to reinvest that, not only in sales, but also in R&D to ensure that we're capturing future value from AI and making sure that we're ahead of the game as it relates to development of next-generation products and technology.

And how are the incentives for the sales organization, particularly given the fact that maybe more beneficial to commit to a higher level of spend, but your customers may not want to do that right away. So how do you manage the conflict between those 2 dynamics?

Yes. That's actually a question that's been asked quite a bit in Q1 because of the outperformance in the revenue in Q1 and what are we doing to incentivize. I think it's become abundantly clear that its usage over minimum commitment. I'll let everybody know, and we'll put it on record, that our sales organization is compensated on commitment, not on usage over minimum commitments. So they're incentivized to get the customer to commit to a higher annual commitment. They're not benefiting from the usage over a minimum commitment in terms of their commissions and what's going into their pocket. So there's a combination of the customer spending more. That's good news for JFrog. It's reflected in the revenue. It's not reflected in our RPOs, and it's certainly not reflected yet in the commission for our sales team, and they're actively working to move those customers to a higher annual commitment.

Got it. Maybe I want to touch as well on emerging products. Obviously, at swampUP, your user conference, you had a number of different releases, a handful of products emerging then. And I think a lot of people have been focused on security, curation, maybe haven't heard a whole lot about governance, but which get you more excited? And how do you think about the way you might disclose the traction of these emerging products on the platform going forward?

Yes. I'll start from the finance side, and Jeff, you can maybe finish with the technical side. But from a finance perspective, it's AppTrust and AI Catalog, were the 2 products that we released during swampUP. We're still in the very early stages of that. We hope to have maybe an update in Q2 with some customers that are considering some of those products. But we released them in September, Q3 of last year. We started to build the pipeline. We're seeing some nice traction in the pipeline, and as I mentioned, maybe an opportunity to convert that. It's certainly not considered right now as a growth driver in 2026. We see maybe in the future in 2027, and we'll update as part of our guidance. But governance is certainly an area of opportunity for JFrog, as Jeff mentioned, that's really where the platform is going. But today, it's not part of the guidance. So anything that we do in those products would be additive.

Yes. Technically, I don't think there's -- we've talked about some of the governance aspects and what they do on a technical basis. I think the key here to understand is that governance becomes much more critical to the enterprise as we move towards autonomous and more agents working autonomously. And I would just point out, I think a unique thing to think about for you guys is we talk to you guys about curation as a security product created by the Vdoo team that we acquired back in '21. Our internal team looks at curation as governance, and we'll see that as a core aspect of governance in the governance portion of the platform that we'll be rolling out in the coming years.

Helpful. Maybe I want to touch on real quick capital allocation. You guys had a pretty well-timed buyback that you announced. What's the structure of that? Are you guys on a grid? Or is this just opportunistic to be used at some future point? Have you dipped into it? How meaningful is that announcement?

Well, it was meaningful because we had an announcement that happened on February 20 that created a reaction in the markets, and it created a reaction within JFrog. And within 6 days, we had an authorization from the Board to allocate $300 million of our capital to buy back shares. We are on a grid, obviously, at those levels that after February 20, that gave us an opportunity to build that grid, and we'll see what happens with the buyback, and we'll report certainly in Q2 where we are with that buyback.

Now, as we think about capital allocation, and you mentioned maybe, what are we thinking about with free cash flow and the amount of cash that we generate, we're very fortunate because of our disciplined approach and how we manage expenses and driving free cash flow and the operations around collections, et cetera, that drive that. That gives us the flexibility to invest in the next generation of products and certainly with governance. There may be an opportunity to do some kind of tuck-in. And so -- right now, there's nothing that we're actively going after, but the nice thing is that we have the flexibility to do that because of the practices and the discipline that we have as a company.

Got it. Maybe one last quick one because we have like a minute left. But I just wanted to ask about the large customer adds in the quarter, 57 net new in the $100,000-plus category. What was the driver of that? And how sustainable is that traction?

Yes. A lot of that has to do with the addition of security and the add-ons. So if you take an enterprise customer, and they're adding security on that, enterprise at the lowest level is below that $100,000, and you add security and that obviously gets you above the $100,000 mark. So security was a brilliant move on the part of the management team when they did the acquisition of Vdoo many, many years ago, but taking that to market and seeing the criticality of it, that's what's driving the ASPs and why we're starting to see the traction, not only in the $100,000, but in the $1 million cohort as well.

And I would just add quickly, I know we're at time, Brian. New -- net new land, that ASP is going much higher than what it was 4.5 years ago when I joined the company. What a cloud win was then for a land versus what it is now has materially changed. And part of that is people are landing on the platform because we are becoming much more strategic of importance to the enterprise, landing with the platform and adding security on the initial land as opposed to 4 years ago, the kind of Artifactory product motion.

Super helpful. And with that, I think we're out of time. So Ed, Jeff, thank you very much for joining us.

Thank you.

Thank you, Brian.

Thank you for all of you as well.

Ladies and gentlemen, thank you for joining us, and welcome to the JFrog First Quarter 2026 Financial Results Earnings Call.

[Operator Instructions]

I will now hand the conference over to Jeffrey Schreiner, Head of Investor Relations. Jeffrey, please go ahead.

Thank you, Nicole. Good afternoon, and thank you for joining us as we review JFrog's first quarter 2026 financial results, which were announced following the market close today via press release.

Leading the call today will be JFrog's CEO and Co-Founder, Shlomi Ben Haim; and Ed Grabscheid, JFrog's CFO.

During this call, we may make statements related to our business that are forward-looking under federal securities laws and are made pursuant to the safe harbor provisions of the Private Securities Litigation Reform Act of 1995, including statements related to our future financial performance and including our outlook for the second quarter and full year of 2026. 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. You are cautioned not to place undue reliance on these forward-looking statements, which reflect our views only as of today and not as of any subsequent date.

Please keep in mind that we are not obligating ourselves to revise or publicly release the results of any revision to these forward-looking statements in light of new information or future events. These statements are subject to a variety of risks and uncertainties that could cause actual results to differ materially from expectations. For a discussion of 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, which is available on the Investor Relations section of our website and the earnings press release issued earlier today. Additional information will be made available in our Form 10-Q for the quarter ended March 31, 2026, and other filings and reports that we may file from time to time with the SEC.

Additionally, non-GAAP financial measures will be discussed on this conference call. These non-GAAP financial measures, which are used as a measure of JFrog's performance, should be considered in addition to, not as a substitute for or in isolation from GAAP measures. Please refer to the tables in our earnings release for a reconciliation of those measures to their most directly comparable GAAP financial measures. A replay of this call will be available on the JFrog Investor Relations website for a limited time.

With that, I'd like to turn the call over to JFrog's CEO, Shlomi Ben Haim. Shlomi?

Thank you, Jeff. Good afternoon, and thank you all for joining the call. We entered 2026 strong. Our first quarter performance reflects both the clarity of our strategy and the discipline in execution. Our continued focus on powering the world software through JFrog Artifactory as a system of record for trusted binaries, software packages and AI artifacts is resonating deeply with market demand. We are seeing growing adoption among the world's leading organizations and AI labs, which are choosing JFrog as they transform to adopt modern software supply chain practices.

Across industries, geographies and deployment environments, whether cloud or on-prem, our customers are partnering with JFrog as their foundational platform while they navigate a complex transition of adding AI technologies and tools to their software supply chain. They tell us they are prioritizing AI adoption while simultaneously maintaining legacy pipelines and open source packages, all as they demand stronger security, governance and fast release cycles. We are working closely with our customers, the broader developer community and AI-native companies to support them through this period of change. Our Q1 results reflect this momentum with AI redefining the software supply chain and powering our continued expansion.

In the first quarter, JFrog delivered total revenue of $154 million, representing 26% year-over-year growth. Cloud revenue grew 50% year-over-year, underscoring the accelerating shift towards our cloud-first platform. This performance was driven by continued strength across our core growth vectors, increasing consumption of our cloud services, rising demand for our software supply chain security solutions, higher ASP on new customer acquisitions and robust expansion within our existing customer base.

We also saw continued momentum at the high end of our customer portfolio. The number of customers with annual spend exceeding $1 million grew to 80, up from 54 a year ago, representing 48% year-over-year growth. Customers spending more than $100,000 annually increased to 1,225 compared to 1,051 in the prior year, representing 17% year-over-year growth. These results reflect our alignment with the evolving needs of modern enterprises. Developers and increasingly AI agents are producing software at scale and speed. This surge in binaries fueled by AI is driving the need for a single trusted system of record to manage, secure, and govern these assets across the entire supply chain.

On today's call, I will walk you through the quarter in detail, and Ed will follow with our updated outlook and additional financial insights.

Now I will highlight the key drivers behind our performance this quarter. First, continued cloud growth, driven by increasing consumption and rising demand for a true system of record as a service, delivering scale and universality. Second, the sustained momentum in our security business as customers prioritize end-to-end protection and governance amid rising software supply chain attacks. And finally, I will highlight our ongoing innovation that leads to solid adoption of our platform and Enterprise Plus subscription growth.

Let me start with our cloud business. As mentioned earlier, cloud revenue in Q1 grew 50% year-over-year, an exceptional result that reflects not one single driver, but a broader trend we have been observing over the past several quarters. As AI makes human to technology interaction nearly costless and source itself increasingly commoditized, binaries become king. Organizations are actively encouraging developers to utilize AI coding agents as well as explore agentic capabilities, causing software output to accelerate, resulting in more compiled codes, a true AI-fueled tsunami of binaries.

Observing our customers' consumption trends, we noticed that this growth is not tied to one package type or a specific AI native workload. It is not a spike in usage or a onetime increase in open source caching. It is the result of a fundamental shift in how software is being generated, delivered and consumed across the software supply chain. We are seeing an acceleration in the volume of compiled software flowing through the JFrog platform. This trend, which began taking shape in 2025 is driven by 2 major forces.

First, developers are being supercharged by AI coding agents. Simply put, the world is creating more software packages. In this AI mass adoption reality, we see organizations willing to accept budget overruns until they gain better clarity on long-term usage requirements and prior to increases in annual commitment. Second, as AI drives more software creation, it is also accelerating the flow of all open source components. Open source consumption by developers and AI agents is rising across nearly every software package we support. And as the ultimate Switzerland of binaries, JFrog sits at the center of this growth. Whether through on-demand increased usage momentum or annual commitments, we believe JFrog Cloud is positioned to benefit from these trends.

Now to the continued momentum we are seeing in security. As we mentioned in previous calls, modern software supply chain security is moving beyond traditional DevSecOps and fragmented scanners. AI coding agents are increasingly securing, scanning and even fixing code rapidly at scale. And while still evolving, we see agents replacing human skills in code protection. We believe a trusted software supply chain requires a single authoritative system of record for all binaries and AI artifacts. Building on this foundation, we deliver protection and governance beyond traditional scanning, analyzing, tracking and proactively blocking risk at the point of entry or before distribution to production.

As AI adoption accelerates in binary scale, the threat landscape is becoming more complex. Software supply chain attacks are rising, increasingly targeting open source creators and package maintainers. This dynamic drives the growing demand for a trusted control layer and stronger DevOps practices. In Q1, we again demonstrated that customers subscribed to JFrog Curation were effectively protected from recent software supply chain attacks. Curation serves as a critical control point at the gate, enforcing policies that ensure only trusted packages enter the system, keeping Artifactory clean. Once artifacts are sold, JFrog Xray and JFrog Advanced Security continuously secure and govern the binary flow, providing ongoing visibility and protection.

In addition, as advanced AI models like OpenAI, GPT, cyber and Anthropic cloud become increasingly embedded in development workflows, we believe modern software supply chain security and governance are defined by 4 core pillars. First, a centralized system of record, a single source of truth across multi-agent environments; second, universal governance, consistent visibility and enforcement across all types of artifacts, whether consumed or generated. Third, predictable and deterministic protection, continuous policy-driven guardrails that prevent malicious or vulnerable components from progressing. And finally, comprehensive coverage, securing both newly generated assets and the extensive base of existing mission-critical legacy binaries.

Our customers tell us they are accelerating software development and generating more binaries through the JFrog platform. As AI adoptions expand, JFrog provides a unified system of record to secure, govern and manage AI-generated open source or legacy binaries in one place. Our customers' adoption, Q1 results, sales pipeline and future road map innovation are aligned with these observations. Looking ahead, we expect security to remain a key growth driver for JFrog. This set the stage for an update on the innovation we introduced at our annual LEAP conference in New York this past March. LEAP is JFrog's top customers gathered by region scheduled globally during H1 every year. At LEAP New York, we demonstrated GA-ready solutions to concrete customers' need for a trusted infrastructure layer for software supply chain management in the AI era.

We introduced the JFrog MCP Registry, the first enterprise-grade registry for MCP servers, extending our platform to support the growing AI ecosystem. As MCP adoption expands, customers need a centralized trusted way to manage, secure, and govern these new assets, which logically sits in Artifactory as a system of record. MCP is rapidly adopted next to agent skills based on AI ecosystem demands. In Q1, we expanded our platform for AI-driven development with the introduction of the JFrog Skills Registry, providing a centralized way to manage and govern reusable AI capabilities.

In collaboration with NVIDIA, we announced the Skills Registry at GTC, enabling the governance and trust layer enterprises need to run agentic workflows securely and at scale. We further announced that JFrog Artifactory will serve as a registry for AI models and agent skills within NVIDIA AI-Q Blueprint, part of the NVIDIA Agent toolkit. The Vice President of Enterprise Partnerships at NVIDIA, Pat Lee noted, "security and governance are key to deploying AI agents in the enterprise. JFrog's Agent Skills Registry for NVIDIA NemoClaw supports security and control for deploying long-running agents to help scale enterprise productivity with powerful new AI tools. "

JFrog unifies all artifact types, binaries, models, skills and MCP servers into a single platform governed by one framework, one set of policy and complete visibility and traceability in one place. These innovations, combined with a growing ecosystem of strategic partnerships are driving increased adoption across the enterprise, amplifying the value of our enterprise class subscriptions and accelerating its expansions within organizations.

With that, I will hand it over to Ed for a detailed review of our Q1 financials and our updated outlook for Q2 and the full year 2026. Ed?

Thank you, Shlomi, and good afternoon, everyone. We are pleased by the results of our first quarter, which exceeded the top end of our guidance range on every metric. It was a strong start to the year, highlighting our consistent strategic execution and ongoing operational discipline.

During the first quarter, total revenues equaled $154 million, up 26% year-over-year. These results demonstrate the continued execution of our go-to-market strategy, fueled by our cloud revenues, ongoing demand for our security core products and growth in our Enterprise Plus subscription. Our first quarter cloud revenues grew to $78.9 million, up 50% year-over-year, now representing 51% of total revenues versus 43% in the prior year. Our outperformance in the cloud was driven by robust usage across our customer portfolio, which exceeded contractual minimum commitments. We strategically work towards converting this usage into higher annual commitments.

During the first quarter, our self-managed or on-prem revenues were $75.1 million, up 8% year-over-year. We continue to proactively engage our on-prem customers to migrate DevSecOps workloads to our cloud or explore solutions better aligned with their specific use cases, including hybrid and fit-for-purpose deployments. In Q1, 58% of total revenues came from Enterprise Plus subscriptions, up from 55% in the prior year. Driven by the ongoing execution of our enterprise go-to-market strategy and broader customer adoption of the JFrog platform, revenue contribution from Enterprise Plus subscriptions grew 33% year-over-year in Q1 2026.

Net dollar retention for the 4 trailing quarters was 120%, representing a year-over-year increase of 4 percentage points and 1 percentage point improvement sequentially. These results highlight the continued adoption of our security core products, increased cloud usage across a broad set of conventional software packages and AI workloads and conversion of customers with usage over minimum commitments into higher annual contracts. We continue to demonstrate that our customers view JFrog as a mission-critical system of record to their software supply chain with gross retention that equaled 97% as of the first quarter of 2026.

Now I'll review the income statement in more detail. Gross profit in the quarter was $129 million, representing a gross margin of 83.8% versus 82.5% in the year ago period. We remain focused on cloud hosting cost optimization as we anticipate a larger share of our revenues being generated from the cloud. Given our expected increase in cloud revenue contribution to total revenue, we reiterate our annual gross margins to be in the range of 82% to 83% in 2026. Operating expenses in the first quarter were $96 million, equaling 62% of revenues. This compares to $79.7 million or 65% of revenues in the year ago period. Our operating profit in Q1 was $32.9 million or an operating margin of 21.4% compared to 17.4% operating margin in the first quarter of 2025.

The continued balance between strategic investments and operational efficiency demonstrates our commitment to profitable growth. Cash flow from operations equaled $38.4 million in the first quarter. After taking into consideration CapEx requirements, our free cash flow reached $37.3 million or 24.2% margin compared to $28.1 million or 23% margin in the year ago period.

Now turning to the balance sheet. We ended the first quarter with $741.2 million in cash and short-term investments compared to $704.4 million at the end of 2025. Given our strong balance sheet, consistent free cash flow generation and confidence in our strategy to execute on durable growth opportunities, JFrog announced in late February, our first-ever share repurchase authorization of up to $300 million in ordinary shares. As of March 31, 2026, our RPO totaled $574.9 million, a 36% increase year-over-year, highlighting the successful execution of our go-to-market strategy as customers continue to make larger multiyear commitments to our DevSecOps solutions. As a reminder, our RPO excludes any benefit from customers' usage over contractual minimum commitments.

And now let's turn to our outlook and guidance for the second quarter and full year of 2026. As we enter the second quarter of 2026, we remain encouraged by the strength in our pipeline and emerging AI workload trends driving increased cloud usage. Even if cloud usage trends accelerate, our guidance philosophy will remain unchanged as we continue to derisk our largest deals due to timing uncertainties and any benefit from cloud usage above contractual commitments. Our outlook reflects growing contributions from our JFrog Security core products, ongoing adoption of our full platform and cloud growth driven from higher annual customer commitments. We are raising our estimated full year 2026 baseline cloud growth to be in the range of 33% to 35%. Given the anticipated contribution from our security core and increased baseline cloud growth assumptions, we now expect our net dollar retention floor to be 118% for 2026.

Turning to operating expenses. We continue to prioritize investments in innovation across our platform. We remain committed to a disciplined spending philosophy and are confident in our ability to manage expenses and drive ongoing efficiency in line with prior execution. For Q2, we anticipate revenues to be in the range of $154 million and $156 million, with non-GAAP operating profit anticipated to be between $28 million and $30 million and non-GAAP earnings per diluted share of $0.23 to $0.25, assuming a share count of approximately 126 million shares.

For the full year of 2026, we anticipate a revenue range of $628 million to $632 million, representing 18.5% year-over-year growth at the midpoint. Non-GAAP operating income is expected to be between $112 million and $116 million and a non-GAAP diluted earnings per share of $0.93 to $0.97, assuming a share count of approximately 128 million shares.

Now I'll turn the call back to Shlomi for some closing remarks before we take your questions.

Thank you, Ed. AI is transitioning from experimentation to tangible revenue, and we are seeing stronger momentum across our business. Looking ahead, demand signals for JFrog remains strong, including the durable cloud growth driven by AI, which is accelerating usage. New logo ASP is rising and demand for our security solutions amid the increasing frequency of software supply chain attacks is growing. To my fellow frogs around the world, thank you. This quarter, you didn't just deliver, you rose above. No matter the circumstances, you kept pushing forward, navigating with resilience, innovating with purpose and try and thing where it matters most for our customers. Because of you, we don't just move forward, we live further. May the frog be with you.

Operator, we are now ready for questions.

Your first question comes from the line of Sanjit Singh with Morgan Stanley.

I had 2 questions for the team. I wanted to start with Ed first. Obviously, great cloud growth, great total revenue growth in Q1. When I look at the outperformance versus what the estimates were, it seems like you guys came in about $7 million above on Q1. Q2, you guys came in ahead by a couple of million bucks, so roughly $10 million. When you look at the raise for the full year, it's somewhat less than that. And so I just wanted just to sort of say maybe check any sort of revised assumptions about the second half ramp. That was sort of my first question. And then I had a more strategic one for Shlomi.

It's a good question. We had a very strong quarter in Q1, as you highlighted, the growth in the cloud is 50%. And more importantly, we now see the mix in our cloud above 50%. We delivered 51% first time -- it's a milestone for JFrog, where we see more revenue coming from our cloud offering than we do from self-hosted. But we also are committed to our guidance philosophy, which we will only guide on those commitments. So while we saw the strength in Q1, much of that was being driven by usage over minimum commitments. We are deploying our sales organization, of course, to convert that into annual commitments. But until it becomes an annual commitment, it will not be part of our guidance, aligned with our philosophy.

That's very clear. And then Shlomi, the question for you is, it's a really interesting time, like some of our own field work on JFrog shows a real inflection in demand for the security side of the portfolio. It seems very clear to us. And I think you highlighted that in your script. At the same time, there's more of this longer-term structural debate on security overall and what the model logs will subsume. And there seems to be a take that seems like scanning, vulnerability management, vulnerability scanning, posture management, code security could be more of the purview of model logs longer term. And so to the extent that you guys have some exposure to those parts of security, I'd just love to get your latest thoughts on the long-term durability of those pieces of the security product portfolio.

Good question. So what we see in the market is kind of flooding of software supply chain attacks coming mainly around open source maintainers and the hackers are going after them. JFrog is positioned to secure our customers with that quite strongly. We called that in the script when we said that all the JFrog Curation customers were actually protected by those software supply chain attempt to be attacked. Moving forward, what's the real question? The real question is that can you really secure and govern the binaries, the artifacts, the outcome of AI. And what JFrog provides is not only a place that scans. Scanners are important, but the system of record of where you secure, manage, store, govern your artifact is actually more important because in a world of multi-agents that are all building and scanning and protecting and even fixing software, you still need to host it in a secure place.

The second thing, you will have to protect yourself from the open source world that will still exist, the Python, the NPM, the Hugging Face, the Docker, which is JFrog is doing at the gate. And the third thing is how you combine security of the new outcomes coming from agents or multi-agents with the legacy that is now being built. You still need to manage dependencies with the binaries of yesterday that are still hosted and still regulated and still are on the servers in your production.

The combination of the expertise that we built around binary security and not source code because this is a big confusion in the market. Coding agents are now securing source code replacing human beings. The combination that we built with that, including the moat around Artifactory, the system of record in a multi-agent world, including the open source on top of it and including the legacy, I think, gives JFrog customers the confidence to bet on that. This is also one of the things we called out, new logos are now buying JFrog with security, knowing that this is the future.

Your next question comes from the line of Radi Sultan with UBS.

Maybe just 2 quick ones. Shlomi, just on legacy code modernization. We've been hearing an uptick in JFrog getting pulled along in AI-driven legacy code modernization deals. So Shlomi, if you could just talk through like how big of an opportunity is legacy code modernization for JFrog? And where do you expect to see the biggest potential pull-throughs to your business? And then maybe just -- sorry, one more quick one for Ed. Could you speak to how impactful your AI native customers were to the cloud strength in Q1? Just want to get a sense of how broad-based the strength was.

Maybe I'll start, and Ed will take it from there. When we speak about legacy, we speak about legacy binary code, not source code. Basically, what you currently have in production is what we call legacy. What you have to regulate for the next 7 years, if you are a bank or the next 45 years, if you are an automaker, this is legacy. These are binaries that were built today or yesterday. And tomorrow, with coding agents, we still have dependencies that are in your servers in production. This means that those binaries need to be also first-level citizens in the system of record. Otherwise, how can you protect what is secured to be shipped.

What was made yesterday and approved and governed by the organization need to still be maintained in the system of record. So it's a very important asset that our customers are protecting still while coding agents are building the new binaries that are also scan and protected by JFrog.

And regarding the question on the native AI companies, we had a successful Q1 driven by a broad set of customers. So not only AI native customers, but traditional customers as well or non-AI native customers. You recall last year, we talked about $1 million land that we had with an AI native customer that renewed, and we're in continuous conversations with many of the large AI native companies, and we'll explain more update later.

Radi, if I may add to it, serving the AI labs is important, and we take pride of it, and we are very honored, but I think that once you become the power grid of this AI labs software supply chain, you learn much more in how you should serve the rest of the portfolio. And that's the big plus, not $1 million here, $1 million there, but mainly what we are building together with them as we power their software supply chain.

Your next question comes from the line of Michael Cikos with Needham.

Congratulations on the strong start to the year here. Shlomi, maybe for you. And one of the things we've been going through this earnings season which is still pretty quick on the heels of the SaaS apocalypse, which seems overinflated at this point. But one of the things we're seeing is the budget is there for strategic vendors. And so I'm wondering when you're going to speak with customers, is it fair to assume that this evolution of the Agentic stack or how AI is playing out is causing customers to rethink or the need to modernize their existing architecture. And as a result, JFrog is being pulled into that conversation and benefiting them with respect to cloud migrations. Can you talk to what the tempo of conversations you're seeing out there actually is like? And then I just had a quick follow-up for Ed.

So what is it that we hear from the market? What we hear from our customers is that every application to your point of SaaS companies, every technology that was built to have human interaction with technology is being question mark. Everything, every application, even source code, source code became cheap. Source code, as we mentioned in the script, source code is something that now you can do on an experimental level and you can do it 1,000x faster. But what happens when the machine language, the binaries need to be maintained, this is where they start to be a bit more cautious about how they plan the future. So for example, in order to enable AI, you need to use MCP servers. This is the interaction between machines and your solution. MCP servers are yet another binary. This is where -- to your point, this is where JFrog comes into the question, can JFrog become my MCP registry for all the MCP servers.

The same thing happened with NVIDIA when they ask us about skills, skills for agents, yet another binary. Can JFrog become the Skills Registry? So all of what we are hearing is that how can I build a stronger, better, scalable, universal system of record to manage all of these binaries because in tomorrow's world, what matters would be the machine language, not sources, not human language, but 0 and 1. And this is what JFrog did for the past 17 years.

And Ed, for a quick follow-up here, just trying to peel back layers of the onion as far as the strength in cloud that you guys saw. Is there any way to further qualify -- I don't know if you could talk to either the size of the cohort that drove the magnitude of that upside or how cloud over consumption trended through the quarter from a linearity perspective? Can you just put any finer parameters around that strength?

It was a strong quarter from start to finish, Mike, to be honest with you. It was very broad-based. It wasn't concentrated on one geography or one industry. I will say that what you saw in terms of the cloud was represented in our increase in the cloud guide. So we were very happy. We're confident with what's happening right now in the cloud, and that's what gave us the ability to raise our guide from 30% to 32% to 33% to 35%.

Your next question comes from the line of Miller Jump with Truist.

Last year, you guys were talking about AI experimentation driving consumption beyond commitments. It sounds very different today from your prepared remarks. So can you just talk about the difference you see in the amount of binaries in your system reaching production now versus a year ago? And I would also say it sounds like there's still a number of customers that are maybe waiting to commit bigger. So what are you hearing in terms of their hesitancy?

Miller, this is an awesome, awesome question because basically, you're saying source code is being produced at a completely different pace, completely different volume. Everything produced source code now. It's not just human developers, but all the coding agents together with the human developers. So the big question is, we see binaries growing at the same time. So you can think about it as like the digital photography replacing film. Film was expensive. You would take one shot on sunset before you print it. And now assume that you can take 200 of them. And instead of one printing, instead of one posting to your social network, you will now have 5.

Binaries are the asset that you will take to production. Source code became cheap, and now you can make more binaries that need to be immutable. They need to be tracked. They need to be governed. And you will see this growth in binaries and what you can take to production because of the change of AI. Same thing goes to governance. How do you make sure with the same metaphor, how can you make sure that the pictures that you posted on your social doesn't carry your home address at the background. This is what JFrog brings, not only dealing with the volume of new secure pictures, but also govern what goes out.

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

I have 2 questions. My first is, I'd like to better understand how exactly JFrog's revenue benefits from curation and advanced security. I believe there are a few parts. The first being you need tier upgrades where you have to be on Enterprise X and Plus to qualify for buying those products. And then as you make commitments, you obviously get that -- you get a commitment. And then there's over usage, I believe that's driven by increased traffic from attacks. So I just wanted to run that by you if those elements are correct.

Yes. So I'll start speaking about JFrog Curation and JFrog Advanced Security and JFrog Xray and Ed can speak about the over-usage and what we found. Well, listen, everything that comes from open source, whether pulled by agents, AI agents or by human developers is something that need to be protected before it steps into Artifactory, your single source of tools. When we built Curation, it was based on customers' request. They asked us to give them a firewall that will enforce policy of what comes in. That was in a completely different volume when it was made by humans pulling open source packages.

Now when you have 1,000x faster pulling request for open source packages from public hubs, whether NPM, Docker, Hugging Face, Conda, PyPI, you know that you are subject to attack. And the attackers are also using coding agents. They also became more sophisticated. They're also going after the maintainers that they know that by an order of magnitude will be the malicious packages. So what Curation did very successfully, not only apply this firewall enforcing your policies, but also scale to this level of AI. And this is why our customers not only embrace Curation, but also increased the demand for it after every attack we saw since last quarter of '25, which I alluded to this quarter with MCP and Python and others.

Regarding JFrog Event Security and JFrog Xray, once it is in, once it's inside your system of, once it's inside Artifactory, you need to still maintain the security of your software supply chain. You need to look for secrets that were exposed. You need to look for composition analysis. You need to look for dependency graph security. And this is what JFrog Advanced Security and Xray is doing. And then when you shift the production, you ship something that you can actually trust.

And Howard, this is Ed. Regarding the monetization of Curation, the monetization is based off of seats. This is a common currency in security, and we monetize based off of the seats. So regarding the attacks and an increase in attacks that certainly drives a demand from our customer portfolio and new customers to take either an increased number of seats or adopt Curation, but it doesn't necessarily drive data consumption. Data consumption is being driven by packages coming in and out of the organization or going into production. So Curation itself is not driving the usage over minimum commits.

Your next question comes from the line of Mark Cash with Raymond James.

Shlomi, maybe I wanted to build off a few previous questions and ask about NCP registry and AI catalog because there's a lot of companies saying don't provide the visibility and security for AI agents. So I guess where in the customer journey do organizations realize they need JFrog governance capabilities? Because what pain points are they seeing that others can't solve before coming to you?

What's happening now is that every software provider already provide an MCP server because we all know that if agents will not have an interaction with your software, that would be the end of your software usage. MCP servers are a binary code. No matter who provides that, it's a binary code. So our customers came to us and asked for MCP registry. As they trust MPM packages inside Artifactory or Python inside Artifactory or Docker containers inside Artifactory, they also want to have a list of MCP servers that they can put in an MCP registry, which is what we released this quarter. And then they can sell all of the AI agents or human developers, this is a safe place to take your MCP servers from.

Same thing happened with Skills, which is a very growing trend when you use coding agents. And there is some kind of a movement now to CLI, which is the third technology. All of the above are binary code, all of the above are a natural expansion of our solution, and therefore, they are sourcing Artifactory.

Your next question comes from the line of Jason Celino with KeyBanc Capital Markets.

The value proposition of Curation is quite compelling, right? And as you noted, these Curation customers were protected in Q1 from the software supply chain attacks that we saw in the news. It seems like a no-brainer to me and to most investors. But to the customer, what might be the alternative if they don't choose Curation or what factors are being considered that might be delaying that customer's decision? And given you are seeing this tremendous demand, do you have the capacity to kind of meet it?

Jason, I think it's clear that for a very long time, JFrog said we are betting on a world of automation, a world where machine will have to manage the asset. And therefore, we never shifted our focus from managing binaries. Every binary management tool is an alternative. So I think that, therefore, the strong differentiators that JFrog brings like universality, JFrog is the Switzerland of binaries. JFrog not only serves all the binary sites, but all the coding agents, human beings and other citizens that are using our solution. JFrog scale, we built 17 years of scalability. We went with the biggest organizations on the planet to scale to their level. And now we are even elevating it more because of AI. So scalability matters.

JFrog is hybrid. We give you the freedom of choice of running it in the cloud, on every cloud and on-prem, if this is what you prefer if you are in a highly regulated environment. JFrog integrates with all your ecosystem tools when it comes to DevOps, DevSecOps, DevGovOps, which gives you also the freedom of choice and not getting into a vendor lock-in. So if they will come a solution that provides all of this in a universal way and go so well complementing the AI change in the world of machine language binary, that would be a threat of JFrog. I hope that we put the moat around what we built best, which is the system of record.

Your next question comes from the line of Kingsley Crane with Canaccord.

So on the Q4 call, you called out that the November NPM attacks had driven both immediate curation revenue as well as building pipeline. I'm just trying to get a sense of if there's more urgency around procuring Curation or Advanced Security versus some of these like larger software architectural decisions that could take multiple quarters. And I guess more specifically, just on Q1, like how much did Curation drive the upside in cloud in Q1?

Yes. Well, listen, Kingsley, every time that there is some kind of a software supply chain attack, we see a rise in the pipeline. And obviously, a lot of our customers are concerned, and that's an immediate impact. But what happened when it's happening every few weeks. This is what's happening now. It's happening every few weeks. And it used to be SolarWinds and then a year after Log4j and then year after something else. Now you refresh your browser, there is a software supply chain attack. And why? Because source code doesn't matter anymore. Source code scanning is something that used to be overappreciated.

Now people understand that what needs to be protected is what's going to production. And the hackers, the attackers also understand that. So they go after the maintainer of the open source packages, and this is what you have to protect yourself from. Now will there be companies that will kind of react based on fear only? I think it's forever kind of the trade-off. But we see more and more responsibility on the customer side, knowing that the magnitude that they are looking at is completely different than what it used to be yesterday.

Your next question comes from the line of Shrenik Kothari with Baird.

So Shlomi, Ed, you have been careful in the past not to oversell AI as an immediate sort of revenue windfall and Ed's own words described 2025 more as an initial spark [ Danfire ]. And Shlomi, you drew a comparison now with the transition from like film to digital. So as once this higher-quality AI code with Opus, Codex is likely reaching more production, and we are hearing anecdotes about it, that definitely creates more valuable binaries for you to act as a system of record. Just where are customers today on that journey from that AI slop to production grade? And what specific indicators are you watching that would tell you that the fire has started or is going to start?

Yes. I think that what we see today is more in an experimental kind of mode. Everybody is trying everything. Not so long ago, we would speak about Copilot and Cursor and now everybody is speaking about Anthropic and Codex. And I think that a lot is being adopted on every organization means that as we used in this metaphor before, like it can take many more features. It doesn't cost you anything. But what we also see is that not even a single customer has a full autonomous process. This is not yet there. It's still a combination of human developers with coding agents. There is no coding agent that starts from scratch and push to production and maintain the production fully autonomous.

So there is still some miles to do before AI will take over completely of the developers' position. But we start to see the collaboration between strong human developers and coding agents, and this is why there is the rise of the water on every front.

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

Shlomi, back to your comment during the prepared remarks about customers willing to, I'll say, absorb meaningful overages in the cloud. What do you think is the gating factor? Or why are they willing to do that and not commit and get a better rate?

Yes. There is a race of AI adoption in every company now. No matter if you are a small or medium business or if you are one of the largest banks in the world with 50,000 developers, it's coming from the Board. It's coming top down. The Board is asking about AI adoption, making sure that you are in the race and not kind of falling behind. So what we see now is that usage in the cloud is also part of this experiment.

Now if you go to the CFO and you say, JFrog asked me to commit, the CFO will ask commit to what. And therefore, they give you -- they leave the meter on, they let you use more and even pay more. And now our key mission is to make sure that we convert this over usage into commitment to gain a win-win situation with our customers. It will come. It will just take a bit more time because of the predictability that is now missing.

Your next question comes from the line of Lucky Schreiner with D.A. Davidson.

Maybe a bit of a follow-up there. Previously, you've spoken to some customers preferring to buy JFrog on a self-managed basis, given the better visibility and cost controls around that. But I didn't get a sense of those trends from the prepared remarks today. So one, is that fair to say? And two, is there maybe any potential reason for a change in those trends?

We still see customers asking for the self-hosted solution or on-prem solution. It's split into 2 profiles. One is the big AI labs that are building their own data centers. They have enough money, they have enough capital. They don't want to share anything with the public cloud from whatever reason you can imagine. And they will take an on-prem solution and embed it into their software supply chain architecture. The second group that we see is a group of the highly regulated companies, government institutes or whatever organization that need to be highly regulated. And they will, for sure, do a lot of tests and experiments on an on-prem environment before they will go to the cloud or to FedRAMP.

And at last, what we see are companies that are well established and kind of seasonal players at the on-prem environment. So they are not looking to have now 1 or 2 entities in their world playing in the cloud. They keep on extending their own on-prem. But as you can see in our numbers, it's not only part of our strategy to migrate our business to the cloud. And this quarter, we also announced first time that it crossed the 50% of total revenue. It's also a benefit that we keep in our pocket to become the only company that gives you a full hybrid solution with a full freedom of choice. Like no matter what you are, no matter who you are, we can give you the freedom to embed AI or to adopt AI in your environment.

Your next question comes from the line of Jason Ader with William Blair.

I wanted to kind of revert to an earlier question, which was asked about the risk that the LLM guys encroach into the binary layer. And Shlomi, I was hoping you could talk about some of the announcements that the labs made during the quarter where they started to talk at least a little bit about binaries. It was too technical for me, honestly. So if you could help enlighten us and just talk about what they announced around binaries and why it's not something that you worry about?

Jason, let me start with the last sentence. I'm worried about everything. There's nothing that I'm not worried about. But I have the confidence that what we are building alongside the company is completing and being complementary to what the world is demanding. What you have heard is reverse engineering binaries, I guess that you refer to the OpenAI announcement about 5.4-Cyber. This is a way to kind of take the binaries themselves and reverse engineer it and to see what they were built from. That's not replacing JFrog solution because even if you are running kind of fast forward 2 years from now, when every organization would use OpenAI next to Anthropic, next to Cursor, next to Copilot, next to Gemini, I guess that we will all agree that even if you have this environment that they all build binaries, you need a governance tool that provides a universal solution to contain them all.

The second thing, who will protect the open source. It's not reverse engineering the open source packages. It's what the agent speaks itself. So when you bring something from NPM or something from Docker, how do you make sure that what you brought into the organization past your firewall? How do you make sure that this is secure?

The third thing is that this race is not just happening between the defenders and the vendors. This race is actually happening between the defenders and the attackers because the attackers will also use cloud and they will use products in order to build a more sophisticated malicious attack. So how can you make sure that the policies that you put to the gate are secure in your system of record?

And last, what happened when one authority should take a decision of what's going to production? Will it be one of the coding agents? Will it be 2 of them? Will it be human being? Will it be the company policy? This is the infrastructure we provide. We are not the policymakers. We are the policy enforcers. And we help you to make sure that what goes to production kind of came out clean from the non-poison reservoir.

So while we are seeing this happening, it's mostly focused on what I built as an agent, replacing human being, replacing human language to what I need to secure at the end and to govern and to trust. So that's how we see it now, and this is what our customers are telling us.

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

Shlomi, on the security side, we've got a lot of questions on how much of your revenue comes from Xray since the labs now have similar products. It'd be great if you could clear that up for people. How should we think about the contribution of that versus Advanced Security and Curation and just the main barrier to entry for the latter, the Curation and JAAS.

Xray was a part of our DevOps offering. And why is that? Because we don't think that Xray should standalone and start to do software composition analysis. We think that Xray should run over your Artifactory, making sure, for example, that your containers 4 years down are secured to be in Artifactory. Can other tools replace that? Yes. But if you get a start of your package using Artifactory built into Artifactory, why do you need another tool?

The second thing that Xray brings is this understanding of what's coming out from the open source environment and to be able to break that in pieces and to secure that. Can it be something that brought from the outside like a point solutions tool on top of Artifactory? Yes. But what we see is that thousands of customers prefer to take it as part of their DevOps subscription with JFrog, knowing that this is a build-in solution on top of your system of record.

Your final question comes from the line of Koji Ikeda with Bank of America.

When I look at cloud, the net new revenue added this quarter, I think, is the most ever in a quarter, let alone a first quarter. And so that absolutely implies customers are spending above commitment levels like never before. And so why is it -- or maybe the better question is, how long do customers typically take before they come to JFrog and start renegotiating their contracts for higher commitment levels, which presumably come with better volume discounts?

Yes. Regarding the why, I'll make it simple. We called it before. We will see more code means more binaries, means more JFrog. And JFrog is well known for being the binary people. Regarding how long it takes, we are actually -- we are not waiting, Koji. This is part of our practices, our enterprise sales practices changed something like 2 years ago. We are going to those customers with a better offer, with a better plan if they are committing. The question would be how long this experiment will be mature to be discussed as a commitment. And this is something that I'm sure that we will keep on following and provide you with more clarity of where the cloud goes.

But if you look at the confidence in our guidance, we raised the cloud, although we see that a lot of it comes from usage over commitment, we raised the guide for the year because we know that this is not a trend, this is not a spike. It's a few quarters already that we see the growth in the cloud.

This concludes the question-and-answer session. I will now turn the call back to Shlomi for closing remarks.

Everyone, thank you for your questions. Thank you for your trust, may the God be with you and may we have a great year.

This concludes today's call. Thank you for attending. You may now disconnect.

All right, getting towards lunch time. Hopefully, people are grabbing some food. We're on with -- we're continuing the TMT Conference, and we're super happy to have the management team of JFrog. JFrog has been at our conference almost every year that since you guys have been public.

We have Chief Executive Officer, Shlomi Ben Haim; and Chief Financial Officer, Ed Grabscheid.

Ed, so let me thank you again for coming back to the TMT conference.

Thank you for having us.

Thank you for having us.

Awesome. Before we get -- so much to talk about when it comes to the JFrog story. Before we get there, for important disclosures, please see the Morgan Stanley research disclosure website at www.morganstanley.com/researchdisclosures.

With that, let's kick off the conversation around JFrog, and maybe to level set, Shlomi, we're out of time in software. There's a lot of uncertainty. And so I think people are coming -- investors are coming back to doing first principles analysis on what these software companies are, how do they create value. And so from your perspective, can you just walk me through the problems JFrog solves for its customers and why the company is adopted by more than 90% of the Fortune 500?

Yes. So great being here again, Sanjit. I think that what we are seeing in this evolving market is the adoption of the trust layer of software supply chain. That's the main reason behind everything, whether it's AI dreaming or development-driven, and JFrog provides exactly that. In the world of software supply chain management, you need a very strong system of record, in order to enforce security, in order to enforce governance, in order to make sure that what you distribute is also safe, in order to enable automation, in order to have universality of tools and universality of packages. And JFrog is the company that govern that with one primary asset in mind, managing your binaries. And binaries are the outcome of whatever source code developers or agents are writing. This is where JFrog is being more appreciated.

Awesome. In your messaging for years, you've been clear that JFrog serves as the system of record, single source of truth for binaries. And increasingly, I think the goal is to become play that same role for AI artifacts and AI models. What's the important -- what's the most important shift in how customers are using you that makes that framing truer today than maybe a year ago? So looking at that shift between containers, managing containers to AI models. I'm sure it's pretty early, but just any -- what's sort of the tea leaves that you're reading from that perspective?

Well, when customers are choosing JFrog platform, they are looking at different aspects. They are looking at the aspect of securing the one system effect or the single source of truth that every artifact, every binary, every software package that comes in and comes out, comes from the same place that was clean, blessed by the organization policy and whatever compliance will they enforce.

The second thing that they are looking at is the security layer, not only how do I secure this system of record, how do I secure this vault of binaries but also what do I put at the gate between my organization and the entire open source Wild West. What do I put outside of Artifactory when I'm distributing binaries because as you all know, the only asset in your deployment environment, in your on-time environment is again binaries. And also what our customers are telling us is that what we see lately in the last years is software supply chain attack. It's not source code attack. Hackers and attackers are not after your source code anymore, not necessarily coming from public hub, but they are attacking the binary. So the threats in the market is also different. So when you think about why customers will adopt JFrog platform, it's because of the system of record, because of the governance, because of the enforcement and because of the security at the gate and before the runtime environment.

Yes. Let's dive into that a little bit more, right? So just over a week ago, a code security solution was announced by one of the leading model providers. JFrog stock was down 20% in a single day. And so can you give us a sense where JFrog plays in terms of the software security supply chain? And do you find your sales in the crosshairs of where the model providers are going from a security perspective?

Yes. Well, so much happened since February 20. And it's just amazing to see, and we are also very excited to see how well AI is accepted and adopted and nobody see it as a trend. People understand that the world of software is different. But what specifically announced by Anthropic, I think you referred to, is that the code agent is now not just building the source for you but also securing it, meaning scanning for vulnerabilities. And not only scanning for vulnerabilities, but also offering you a fix. And not just offering you a fix, in one click of a button, you can co-fix and have a better source code, that's amazing. I think the impact that we saw in the market is the impact of those who confused source code security and binary security. And then the more sophisticated shareholders or analysts ask that, okay, you know what, we got it. You are binaries, they are source code. But what promised me that tomorrow AI becomes better and smarter, why AI would not become the system of record. And the answer -- first of all, I want to be honest. OpenAI just raised $110 billion. They have brilliant people. If they want to become JFrog, they can become JFrog. I don't think that there -- this is their core business. But when you think about source code scanning versus binaries protection, you have to go down to the core. The core is the system of record, the single source of truth you protect.

Now why it's important? Because there is no one I hope in this crowd or anywhere else in the world to think that whatever company will have only one agent. You will have Anthropic, you will have OpenAI. You will have Copilot from Microsoft. You will have Gemini from Google. You will have a multi-code agents environment. So who's governing who. Unless you have a universal to integrate it to fail system of record that all of them are working with this same system of record.

Now let's say that this is also done. What happened when you combine that, this 10%, maybe 15% of your code with the rest of the open source packages, the 90%, the NPM, the python, the containers from Docker, the Hugging Face model, what happened with them? Who protects you from combining them? That's a second question that also being sold in a universal solution like JFrog that set as your single source of truth.

Third question, Anthropic generated with Cloud -- by the way, JFrog developers are using Copilot and cloud. Anthropic generated an amazing source code secured 10x better, created the binary out of it. Binaries pushed into Artifactory, great. Then OpenAI comes and use this binary and build the other source code with it and create another binary. Now you have a dependency, who's managing who.

You must have the enforcement layer that's not just protecting the company from vulnerabilities, but also managing, orchestrating governance and security all the way from the creation to production throughout the software supply chain. And I'm happy to see that not only JFrog believe that some of our customers are those exactly native AI companies that are practicing the same.

Yes. That's a great point. I mean, as you know, I've been around for every single JFrog quarter, you and I met before the IPO. I mean, this just seems like another permutation of a classic question concern around JFrog. I mean, remember, it was Microsoft, right? Like why can't Microsoft do binary package management? And how that story resolved is strategic partnership, right? And so -- and we sort of point out that you guys have the leading model providers as customers as well. So I think -- when we think about the nuances of the debate, I think the next angle is if AI native companies don't outright replace JFrog, will it erode pricing power and pressure what has historically been a high-margin business? What's your view against the notion that AI significantly reduces the terminal value of incumbent software vendors like JFrog?

Yes, I think that what happened on this Friday, the 20th was that people started to ask questions about the terminal value, right? So we already saw that you guys know how security is going to look like. 2 years ago, 3 years ago, people ask us, would you be able to sell security? Will you be able to address the [ upseec ] and the CISO pain? We prove that, not only with what we delivered, but also with the ARR numbers we shared, with the RPO number we shared. We show the world what it means to have a holistic solution and not a point solution, covering your software supply chain. But the terminal value question is, okay, so I thought based on your success that it's 10%, maybe it's 3%. And now I have some fears that JFrog will be replaced. And from what -- from where I see it, and what I see, and it's simple math.

Code agents are being adopted on an hourly basis. It's amazing to see how fast this innovation is being hugged by the industry. Code agents are not going to sleep. They are not going to eat. They are not taking PTOs. They are not going anywhere. They just create more and more and more source code. They build with more and more source code and create more binaries.

This Tsunami of binaries, where will it land? Where will it land? Storage, maintenance, dependency maintenance, security, distribution. Where will it go? Now not only did JFrog is a universal solution, it's the Switzerland of agents. It's also the database of DevOps. These agents are also building with what you created yesterday and approved by the organization, and it's in Artifactory. Not only that, it's also a combination of open source and agents. So I think that from terminal value, the autonomous -- the more autonomous our world will become, the greater the need for governance and enforcement and rules that you will have to apply to make sure that your organization is secure. So we are very excited about it. I understand that the market is showing some feel or panic and it's on us to execute and to pull...

Awesome. To sort of wrap up this line of question, I wanted to bring Ed into the conversation. So on the -- post February 20 in the last week or so, you announced first ever $300 million share repurchase program. Why the decision to pull the trigger on the share repurchase? Will it go beyond just managing dilution? And what time frame are you considering to complete the program?

Well, thanks for the question. There's so many questions about technology. There's actual financials and fundamentals that go behind that. And part of the reason why we did this is not because we saw great value with where the share price was. But the fact that we continue to generate cash that we have strong fundamentals. We manage with discipline, we continue to generate cash. This gave us the flexibility to do that. Now what happened on February 20 with the stock declining at the rate that it did, we saw an opportunity to deploy capital and stabilize in a sense, putting a stake in the ground saying, we firmly believe in our ability to execute going forward. Therefore, we're doing a $300 million share buyback program.

How long that program will be? It's open ended. We are certainly going to look at opportunities as it continues to present itself with a favorable share price and we'll build it accordingly. But the time will be determined based on the price, and we'll continue to keep a close eye on that, and we think it's a great use of capital.

Awesome. Maybe sticking with you, Ed. As we think about fiscal year '26 guidance, you guided total revenue 17%, 18%. The cloud business -- excuse me, between 30% and 32%. How should we think about the level of conservatism against what appears to be a more -- excuse me, constructive demand environment versus a year ago? And the security business is frankly also gaining more traction.

Yes. So the philosophy didn't change. In fact, the philosophy remains exactly the same in 2026 as it did in 2025. What happened in 2025 as we started to see usage over minimum commit, something we didn't see in 2024. And as we step into 2026, the sentiment is better. We see a better environment. We see momentum that's being built in usage, but the philosophy itself didn't change. So assuming that things continue to progress the way they did in 2025, we would expect to see better performance against what we guided and because of the usage over that minimum commit. We also see our cloud -- I'm sorry, our security, building momentum. And as long as we continue to convert those opportunities, we'll see an outperformance against the guidance that we provided. But the philosophy itself remains the same.

Can I ask one follow-up before, Shlomi and I talk more about where the market is headed. If I go back to like 2024, you guys won some really large migration opportunities that help business growth. The theme for 2025 was really customers using an excess of commitment, maybe less on the migration side. With respect to those 2 particular vectors, migrations and excess commitment, how do you see that? What's your initial hypothesis of how that plays out in 2026?

Yes. So our existing customer usage and expansion of those customers is going to outpace what we're seeing today in migration. So customers today from a migration perspective, the very large migrations are being paused. We still see it many customers moving and migrating from self-hosted to the cloud. That hasn't changed, but the magnitude of the dollars and the large projects, those are on pause. So most of the growth will come from expanding of those existing cloud customers, some smaller migrations from self-hosted to cloud. But in terms of the very large cloud migrations that we saw during 2024, right now, those are due to predictability questions and uncertainty that these emerging AI trends are driving. Those will most likely be on pause during 2026.

Understood. Okay. Good to understand the dynamics going into next year. Shlomi, I want to talk about where the software development cycle is headed and what role JFrog will play? So I think a lot of confusion when it comes to this space that I deal with all the time when talking to investors about this category is. Code is just one piece of the process of getting software into the hands of customers. What role will AI play in the broader software development cycle? And how will the role that JFrog plays change in an AI-powered AI agent software development life cycle?

Yes. Well, listen, if anyone in this conference or anywhere else in the world, will say that they know where AI will go. I think that it's a bit too early. So humbly, I will say that it's amazing to see how fast AI is doing exactly what they said that it will do and replacing kind of a human labor and simple tests. We will see developers -- first phase will be developers empowered by agents. And second phase will be developers moving from being players to being coaches. They will start to manage agents. And the third thing that we will see is that agents are having a full autonomous power, not only to build, but also to take it all the way to production. At JFrog, we are looking at this phase already. Because I think that the movement that we will see is more and more business will understand that B2B is over. You have to think about business to agent and how the agent will pick me as a vendor because of whatever I can provide.

So that's our focus on our JFrog 2030, the next 5-year strategy is about this shift, but it will take time. And until then, we will see a growth of the asset, the primary asset that this agent will generate, which is the binary. JFrog is built for this scale from day 1, not only because of the hybrid and multi-cloud, the solution that we bid, but also the storage layers that we build that scale better than everyone. And how well we know this asset we call binaries.

The third thing that we will see is that the security aspect will be different, and there will be much more kind of focus on enforcement and governance and making sure that no agent is doing crazy stuff. And this is, again, when you need the system of record. So when we are looking at the future, I think -- it's a simple math. You will see more binaries. You will see more requirement for universality and flexibility and faster adoption of AI technology.

Now one last sentence, Sanjit. We also have to remember that with all of these good things that are coming, attackers and hackers are also going to embrace AI, okay? So the race and the pace of the attacker versus the organization will stay because there's no hacker that will say, I'm not using cloud or I'm not using OpenAI or I'm not using Gemini. The malicious side of software is also going to be more sophisticated. And we have to put some guardrails around shadow AI, where AI was used, how it was used, identify that, trace that, make sure that governance also comes with the right auditable signed artifact. So I can trust this and not just saying that someone check the box on it.

So that next evolution that you speak to often is something from going from DevOps, DevSecOps and DevOps, which is a very interesting point. I also wanted to get your take on this sort of structure of the market. And what I'm referring to is that I would say pre-pandemic, if you looked at the DevOps market, it was pretty fragmented, right? If you look at the Infinity Loop and overlaid all the vendor landscape, you probably count 50 different vendors. If there was -- coming out of the pandemic and moving to a higher rate environment, budget is a little bit tougher. We did see a move towards consolidation. You guys have been benefiting from that from the security side. The question is that do we consolidate further? And like some of the start-ups of the Silicon Valley kind of expressed this view of a invisible SDLC where all of the workflows get executed in a particular agenetic platform. What do you view -- like as we go into the area, do you think this is going to be a multiple vendor environment, a heterogeneous environment? Or is it going to be kind of a winner take all?

What a great question. JFrog was founded 15, 16 years ago before DevOps was even a phrase. We called it developers acceleration, automation, whatever. And then it got so well adopted, automation was so much required and became a domain. And then it was evolved to DevSecOps because DevOps build speed then developers became fast and dirty and everybody wanted security to be enforced. So DevSecOps came in. Now we spoke about governance and DevGovOps. And the evolution is there, and you see a lot of companies that are not here anymore. Where are all these pioneers, they were either acquired or left somewhere. And a lot of these tools became commoditized. Think about CICD. I remember that people told me that the CICD of the world will acquire JFrog. And think about containers. I remember that people told me the Docker is everywhere, and by the way, Docker is everywhere until today, even in AI, you use Docker. But where is the consolidation that everybody spoke about. We will not need universality of software packages. We would not need NPM may even go and Python because of Docker. And still, it happens. So what really to your question, what really got commoditized or what kind of commoditization makes sense.

It's around an asset. And what we start to see, especially emphasized by AI is that the world is divided to two. Are you an infrastructure company or an application company? Infrastructure, thumps up, great. If you're an infrastructure, are you a platform or a point solution? If you are just a scanner of source code, done, you are out. There is no one in the IPO pipeline to say, I'm a point solution security anymore. It was just the reality 2, 3 years ago.

So are your platform? Yes, I'm a platform, check, thumbs up. Next question. Are you a foundational platform, meaning do you have a source code? Do you have a system of record that you provide that you can build value on top of it? If you are a CRM system of record, you are probably Salesforce. If you are a finance system of record, you're probably Intuit. If you're HRIS system of record, you are probably Oracle ERP. In the world of software supply chain since binary are the primary asset, JFrog became the system of record, and this is how we are powering most of our customers. So you will see more consolidation around the system of records and not necessarily the solution because solution will be commoditized.

And by the way, how easy it is to move from Copilot to cloud, from cloud to OpenAI, a matter of hours. How easy it is to move from your system of record? Impossible.

Yes. That's a great point. Let's get an update on the security business. Every Q4, at least for the last 2 quarters, you've given us some really great metrics. And so in terms of where the security business stands, in terms of ARR, we're up to 10% of ARR versus 5% last year. Security accounts for 16% of RPO versus 12% last year, and it's increasingly driving your larger deals. How should we think about the attach rate of the security business going forward with security incidents like with NPM and pipeline over the past couple of months? Do you see security now as a structural growth driver rather than driven by kind of onetime events?

Yes. So I can only look at the pipeline. And by the way, an anecdote, none of the opportunities in our pipeline, none of our wins in the past were due to static code analysis, which is the Anthropic announcement. But -- looking at the future, I'm looking at the pipeline, and we are very optimistic because there is a real use case there that is looking at JFrog as a holistic solution. Our customers are not just looking for secret detection, contextual analysis, binary scanning, container scanning, they are looking at the full software supply chain protection from the creation of the code all the way to the production. This is what we see in the pipeline.

The other thing that we see is that the attackers moved completely to attack software supply chain. It happened with Log4j, as you mentioned, Sanjit, last quarter of 2025, NPM [ Saikulud ], in between MCP attack, Python attack, SolarWind attack, everything is software packages. Everything I just mentioned is a binary attack. Not only that, when you try to remediate with the software package, it's different than source code. You have to open it and to look at all the dependencies. So the value that JFrog now brings in terms of ROI, in terms of enforcement, in terms of governance, is very clear. And we also understand humbly that all of our customers had a security solution before JFrog, and migration in the world of security is not happening in a day. And we have patient and we trust our value, and we see the adoption growing. And this is why it was important for us not only to provide you with the revenue numbers, but also the RPO numbers and the ARR numbers.

I mean to that point, you mentioned in the past is trying to find new ways in terms of customers, helping customers find budget for the security add-ons? Can you give us an update on how you're solving for this? Are there incremental investments needed beyond the security overlay team and some of the incentives you're rolling out for the sales force?

Yes. What we see is a very, very intensive collaboration between the CIO and the CISO of the organization. It's already kind of a mixed budget of who owns what. And yes, there is an incremental -- it's a growing addressable market. And why is that? A, because attacker think different. So introducing you to a new world of threat. And the second thing is nobody that I know is willing to adopt AI without first having it secured, trusted and governed and that all fall under the security. So on the gateway, the firewall part, which is JFrog Curation, you want to make sure that everything that comes in is kind of blessed, approved. The passport control is approved by the organization policy. JFrog Curation is an innovative tool. There is one or two other companies in the world that suggest that. JFrog brought that together with Artifactory. The other companies, even if they have that, they have to integrate with Artifactory, like you put a gate before of what, before of your system record. This is JFrog. So when it comes from JFrog, it's clear.

Now what happened inside Artifactory. There is a new set of risks that you have to mitigate. How do I make sure that agent didn't bring a GPL license from whoever or violating someone's IP that I will be sued. And the last piece is the distribution what goes to production. Again, binary, how do I make sure that what happened in AWS 2 months ago, Kroll an agent decided to be completely autonomous, push something to production, wiped out the entire region. These need to be governed and secured in a holistic way, and this is why the budgets are growing, but so is the life cycle. To be honest, we will not see it immediately. It's an adoption, education, enablement process. And we see a great horizon ahead.

That's great context. I wanted to talk a little bit about how investors should think about the growth equation for -- the basic growth equation for JFrog. I'll hand it over to you, Ed, to opine on. But essentially, what I found very interesting last year is that your $1 million customer cohort grew fantastically well. So I think up [ 49% ], 100,000 customers, up 15% year-over-year. Total customers account actually declined, and that's in part due to pruning some lowest ASP accounts. So with that context, you have an NRR that stands currently at 119%. As investors think about what kind of growth equation to underwrite for JFrog, how should we think about the contribution of growth coming from existing versus new customers?

Yes. I'll start with the strategy. Ed, if you want to add about the numbers and the ARR growth retention. Listen, we told you guys around 3 years ago, and I remember it clearly because we completely changed our sales and marketing focus. We are going after the enterprise. And when you build a solution, not just the product, but the service around it, the go-to-market around it, everything, when you build for contracts that justify this growth that you mentioned with over $1 million, consistently growing over $100,000 to $1 million, consistently growing; ASP, consistently growing. When you build for that, when you build for $1 million, $2 million, $5 million, $20 million customer, you cannot be focused on the $150 per month customer. And I will never fire a customer, but I understand why they don't see a value when I'm building a mothership, and they only need a bicycle.

So the proof of that was last year when we raised the price, the basic price from $3,000 a year for Artifactory to $6,000 a year, some of them left. Why? Because 100%, even if it's only $3,000 a year, 100% for them was too much. For me, it's signaling something else. If JFrog's strategy is to become your system of record, these guys are not adopting JFrog as a system of record. So this is not even a matter of who is the customer, it's a matter of, is it aligned with our strategy.

The second thing, and then, Ed, you take it from here. Second thing is that back to 2021, people bought revenue. If you want me to get to 8,000, 10,000 customers, easy, easy. I'm just dropping the price to $50 a month, and the number of logos will explode. But this is not what we do. We build something for the enterprise. We are very serious about it. This is why we also had to kind of remove all frictions internally. We had to take 300 entities and put them under their parent companies because I brought a field sales enterprise field guy that comes into the organization as an octopus and work with the CISO, with the risk team, with the governance team, with the DevOps team. And then you hear that there is an SDR that is doing inside sales with some entity. Of course, I had to kill this kind of phenomena, and to consolidate that, that's by 300 logos.

I'll be very quick. The growth algorithm, very similar to what we saw in 2025. Usage that continues and we see the cloud growing. I want to remind you, we started in 2025, 31% on the guide. We ended at 45%. We see something very similar. Security, although we don't give the number of customers in security, we certainly see a very long tail. We have over 3,000 customers today that are enterprise customers. They have the ability to cross sell and grow through security products, and we're actively pushing that. So if we continue to do that, I think there'll be a very similar outcome in 2026 that we saw in 2025.

And you see the gross retention, right, over 97%. You do the math, you understand that all of the customers that we bought hundreds of them that we bought in 2025, landed with a much higher ASP and the net dollar retention also grew, so simple math.

Awesome. Well, thank you so much, Ed and Shlomi, for giving us the update on the JFrog story.

Thank you for having us.

Thank you.

May the frog be with us.

Awesome.

Ladies and gentlemen, thank you for joining us, and welcome to the JFrog Fourth Quarter and Fiscal Year 2025 Financial Results Call. [Operator Instructions]

I will now hand the conference over to Jeffrey Schreiner, Head of Investor Relations. Jeffrey, please go ahead.

Thank you, Nicole. Good afternoon, and thank you for joining us as we review JFrog's Fourth Quarter and Fiscal Year 2025 financial results, which were announced following the market close today via a press release.

Leading the call today will be JFrog's CEO and Co-Founder, Shlomi Ben Haim; and Ed Grabscheid, JFrog's CFO. During this call, we may make statements related to our business that are forward-looking under federal securities laws and are made pursuant to the safe harbor provisions of the Private Securities Litigation Reform Act of 1995, including statements related to our future financial performance and including our outlook for the first quarter and full year of 2026. 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.

You are cautioned not to place undue reliance on these forward-looking statements, which reflect our views only as of today and not as of any subsequent date. Please keep in mind that we are not obligating ourselves to revise or publicly release the results of any revision to these forward-looking statements in light of new information or future events. These statements are subject to a variety of risks and uncertainties that could cause actual results to differ materially from expectations.

For a discussion of material risks and other important factors that could affect our actual results, please refer to our Form 10-Q for the quarter ended September 30, 2025, which is available on the Investor Relations section of our website in the earnings press release issued earlier today. Additional information will be made available in our Form 10-K for the year ended December 31, 2025, to be filed with the SEC on February 13, 2026, and other filings and reports that we may file from time to time with the SEC.

Additionally, non-GAAP financial measures will be discussed on this conference call. These non-GAAP financial measures, which are used as a measure of JFrog's performance should be considered in addition to, not as a substitute for or in isolation from GAAP measures. Please refer to the table in our earnings release for a reconciliation of those measures to the most directly comparable GAAP financial measures. A replay of this call will be available on the JFrog Investor Relations website for a limited time.

With that, I'd like to turn the call over to JFrog's CEO, Shlomi Ben Haim. Shlomi?

Thank you, Jeff. Good afternoon, and thank you for joining our call. 2025 was a remarkable year for the frogs. We didn't just fire on all cylinders. We set the pace. JFrog paved the way for securing and managing the software supply chain in the area of AI, expanded our product portfolio, innovated at a faster pace than ever before and build deep partnerships with the world's leading companies to strengthen our value to enterprise customers.

We delivered quarter after quarter, exceeding our commitments and pairing solid growth and expansion with strong business discipline and operational efficiency. This is what strategy focused execution and share belief look like. And it's what makes this journey special for us.

On today's call, Ed and I will walk you through our fiscal year 2025 results in more detail, share our fourth quarter performance and discuss our outlook and guidance for 2026. In fiscal year 2025, JFrog's total revenue was $531.8 million, up 24% year-over-year. Cloud revenue for 2025 was $243.3 million, representing 45% year-over-year growth.

In Q4, greater than $1 million customers grew to 74% compared to 52% in the year-ago period, equaling 42% year-over-year growth. Customers spending more than $100,000 annually, grew to 1,168 compared to 1018 in the year ago period, equaling 15% year-over-year growth. These strong consistent results reflect our alignment with modern enterprise market needs as software continues to transform how the world operates. JFrog's 2025 performance highlights a clear trend as human developers and AI agents generate software at a massive scale, the resulting surge in binaries demands a trusted platform to manage, secure and govern them end to end. This reinforces JFrog's leadership as the foundational infrastructure for software delivery in an AGM-driven wall.

Now I will discuss our success in Q4 in security, cloud and AI. Security first. In 2023, we set out to support our customers by giving them a complete chain of cyber trust from code creation to production. When we launched JFrog Advanced Security and JFrog Curation, we evolved rapidly from securing artifacts within JFrog Artifactory, with JFrog Xray to offering end-to-end trust and governance. JFrog has now has established itself as a complete system of record for software supply chain security that protects companies binaries, software packages and AI models even before software enters the organization.

We strategically challenged the security point solution market by unifying security and DevOps on a single platform built on one source of truth. We believe that's the only scalable way to protect our customers' software supply chain. The pain was real, the threat was growing and our results in 2025 reaffirm that this approach was the right one. In today's AI-driven software environment, that source of truth becomes mission-critical. Organizations need a secure center of gravity and the foundational platform that stores, manages and governs all artifacts where they're created by humans or machines.

Against this backdrop, we will continue to build JFrog as the system of record, the single source of truth that all developers and code agents rely on to securely manage and govern every binary and AI artifact in the software supply chain. Just as we reported, strong early results for JFrog Security in 2024, our 2025 annual reporting reflects another year of significant momentum and success of our security solutions.

In 2025, JFrog Security Core products, excluding contributions from JFrog Xray became an even more meaningful growth engine for the company. As of December 31, 2025, JFrog Advanced Security and JFrog Curation comprise over 10% of our total ARR. This past year, we built on strong RPO growth momentum, driven by increased cloud usage, DevOps adoption and our security core products. We are pleased to report that in 2025, JFrog Security represented 16% of our ending RPO compared to 12% in the prior year and nearly doubled long-term revenue commitments.

This growth highlights not only the rapid adoption of JFrog Advanced Security, JFrog Curation or both by hundreds of our customers, but also the broad opportunity ahead to provide value to the thousands of existing enterprise customers over time. JFrog Security Solutions tightly coupled with JFrog Artifactory, enabled end-to-end software supply chain protection, and are now offered as a bundle add-on with our enterprise subscriptions, increasing ASP, expansion in cloud usage and triggering multiyear commitments.

This approach resulted in tangible growth for JFrog [indiscernible] indicates that customers adding JFrog Security drove a strong account expansion with growth extending beyond security and into broad product portfolio usage. We are pleased with our financial performance and the execution of our enterprise go-to-market teams. Our security, RPO and pipeline numbers show strong momentum carrying into 2026.

Our success in security also reflects the deep alignment with the evolving threat landscape. Attackers are increasingly targeting the software supply chain through software packages. Our customers and community described the recent [ NPM ]. [indiscernible] incident as a mega attack on software supply chain exposing millions of Javascript developers echoing the impact of [ LOG ], PyPI and other recent security events. JFrog customers using JFrog Curation as the firewall remained protected from the world's leading enterprises with tens of thousands of developers to a small development shop, JFrog Curation in-force policies, secured the software supply chain and prevented business impact.

Expanding our security offerings in 2025, we also announced the availability of AI catalog and agentic remediation capabilities to address emerging challenges created by the introduction of AI models and agent-generated code into the software supply chain. JFrog is the mission-critical infrastructure of the company's primary software assets, the binaries. As more code is generated by human developers and AI coding agents as tsunami of binaries is being created. More binaries, artifacts and model lead to a greater need for trusted infrastructure that manages secure and govern the software supply chain at scale. We anticipate these growing needs will drive sustained customer adoption of our holistic security solutions through 2026.

Second, cloud. As we noted earlier on the call, 2025 was a year of high-quality sustained growth in our cloud business. We delivered 45% year-over-year growth while remaining highly disciplined in usage management and continuing to migrate customers toward annual commitments. In addition, we drove a higher volume of security deals in the cloud, strong new logo wins and deeper marketplace partnerships. During the year, we strengthened our partnerships with all major cloud providers, improved our commercial terms and established a stronger long-term gross margin strategy. Our cloud momentum was also supported by focused investments in service performance and availability, the continued build-out of our global SRE organization enhanced sales incentives and the consumption-based pricing model aligned with the market standards.

Customer purchasing decisions change in 2025 as CIOs with less focus on mega cloud migration initiatives and instead, increasingly emphasize building fleet to purpose hybrid and multi-cloud architectures as they adopted new AI solutions. Looking ahead, we see trends that we believe will continue to mature. First, clarity. While cloud remains the preferred deployment environment, CIOs are seeking greater clarity, cost predictability and ROI as they assess the full impact of AI adoption at scale. along with the evolving security and regulatory requirements that comes with it.

As AI adoption matures and clarity improves, we anticipate CIOs will increase investments in cloud infrastructures. Second, AI-driven software package ecosystem, such as PyPI, Hanging face, NPM, Conda or Docker are driving consumption spikes from both developers and AI agents in some cases, going beyond customers' commitments. This has the potential to be a tailwind for JFrog, and we expect customers to align commitments once usage patterns stabilize. We will continue to provide guidance based on customers' annual commitment and proactively derisk exposures to volatile usage-driven and sizable deals. Our go-to-market strategy will remain focused on converting usage overages into annual commitments. Looking ahead, we believe that in 2026, we will continue to deliver durable growth in the cloud.

Finally, I will address AI and MLOps. In 2025, we strategically built the foundation of JFrog ML as part of our platform to deliver the capabilities, enterprises will increasingly require to automate speed, trust and control across the modern life cycle. At the same time, we introduced advanced functionality and integrations that extend beyond traditional B2B workflows into the emerging business to agent market. We released JFrog's MCP server as a core integration layer, enabling AI agents and LLMs to securely interact with the JFrog platform.

We introduced the JFrog AI catalog for model discovery and governance, extending the platform to manage AI and ML models like other software packages. We also announced agent-based security and remediation, leveraging agent capabilities to drive detection and automated recovery. To strengthen our position as a system of record for all AI artifacts, we partnered with NVIDIA Enterprise AI factory to serve as its secure NIM model and artifact registry while also partnering with Hugging Face to secure the world's leading open source hub for models supporting trusted enterprise consumption of AI.

Looking into 2026, our road map includes capabilities designed to expand our growth and to support JFrog's users needs. Whether human machines or hybrid engineering teams of developers and AI agents. These expanding use cases represents a significant opportunity for JFrog. The market is experiencing a tsunami of binaries accelerating by AI. JFrog was built from day 1 to handle exactly this asset at this case. And we are focused and fully committed to providing the infrastructure modern software organizations need to confidently embrace the AI-driven revolution.

With that, I'll turn the call over to our CFO, Ed Grabscheid, with an in-depth recap of Q4 and 2025 fiscal year financial results as well as our updated outlook for Q1 and the fiscal year of 2026. Ed?

Thank you, Shlomi, and good afternoon, everyone. We are very pleased by the results of the fourth quarter, exceeding the high end of the range on every metric we guided for the quarter. It was a strong finish to an outstanding year, highlighting our consistent execution, operational discipline and durable business model.

During the fourth quarter of 2025, total revenues equaled $145.3 million, up 25% year-over-year. For fiscal year 2025, total revenues were $531.8 million, up 24% year-over-year. Fourth quarter cloud revenues grew to $70.2 million, up 42% year-over-year and represented 48% of total revenues versus 43% in the prior year. Our growth in the cloud was driven by customers expanding annual commitments and increasing demand for JFrog Security Core as ongoing NPM incidents during the quarter accelerated customer adoption.

For the full year 2025, cloud revenues equaled $243.3 million, up 45% year-over-year. Full year cloud revenues equaled 46% of total revenues versus 39% in the prior year. During the fourth quarter, our self-managed or on-prem revenues were $75.1 million with full year 2025 equaling $288.5 million or up 11% year-over-year. Aligned with our strategy, we continue to proactively engage our on-prem customers to migrate DevSecOps workloads to our cloud or explorer solutions better align with their specific use cases. including hybrid and fit-for-purpose deployments. We experienced another year of strong customer adoption of the complete JFrog platform driven by customers looking to consolidate their technology stack and secure their software supply chain.

In Q4, 57% of total revenues came from Enterprise Plus subscriptions, up from 54% in the prior year, while delivering year-over-year revenue growth of 33%. Driven by the ongoing execution of our enterprise go-to-market strategy and broader customer adoption of the JFrog platform, revenue contributions from Enterprise Plus subscriptions grew 36% year-over-year in 2025. Our security core products, which exclude any benefit from JFrog Xray, continue to gain momentum as customers actively consolidate point solutions.

For the full year of 2025, security core revenue was 7% of total revenues, with our security core products now comprising more than 10% of our ending total ARR. Driven by an increasing number of large multiyear commitments to JFrog, our security core represented 16% of remaining performance obligation, or RPO, as of December 31, 2025, compared to 12% in the prior year. Net dollar retention for the four trailing quarters was 119%, an increase of 1 percentage point from the prior quarter highlighting the continued adoption of our security core products and increased cloud data consumption, resulting in higher customer commitments.

We continue to demonstrate that our customers view JFrog solutions as mission-critical to their software supply chain with gross retention that equaled 97% as of the fourth quarter of 2025. Our customer count in fiscal year 2025 equaled approximately 6,600. During the year, we continued to execute on our strategy, focusing on our go-to-market initiatives around the enterprise. We further matured our approach by refining our customer logo methodology to eliminate friction for our customers and sales teams. This resulted in the consolidation of approximately 300 lower ASP subsidiaries into their parent entity. Our team also prioritized new customer acquisition, specifically targeting opportunities that land with higher value and greater expansion durability.

Now I'll review the income statement in more detail. Gross profit in the quarter was $121.6 million, representing a gross margin of 83.7% versus 83.2% in the year ago period. We remain focused on cloud hosting cost optimization as we anticipate a larger share of our revenues being generated from the cloud. Given our expected increase in cloud revenue contribution to total revenue, we estimate annual gross margins to be in the range of 82% and 83% in 2026.

Operating expenses in the fourth quarter were $95.8 million, equaling 66% of revenues. This compares to $75.6 million or 65% of revenues in the year ago period. We remain focused on expense discipline while we continue to invest in strategic initiatives. Our operating profit in Q4 was $25.7 million or an operating margin of 17.7% compared to $20.9 million, an 18% operating margin in the fourth quarter of 2024.

For the full year 2025, we delivered non-GAAP earnings per share of $0.82, a 26% increase year-over-year, assuming approximately 122 million weighted average diluted shares. This compares to $0.65 in the prior year and 115 million weighted average diluted shares. Cash flow from operations equaled $50.7 million in the fourth quarter. After taking into consideration CapEx requirements, our free cash flow reached $49.9 million or 34% margin compared to $48.4 million and 42% margin in the year ago period. For the full year 2025, we generated $145.7 million in operating cash flow and $142.2 million in free cash flow, a 27% margin.

Now turning to the balance sheet. We ended 2025 with $704 million in cash and short-term investments compared to $522 million at the end of 2024. As of December 31, 2025, our RPO totaled $566 million, a 40% increase year-over-year. This performance highlights the successful execution of our go-to-market strategy as customers continue to make larger multiyear commitments to our DevSecOps offering.

And now let's turn to our outlook and guidance for the first quarter and full year of 2026. As we enter 2026, we are encouraged by the strength in our pipeline and a stabilized purchasing environment. While we are monitoring the increased cloud usage and early AI workload trends that could result in a tailwind for JFrog, our guidance philosophy will remain unchanged as we continue to derisk our largest deals due to timing uncertainties and any benefit from cloud usage above contractual commitments.

Our outlook reflects growing contributions from our JFrog Security Core products, ongoing adoption of our full platform and cloud growth driven from higher annual customer commitments. We estimate full year 2026 baseline cloud growth to be in the range of 30% to 32%. Given the anticipated contribution from our security core and baseline and cloud growth assumptions, we expect our net dollar retention rate to be 117% for 2026.

Turning to operating expenses. We will remain focused on investing in innovation across our platform and reinforcing JFrog's role as a system of record for all binaries and AI models. The weakening U.S. dollar against global currencies has created a year-over-year headwind for our operating expenses and is reflected in our operating profit guidance. We remain committed to a disciplined spending philosophy and are confident in our ability to manage expenses in line with prior execution.

For Q1, we anticipate revenues to be in the range of $146 million and $148 million with non-GAAP operating profit anticipated to be between $25 million and $26 million and non-GAAP earnings per diluted share of $0.20 to $0.22, assuming a share count of approximately 127 million shares. For the full year of 2026, we anticipate a revenue range of $623 million to $628 million, representing 17.5% year-over-year growth at the midpoint. Non-GAAP operating income is expected to be between $106 million and $108 million and non-GAAP diluted earnings per share of $0.88 to $0.92, assuming a share count of approximately 128 million shares.

Now I'll turn the call back to Shlomi for some closing remarks before we take your questions.

Thank you, Ed. In today's market, nearly every company is looking to capitalize on AI. But in 2025, when the world's leading AI native company selected JFrog as the core infrastructure for the software supply chain, it was clear validation of our strategy. They became JFrog customers, not only for our industry-leading platform, but because we have become the trusted system of record for binaries, the foundational asset powering modern software delivery.

The adoption of our platform indicates that our road map is strongly aligned with where the industry is headed. As [ by ] engineering and coding agents accelerate software creation enterprises are facing a massive surge of binaries that must be managed, automated, secured and governed across the software supply chain. This dynamic clearly differentiates JFrog as the trusted enterprise platform in the age of AI.

2025 is now in the rearview and I'm proud to say that what we committed to our partners, customers and shareholders was consistently delivered. Despite significant macroeconomic and geopolitical challenges the JFrog team roles to new heights. This period has also carried deep personal weight for many in our Israeli team. After years of paying, finally, all hostages are back at home. We move forward with renewed hope for peace, stability and a better future for the region and the world.

To my frogs, thank you. 2025 was a challenging year of uncertainty, yet a remarkable one for us. You didn't just live through it, you want it. And for that, I'm proud and grateful. As we step into 2026, we remain committed to quality growth, responsible investments, expanding our solutions to meet emerging needs and bringing us all on lip closer to realizing our liquid software vision, a world where every software flow is effortless.

And with that, thank you for joining our call and may the frog be with you. Operator, we are now open to take questions.

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

This is [ Oscar Savera ] on for Sanjay. Thank you for taking my question and congrats on the great results. Really nice to see the continued growth on the cloud. I wanted to touch maybe on the customer count. I understand the intentionality of you're focusing more on the higher propensity to spend larger customers, and we see that on the 100,000 customer add. But -- how should we think about that going forward? And how far along we are in that transition? Should we continue to expect that count to come down while the larger customers go up? Any additional color there would be helpful.

Yes. Thank you. This is Shlomi. I'll take this first question. So we are focused on our strategy, building for the enterprise as we said 3 years ago, and you can see the results on the customers over $1 million to significant growth of [ their ]. You can see the results on the customers that are spending over $100,000 R&D increasing spending over there. This is our strategy. This is where our go-to-market is focused on.

And sometimes, it means that we will have to let go to low ASP customers and to be focused on our customers. We also noted that we've made our internal consolidation to avoid friction in our go-to-market team and to focus our team on the big enterprise opportunities. And that by itself was at approximately 300 logos.

And also, I should note that this includes the geographies regulation like churn in China, Russia and other regulatory decisions. So we are very pleased with the stability and the growth. But more important than that, the fact that the customers that we brought in, hundreds of new customers that we brought into our portfolio are spending much more on ASP growing faster and adopting not just DevOps, but also DevOps Security and other services in the cloud. So in the bottom line, it's very much aligned with our strategy and reflects our commitment as we enter deal.

Your next question comes from the line of Radi Sultan with UBS.

My question is for Shlomi. Like it seems like every quarter we're seeing in other security incident, you had the NPM attacks. And Q3 and Q4, you mentioned another one. I mean, are these becoming less like onetime events and more just like structural growth drivers? Like are we just permanently in a higher threat environment and maybe we should be penciling in a more consistent contribution from these types of incidents going forward? Or how should we think about sort of the long-term contribution from just the higher threat environment we're in today?

Radi, this is a great call out of what's happening today. There are more and more attacks over the software supply chain. Attacking the software supply chain by hackers means that they are going after the software packages, the binaries. It started 3 years ago, this trend with [indiscernible], then python, then and now NPM and [ Chaiklin ] in three different incidents, every customer, every enterprise, every software organization understand now that software supply chain is not protected if the software packages if the binaries are not protected.

Now putting aside this threat that is kind of floating over every company now code agents become far more faster in how they create code and therefore, create more binaries, using more open source software packages. So this threat and this trend is something that we anticipate that will only grow. And therefore, we build our JFrog security suite to tackle this threat.

Your next question comes from the line of Michael Cikos with Needham.

Some very encouraging data points and appreciate especially the RPO growth and ARR contribution when thinking about security as far as demonstrating that this truly is a platform and the attach you're seeing.

One of the things I wanted to come back to, and we've been feeling on our side, with the heightened threat environment and some of these hacks, which unfortunately feel a little bit more commonplace now. Is there a general rule of thumb where we would be able to draw parallels between a hack of the NPM size would -- what that would equate to in revenue. Is there any broad parallels we could draw from? Or is it really just happenstance, depending on what part of the ecosystem is being hacked -- anything there would be beneficial. .

Thank you, Mike. It's Shlomi. I'll take this one. The only responsible way to look at the potential revenue growth is to look at the amount of enterprise customers in our portfolio that still didn't adopt the JFrog security. And while we are growing by 100 every year and the ASP is kind of growing significantly, there is still a lot of room to grow within our portfolio.

JFrog Curation in the last two quarters is exploding mainly because of the fact that the threat is real. The scale is required and you have to move fast and big companies are not taking any risks, especially in today's changing environment. The second thing that I should note, Mike, is that most of our new enterprise lenders, what I referred to before when we spoke about the new logos are already adopting JFrog platform with security.

So there are three avenues of growth. Number one, customers that are still not using JFrog Security and will, hopefully. Number two is with new logos that are joining enterprise logos. Again, we are not buying logos. We are going after the enterprise -- enterprise logos that are subscribing from the get-go with security. And number three is the number of projects and the scale within the companies that already adopted JFrog Security, and we still have room to go over there.

Your next question comes from the line of Miller Jump with Truist.

I'll echo my congrats for the acceleration this year. So Shlomi, I wanted to come back to your comment about the tsunam of binaries accelerated by AI. For the customers that have increased their commitments, are you still seeing them grow and consume beyond these re-up commitment levels? And can you give any more color around the consumption change for customers that have leaned into coding agent deployments versus the broader customer base?

Great question, Miller. This is going to grow, and we call it tsunami because this is what we see when we are reviewing the data on a weekly basis, we see how much more software artifacts are being created, how many binaries are being compiled and that's mainly because of the fact that every developer now became a super developer. And we hear how well the code agents are doing the Anthropic, the Open AI, the Gemini AI, these agents are faster than developers. They are powering developers and therefore, creating more binaries. When you create more binaries, you need a single source of truth to go to whether you are an agent or a developer. And this is where JFrog into the picture.

To secure, to govern, to build a trusted software supply chain for you and for your agents. So think about the number of developers, call it, millions of developers in the world today, powered by millions of agents and therefore, the result would be many more binaries. And hopefully, we will see it also in the scale of JFrog.

Your next question comes from the line of Mark Cash with Raymond James.

Yes. I think just on me, if I could ask, with all the [indiscernible] tools, we're just kind of talking about there, creating more software, you have new bottlenecks emerging in the software pipeline. So just kind of curious if you're thinking about AppTrust in 2026 being maybe the next catalyst to alleviate governance and regulatory pressure in adopting AI agentic tools organizations are going through right now?

Most of the very important highlights of our road map throughout the year since we established JFrog, came from a real enterprise need. And our customers told us last year, listen, we are faster now with DevOps and we are more secured with DevSecOps.

But there is a new bottleneck, which is governance, and this is going to be even more painful when it's not only humans that are building code and compile to binaries, but also agents. So AppTrust is tackling that exactly. We are addressing this pain of automated governance with all the evidence, again, storing JFrog Artifactory is a system of record. And whether this code was generated by a developer or by a code agent the governance and the regulation before the binary go to production will come out from a single source of tools, and this is what AppTrust is addressing to solve.

Your next question comes from the line of [ Zane Meehan ] with KeyBanc Capital Markets.

[indiscernible] I'm on for Jason Celino today. I wanted to ask about your AI native customers. In the past, you've talked about having three of the top 5 AI natives, which is obviously great. But for the other two or maybe any of the adjacent players that aren't using JFrog, what would they be using? And how confident are you that you can eventually win them as well?

This is -- every company now want to call it an AI-native company. When I'm speaking about the big players, I'm speaking about those that are moving the market and calling the trend and setting the standard. And they chose JFrog to be their power grid, to be the solution that will run their software supply chain. They build around Artifactory as their system of record, and they are using JFrog Xray to scan that and to collaborate and integrate with the ecosystem.

The other two, I don't know if it's two or more. I don't know what they are using. But from what we hear, some of them are evaluating tools and some of them are building their own tools. slowly, as we are going over release that our go-to-market team built together throughout the year, we're bringing them one by one. So I hope that every quarter, I will report another one that chose JFrog not only as a partner or an ecosystem player but also as a customer.

Your next question comes from the line of Kingsley Crane with Canaccord.

Shlomi, you've had JFrog Fly in beta a bit now. You've had MCP server integrations out for a bit now. I'm just curious what kinds of usage trends you're seeing with either Fly or the MCP server and just what that tells you about how customers workflows are changing and then how you're positioning around that?

Yes. Well, this is, Kingsley, a very important move that we've done during the year we understand, and I call it out on the script, we understand that the business to business is how we build JFrog [indiscernible] in the future, it will be a business to agent market.

And if I want agents to become my person of my customers, then I need to give them access to interact with my platform. This is where the MCP server comes in. but it's even greater than that. MCP itself is also a binary. So we start to see more and more of our customers using JFrog Artifactory, as the single source of tools for all MCPs in the market. So it's not only that agents will use JFrog as the best kind of area to pick up binaries from but also other players will use to place the MCP servers in.

Now if you do that, if people work with one system of record, then consistency happens, secured solution governed software supply chain before the release production is important. And with the MCP server, agents can now interact with the JFrog platform as well.

Your next question comes from the line of Brian Essex with JPMorgan.

Maybe one for Ed. A bit of an acceleration in both sales and marketing and R&D in the quarter. Would love to get maybe a framework of how you're thinking about investment on both sides of those and how you expect that to materialize through fiscal '26? And maybe as part of that, maybe if we get the FX impact both for the quarter as well as what you're thinking about contemplating in your guidance.

Sure. Let me first start with Q4 in the sales and marketing. What you're seeing is end of the year bonuses that flow through in our expenses, and that's what's actually happening there in Q4. But as we step forward into 2026, the reason we called out the weakening U.S. dollar, as you know, more than 50% of our head count is distributed globally.

So we wanted to call that out. We have a very strong hedging program. So most of that risk is already covered and it's captured in our operating margin guide already. We also have a very disciplined operational execution in terms of maintaining expenses. So we continue to invest, but we continue to make sure we're doing it smart. Anything that we see in terms of the potential outperformance on the top line, we would still consider a meaningful portion of that to flow into operating margins.

Your next question comes from the line of Shrenik Kothari with Baird.

Thanks a lot. Apologies for my bad voice. But just beyond the foundational models that you touched upon, we are now a secure model registry for NVIDIA AI factory for hugging phase, arguably two of the most strategic on-ramps to model consumption. Just curious, Shlomi, how are these top-down partnerships driving new logos for you, number one? And two, when it comes to monetizing this unique position where are they on the adoption curve for the additive legos that you've described in terms of AI catalog adoption then app trust and agentic to remediation.

Yes, Shrenik, we managed to expand our platform so successfully with security and with MLOps and with DevOps and distribution that every time that we have another partnership powering the community, it's kind of spread across all of these elements of the platform. Specifically with NVIDIA, beyond the fact that it validates JFrog as the single source of tooth for all AI tools and for models, it's also aiming to the enterprise, as we mentioned before.

This is our strategy. NVIDIA is being used by the enterprise. They are selling GPUs. We are providing them with a secured area for models interaction. And this supports the enterprise. Hugging Face, however, is a different scenario. Hugging Face can be a top of funnel for JFrog because it's the open source hub, but it's also support most of our customers that are using Hugging Face as a local repository for the models that they bring from outside. So they will have a secured environment.

Now when we are looking forward into 2026, the two integrated to fail philosophy is something that we are not only collecting feedback from the market of what they would like to see, but also how can we serve them better not just the human developers but also the agents. How can we serve better and Anthropic Opus 4.6, how can we serve better an open AI agent because these are the new tools that are now using JFrog as their system of [ ego ].

Your next question comes from the line of Ittai Kidron with Oppenheimer.

Congrats guys. Real nice solid finish for the year. I guess I got a couple of questions, one for each of you, for you, Ed. Nice to see the progress there on the security, and I appreciate the disclosure there. If one would do a back of the envelope, and strip out your security out of your business, what would that imply that the core business, excluding security is growing? Is there acceleration there, leveling off deceleration? Any color there would be great.

And then for you, Shlomi, I would love to think about '26. What from your perspective, are the greatest risks to your business right now? And maybe clearly, with what's going on in the world right now in the context of AI and how they could potentially do everything, including our laundry. I guess my question is to you, in what way do you view -- in what way could AI be a risk to your business?

I'll start here, and then Shlomi can finish. We gave you the results at the end of the year for the revenue contribution that's coming from security. And it's still a relatively small piece in terms of that contribution to the top line. We go to the market as a platform, selling together both Artifactory and Security together. So we don't want to carve that out. The results that we show you on the top line were impressive. That included security. We're very happy with the direction that it's going. The sales team are instructed to sell together the solution and the offering, which we believe when you go to the market together with Artifactory in security, you secure your software supply chain at scale, what is needed for today's customer.

Yes. I'll answer the risk question, Ittai. It we are all standing at the edge of a cliff. Some people will tell you that -- we are going into a word of productivity. You mentioned laundry. Some people gave us other examples. And some people will tell you that we are all going to be replaced by robots.

I think that for the business, and this is my job as the CEO is to make sure that we are focused. That JFrog is differentiated and bring value to our enterprise customers. And what we see now is that we keep building for the future with them. Just a few days ago, Elon Musk tweeted that code is going to replace the only outcome will be binary. It's reinforce again what we keep saying for the last 15 years, binary is the primary asset. This is where we should be focused. And my biggest risk is that I will get confused and start to follow trends and not serve my enterprise with what we do best, being the system of record.

Your next question comes from the line of Jason Ader with William Blair.

I hope I could speak to it as well. One quick one for Ed. One for Shlomi. For Ed, trying to understand the $800 million FY '27 revenue target. Is that just kind of aspirational at this point just because it would require a pretty massive revenue acceleration in '27?

And then for Shlomi, I know you guys price Artifactory on capacity. So it should logically follow that the tsunami binaries that you called out should accelerate growth in your core solution. Are we seeing that at all yet? Or is that still to come? And when do you think that could come?

Mike, Jason, I'm sorry, I'll start with the question regarding the $800 million on the long-term model. Listen, at the end of the day, we're focused on delivering and executing in 2026. That's what we're guiding to at this point. When we look at the results over the last 3 years, we're on track with that number. It does not reflect our conservative or responsible guidance philosophy, but the focus right now is on delivering in 2026.

I'll take it from here and speaking about conservatism. We see, as we mentioned in the call, we see a rise on the AI-related binaries being consumed by our customers. We manage PyPI, MPM, Docker, Conda and others. But as Jason, we are not kind of promising metrics. We are being very, very disciplined with our guidance. And we commit -- when the customers commit, then we guide the market. We guide you by commitments and not by usage. So yes, we see spikes and yes, it's an uplift to our performance, but we will not guide based on this data transfer until it will follow a customer commitment and customer commitment usually comes when usage stabilized and not based on trends.

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

Great. Shlomi, has demand for Curation continue to accelerate post the second NPM attack in late November. Does this pull any deals forward into Q4? And how much is the pipeline for that up year-over-year?

Yes. So Curation, listen, we are not celebrating these attackers coming after our customers. But obviously, we benefit from it because there was a very clear value that Curation both -- Curation is a tool that is out in the market for almost 2 years, and it's mature and it was ready, and it was ready not only in terms of the risk but also in terms of the scale not only that we performed amazingly in Q4 because of the NPM incident, but also build the pipeline moving forward, as I mentioned in the call.

Your next question comes from the line of Jonathan Ruykhaver with Cantor.

Yes. Congrats on a fantastic 2025 guys. Shlomi, I would love to hear more details around what you're seeing in terms of the potential convergence of the DevOps tool chain to address MLOps and specifically, are they seeing the benefits of a -- or do they understand the benefits of a potential unified pipeline, particularly around security in governance. And I guess, lastly, as a part of that, just what are you seeing in terms of adoption trends for JFrog ML and your expectations for that in 2026?

Yes. JFrog ML was included in -- I think it was Q2 or Q3 this year in our platform. And some of our customers are already using JFrog ML to manage their full models life cycle. We treat model as a package model is yet another binary. So by providing them with these capabilities, we are reinforcing the fact that JFrog is the central trusted source of tooth not only for the legacy artifacts, but also to the new artifacts, which are modeled. I think that this will be evolved as the market evolves.

MLOps will not stay as it used to be before the days of LLMs and what we see now is that code agents are also starting to kind of interact with the JFrog platform for any push and pull of binaries. So overall, it's growing, it's evolving. This entire landscape is changing, and we are tracking it, and we will keep you posted on it.

Your next question comes from the line of Eamon Coughlin with Barclays.

Congrats on the continued execution. I just wanted to go back to the MLOps motion and how to think about that opportunity. Can you help us understand that consumption profile, like what that would look like for an LLM maybe compared to a traditional binary? And then what makes JFrog well positioned to be the [indiscernible] for these enterprises?

Yes. The MLOps solution is actually going after providing the CICD experience for models. It started with small models and then the world in the last year evolved and so is our tools and our platform. The idea around it in terms of consumption is that if you treat model as a binary, which it is, then we should see more data transfer and we should see more storage. I think that there is a better potential for monetizing on storage because models by definition, are bigger binaries than the others. So we see -- since our pricing model is a consumption-based model, if we drag models to use the MLOps capabilities of JFrog together with the security and the storage you should see our consumption going high. And therefore, the commitment of the customers will go higher. And this is something that we are tracking closely and looking forward to see the results.

Your last question comes from the line of Lucky Schreiner with D.A. Davidson.

Perfect. Congrats on the strong results here. If I look at cloud revenues, they seemed a little bit maybe the more stable quarter-to-quarter this year compared to historically. Maybe just -- on the seasonality side, as cloud revenue becomes a bigger portion of revenue, should we expect a little bit more stable seasonality? Or is it still the same as you target large enterprises in these larger deals no change on that front?

Well, when you think about the cloud revenue, you need to think about three different things. Our first is our guidance velocity, which we derisk for those largest deals as well as usage over minimum commitments, including emerging AI trends, that is what creates variability on a quarter-over-quarter basis, and that is excluded from there. And as you mentioned, it's also now a greater portion of our revenues. But when you think about sequential growth today, it would be linear until we start to layer in any usage if that potential continues or these large deal wins.

There are no further questions at this time. I will now turn the call back to Shlomi for closing remarks.

Thank you, everyone. 2025 was a great year. We are looking forward to 2026, and we are very excited about the changes in the market and the new players in the market that we are looking forward to collaborate with may the frog be with you and maybe you have a wonderful Valentine's Day.

This concludes today's call. Thank you for attending. You may now disconnect.

Hi, everyone. Welcome to the Barclays 2025 TMT Conference. I'm Eamon Coughlin, software research analyst here at Barclays. Very excited to have Jeff Schreiner, Head of Investor Relations at JFrog. Thanks for being here, Jeff.

Thank you. Thank you for having us.

I guess just to set the stage, JFrog has been quite an incredible story for 2025. I guess set the stage for anyone in the room that maybe is new to the story, like how has the year been? What are the key takeaways for you? What have been the key drivers for growth acceleration throughout the year? Anything maybe you'd point out?

Yes. So I think as it relates to JFrog and '25, it certainly, as it relates to the stock price and certainly the execution we've been able to deliver has been quite strong. And those are all great things that coincide with one another. I think that when you look at '25, it's been somewhat unique for us as relative to the last few years, where within our cloud business, there's 2 ways in which we can really grow. There's either migration or consumption.

And over the last 2 years in '23 and '24, it was very heavily weighted to winning a few big deals to make the year related to very large customer migration activity. Whereas in '25, what we've seen is we've seen much more of a consumption-driven contribution to revenue. There we go. Technical aspects solved. All right. Now you guys can hear me better.

But -- so I think that the drivers this year have been different than the drivers we've seen in the past. Migrations have played some role, but not the role of the magnitude that we saw, let's say, in Q3 of last year, where the 3 biggest deals in the company's history were signed and then the full value of those recognized in Q4.

This year has been a year of increasing consumption as our customers look and experiment with AI and ML and also are starting in our view to experiment with the use of coding assistant tools, which then use more of the traditional type of open source packages. And the fact that we're seeing a strong interest and adoption in a lot of our security offerings.

And as we've moved through the year, and I'm sure we'll talk about some incidents that happened in the industry. But as of a recent industry attack on a repository has generated a significant amount of interest in one of our security products to protect those organizations.

I guess just unpacking there's a lot to unpack there. But as you think about like the right pricing model for some of the influx of demand that you're seeing from your customer base, is consumption the right pricing model? Is it more -- I don't know if there's like a server model that you may move to across your platform. But I guess I guess how do you think about that going forward?

Yes. So today, JFrog monetize it in 2 ways. So we have self-managed or self-hosted in which you're having a team of IT professionals manage your software development organization, which is probably on the cloud, but you're managing that or there's our cloud offering in which you're turning that management over to JFrog and putting that stuff into our cloud.

And as it relates to the monetization of each of those methodologies in the cloud, it's based on data transfer or consumption of the data package that you commit to JFrog in which you get a pricing benefit for doing so. In self-hosted, it's based on an incremental server, generally project-based to where I take that server. I've now increased my usage or increased the development team, so I need another new server.

In the world of AI and ML, we found that, obviously, a seat-based model may not work. We've seen the head count reductions talked about. Obviously, there's a lot of bug a boo concern. I think this year for some of those more exposed to seat-based models. We know the consumption works for JFrog in the sense that if you're now using JFrog and your cloud consumption contract and move large language models that can be the 10x the size of packages you move before, you're quickly starting to consume at a higher rate, the amount of data consumption that you've committed to JFrog.

So I think it's begun a discussion, at least an initial sense, there's nothing that JFrog is going to lead, but I think in an initial sense, there may be a discussion amongst the titans of the industry, the hyperscalers that is there another way, and we could go 100 different variables with what that may be of monetizing the AI/ML world that may not be consumption based. But that's to be determined. It's something that will be discussed. And I think what JFrog will do is look to how the industry handles any changes and try to tailor that model to our own model.

Absolutely. I think over time, especially over the last 3 quarters, you've seen that increased commitment from some of the overages from consumption. Can you just walk through maybe how you saw those overages in 1Q, 2Q, 3Q? How they play throughout the year and how they might impact the fourth quarter?

Sure. So as you noted, we get the benefit typically if you are a customer, you come to us and the benefit to JFrog versus working with other maybe consumption model guys out there that are pure consumption based. You're coming to us and you're committing to a usage level in which that we will then give you a better pricing mechanism for the more consumption you commit to.

And of that commitment, you are ratably recognizing that on a, let's say, if it's a 12-month contract to be quite simple to use round numbers for everyone here, if I have a $1.2 million contract with customer X, I'm ratably recognizing that at about $100,000 a month until they use above that commitment. Now this use it or lose it mechanism that Eamon's is talking about tends to be the fact that I'm committing to you annually for, let's say, 10 petabytes. But then that's recognized on a user it or lose it basis each month.

And so to the extent that I start to accelerate my usage and go over commitment, I start to pay a penalty rate that can be somewhere anywhere between 20% to 25% higher than the negotiated rate that you may have with JFrog and what that leads to is as you're building into that, we certainly try to approach you and engage with you to say, hey, look, your usage is starting to climb, good for you. We hope things are working well with the organization. But you may want to look at upping your commitment with JFrog.

And so you can see aspects that happened where Q1, we saw strong usage that we felt was really driven from experimentation of AI and ML. The reason we say that is we saw a strong increase off a small base in packages such as Hugging Face and PyPi and Conda. Q2, the usage kind of flattened out, but the benefit was that we captured some of that over usage into higher committed contracts from some of those customers.

In Q3, we saw an influx now again in usage and the revenue generation that it delivered, but we haven't yet been able to reflect the benefit of what we may be able to do in signing higher committed contracts which once those are signed, then they have become something in our CRPO, whereas the overusage is just recognized in revenue and is not something that's captured in CRPO.

So a little bit of a lag?

It can be. Yes. It can be a quarter or 2 quarters. It also ties very well with where you are in your renewal cycle. With the resources we have, we typically talk about the fact that we go after the guys that are 3, 6, maybe 9 months at the longest away from their renewal who may be over users because they are most likely the ones to be most acquiesce to say, yes, okay, I see I'm overusing. I have a vision that I'm going to continue using at this level or higher. Let's go ahead and renegotiate.

Let's take a step back at something you said earlier. Just understanding like why JFrog has a right to win in security. And then maybe some of the recent news with the NPM attack. Maybe just dive into the first question and then some of the takeaways that you have from the recent NPM attack.

Sure. So security was something that is new to JFrog. Last year was the first year that we disclosed the contribution from security, and it was really the first year that we saw really meaningful acceleration of that product. And that was a product brought about by an acquisition that was done in 2021. Then initially when done, I think the Street had looked at it kind of skeptically saying, at that time, hey, you're a DevOps company, they're security companies. Why don't you just stay in DevOps and didn't see the vision we saw that eventually the world was going to be much more based on core platforms and core assets that they needed to be secured.

And that's the reason we feel the reason that we have the right to win, let's say, in security and as it relates really to the security of the software supply chain and binaries is because we are the core manager of that asset. It's what we do, it's all we do. And so when you are a CISO and you've brought on all these disparate point solutions to possibly protect your software supply chain in your development organization. You've got many disparate databases that are all signaling different vulnerabilities for you. You're almost frozen in place as to what to do.

With JFrog, what you're able to do is consolidate the 7 of those tools within the technologies we offer, into 1 solution. And if you combine it with our relationship with GitHub and utilize their GitHub advanced security for, let's say, source code and static code analysis, you basically can take 7 to 10 tools consolidated it to 2, but you as the customer looks as if I'm using only 1 tool. There's 1 pane of glass in which I can interface all and receive vulnerabilities and remediate those vulnerabilities.

So I think the uniqueness that we have bought and shown that we do have a right to win after the numbers we disclosed in '24, which was 3% of revenue, 5% of ARR, 12% of RPO. And we'll update those again here as we report 2025. We've shown that there is a real interest for the products that we do.

And I think alluding into and transferring to your question about NPM and the impact that securities had for us there. There's been a recent coordinated attack, a very malicious attack that has had various multiple attacks over the last 2 months that began in August 28 of this year.

And that was a hack of the NPM package, which is if you look at our website, at jfrog.com, we have a report called the State of the Union. I think we released it about midyear and we tell you in there a lot about what's going on with the binaries and we show you what are the most used programming languages amongst our customers and Artifactory. And NPM is 1 of the top 3 open source programming languages.

So this group who must be very sophisticated, how they've developed and orchestrated this attack went after one of the leading packages out there utilized by enterprises and organizations to build software today. And they did so that created a very fear-based or fear-driven want for customers to look at one of our security products in general, and that product is curation. And the reason that is, is because curation is essentially a firewall for your software development organization.

It's a centralized policy and where I, as a CISO, can set a centralized policy for that product to say, I will only interface with these repositories and these packages. And so curation is actively managing and scanning those repositories and looking for any discrepancies or vulnerabilities that may be introduced into those to protect the organization.

And since this NPM event which, again, it started out as more of a basic attack where they were targeting specific repositories and looking just to steel to moving now towards the one that happened in the end of closer to Thanksgiving adjusting that attack to be random to any repository. And if there's nothing to steal, wipe everything out, it's got a fear-driven demand in our pipeline growing for our curation product.

Now that being said, that demand is certainly growing, and we're excited about that, but there's a need for our customers to find budget. And I think that's the challenge that we're hearing as a pushback from our customers. I really want it but I need to find budget because many of these customers had committed to their organization and this was a product that they may, in fact, deploy in the second half of '26.

We, in fact, had a customer, a large financial services company that was scheduled to deploy in Q2 or Q3 of next year, closed in 2 weeks in Q3 because of this NPM event and the need that they felt the curation brought to them. And so I would say, when you look at our security product and where it's gone, since introduction, deployment and pipeline has probably been 50-50 between our advanced security which is kind of protecting the inside of the castle and looking for good citizens gone bad and curation, the firewall or the wall around the castle, keeping the salvages out, it is about 50-50 in terms of deployment of pipeline. I think post NPM in September, it's substantially much more weighted to curation today.

And that's incredible. And when you're thinking about some of these customer base, like are they using security today for their binary management? Or are they not using any tool at all? Or like...

Everyone has security.

Are you displacing another tool that maybe is not strong enough to handle the NPM attack.

As it relates to NPM and that attack, curation has no alternative. There was not an alternative to curation. It was a product that we developed at the behest of some of our customers, because why was it created originally. It was created because I, as an organization did binaries in 1 of 2 ways. I either brought in everything and scanned them through x-ray and knew what was malicious and what wasn't malicious and then went from there or in some of these large financial institutions, I allowed nothing in.

And the developer had to make an application to have this package approved and hope it was approved by the time the software that he was building was done. And they came to us and said, hey, it would be great if we could have some type of centralized control over what packages are brought in organization. And the other thing I'm kind of alluding to you guys that you should watch curation for is the fact that as code quality starts to get better from these coding assistants.

And I think there was some impact from that, I can't point to it or show you what code was created by a machine or a human. But I think we know publicly some of our customers are starting to experiment in utilizing these coating assistance. If I want to start turning the machines more free in my organization, curation is certainly a step I need to go in because at that point then, I have curation.

It's integrated with my IDE station, GitHub. It's integrated with AI catalog in Artifactory. And I know that to the extent the machine is building, it's only going to be building with packages that I allowed into the organization.

Interesting. Yes. We recently hosted a call with the CEO of Sonar, code quality tool. Is that -- I mean, obviously, there's a key difference between like code security and code quality. Would that ever be an interesting aspect of expansion for JFrog?

We -- one of the technologies that we offer in our advanced security is static code analysis, the SaaS technology, which is generally a source code based security. That being said, I think that our chops are still much more binary based. And in the relationship we have with GitHub, we're not conceding anything. But certainly, when 80% of our customers use GitHub and JFrog the easier lift, and I don't have -- Ed doesn't have a [ contra ] account to basically make this as a revenue equation, right? But the easier lift is to say, hey, Jose, so replace all these tools with GitHub and JFrog security and let GitHub do what it does really well in source code and let JFrog do what it does really well in binaries.

And I think you heard something similar to that out of GitHub universe over the last few weeks where the GitHub employees, by no means are waiving any white flag publicly to say that we've conceded binaries to JFrog. But I think there was rumblings if you were there and you were attending and speaking to those individuals that we recognize that JFrog is, in fact, a binary expert and does those very well, which allows us to turn focus on the business that we do, which is source code.

Yes. No, actually, I was there, too, and that's the thing I heard. I guess just going back to swap up. There was a time of announcements at the conference, like definitely one of the more exciting conferences and swamp ups in JFrog's history.

Maybe can you just walk through some of those announcements. So JFrog Fly, I thought was really interesting and diving to the SMB customer base, which is a little bit absent from enterprise Q historically and then maybe AppTrust and then AI catalog, maybe what could that drive in 2026?

Obviously, still early on, in this customer journey with some of these products, but how do we think about some of the motions with each of those products?

Yes. No, thanks for bringing that up. I mean, I think it was a very successful swamp up. I mean let's start with Fly, where there's been a lot of questions about it. But there's -- I think that's probably the furthest from revenue contribution. And what is Fly? Why is something that we're creating so that we can better understand software development in the Agentic world. And what we mean by that is that you could now -- binaries are for companies that have very complex software development organization that become a pain point.

But you could become a very -- excuse me, got a little frog on my throat. But you could very much become a complex development organization as an Agentic firm, a firm that is a start-up last year with 50 guys. That could be a startup now with 2 guys in 48 machines in the future in the world of Agentic AI. And what we want to understand is how they're utilizing interfacing with Artifactory, open source packages, are they complex in nature but only using a few programming languages.

So an Artifactory light that could be tied with Fly is the right method forward. I think what Fly is going to do for us is to have real-world deployments, real-world kind of knowledge about how to handle some of these agenetic capabilities and then bring them into the enterprise and say, look, we've already done this with this customer over here. In what you're talking about trying to do, let us bring in help it to you and bring it into the Enterprise+ program.

Maybe later, a few years down the road, there may be some way to monetize that as an add-on or something additional to the JFrog platform. I think that, as you know, the biggest thing that got back to me even and not being able to swap up this year, but was AppTrust and that after that presentation, the customer feedback that many of you heard in speaking with our customers about that DevGovOps product, right?

It's now bringing the operations organization into the JFrog platform and putting inside the development process the checks to know that each of these gates have in fact been completed and that there's a digital record of that, not the way it's done today manually through Excel or yes, I think it's good enough. So even let me sign that DocuSign and say it's good enough. As it relates to AI catalog, that's -- you heard me talk about that earlier. Why is that important? That's going to be very important for large language models and software development in general because that's your Wikipedia of binaries.

So that's an area I talked about where you integrate curation and co-piloted maybe that could be other coding assistance in the future. And that's integrated into your CI/CD flow. You're able to go into a catalog and say, okay, here's my build. What were the last 5 builds. Okay, this is the package that I was using. And so now going to Artifactory and grab that package.

So basically, AI catalog is the world exploding in terms of the volume of packages used and code creation is going to give you a constant evolving database of what is going on in the world of open source packages. So as you do move to more of a machine-generated type cogeneration, this machine will have a database of which to go back and check. What is this package? What was it used for? How have we used it in the past? Is that applicable to the application I'm building today?

To my knowledge, there's no other company innovating on those types of products today. Like how should we think about the initial pipeline generation that you've seen? And then particularly for AppTrust and AI catalog, how should we think about the pricing and monetization aspect of that?

Yes, good question. We'd love your feedback on that because those are things that we're still working on right now, better trying to understand how we, in fact, monetize AppTrust to generate this groundswell that seems to be growing post-swap up for this product given there really wasn't a product such as this AppTrust product.

I don't know that it's going to be a major contributor to '26. I don't think AI catalog is either. I still see '26 being very similar to the fact that as long as we continue to execute, I think the drivers remain somewhat constant. We are in a very fast-changing world. things could evolve in a quarter or 2. But as we sit here today, I think it's going to be much more consumption-driven if we're moving more to co-development through coding assistance and the adoption of our security.

Got it. Yes. I mean I think as you think about the future, maybe the next 18 to 24 months, how might that growth levers look like compared to maybe the first 3 months of this year. They were primarily driven by, I guess, security commitment above -- use above your commitment and then maybe some larger deals. Like any change in momentum that you that you expect for?

No. I think we chug along, and I came up with this, and it may be just basic, but it's our LEGO strategy, right, where you have the base platform as your core LEGO, and now you're adding on to those LEGOs to add not only new incremental revenue opportunities for JFrog, but value to the customer and incremental value to retention for us as well. So I can add on security. I can add on AppTrust. I may take MLOps out of the subscription at some point in the future and monetize that separate as an add-on. I might do that with Fly.

So I think what we're trying to do is constantly evolve and become more to our customers than just the basic infrastructure level. The way that you retain and keep customers and maintain a gross retention rate of 97% like we do, is you are that core plumbing but you want to make sure there's no reason that they ever want to look to replace you because you continue to add value each year with new technology that you add to that platform.

Is that platform has been like a key driver of the strong retention that you're not just the binary management product security aspect is scanning, there's vulnerability management. Like is that the key aspect, the architectural moat that I think we talk a lot with JFrog compared to something like a lab package or even like the package of offering that GitHub has or even like a Sonatype.

I think it is. And the reason I say that is, even you know us, I mean, three years ago, it was a totally different sale process. We were selling new Artifactory into a group, and that's how the industry kind of was purchasing things to where now we had seen this evolving nature turning to a platform of the key aspects of software delivery, source code, binaries, observability run time.

And I think now the difference in buying and you see it. In Q3, we landed a new oil and gas logo customer that was a 7-figure land that is a whole new world for us. And why was that? Because when customers start to buy on platforms, they already kind of have an idea of who the leaders are in each of those sectors.

And they know that they're going to likely need to commit and use those leaders and they're willing to commit at a bigger land than they would have when it was a JFrog Artifactory product that I'm testing out in my group to demonstrate that it offers productivity gains -- and that's kind of what I'm buying is a product versus now I'm buying a platform. And what does platforms also done?

It's recognized that you're the leader, so I'm willing to sign a 3-year deal with you as well. That has incremental step-ups for the use of either developers in terms of the seat count for security or my commitment to consumption in the cloud. That's what happens when you start centralizing on platforms and consolidating these point solutions.

Story for JFrog in the last 12 months. I guess just thinking about some of the recent momentum with like becoming a repository of record. I think you mentioned this week like 3 out of 5 of the big native companies, customers of JFrog.

Can you just talk a little bit about that sale what they're adopting initially. I know that 1 of the customers doubled their license with JFrog in 2Q after signing in 1Q. I guess, just talk about that sale a little bit.

Yes. So I'll quickly try to cover that here in the time we have. And yes, we do have 3 of the 5 kind of native AI foundational model type companies that you all know as customers. I tend to say that, one, people ask me now about the other 2, where I don't think I have as much visibility as to where they're headed and that's something for me to go back and speak with the team about it a little bit more.

But I think one is obviously led to much more excitement and interest within not only in the investor group, but certainly with what they brought to JFrog and what they're looking to do. And that particular native AI customer had tried originally to build something Artifactory like on their own and failed. And thus, we didn't even know that they would be an opportunity and approached us in January and started talking to us about what their vision was of where they wanted to drive what they were going to do and how they would use Artifactory.

And the nerds inside JFrog started hip hopping around a across the Lilly pads because of all the new exciting things that these guys were talking about doing with Artifactory. And it was very unique in nature. And we see these guys which, ironically, a lot of these native AI guys for other people are in the cloud.

For us, it's a self-hosted deployment because the ultimate goal for this customer is to build their own data center. And if we're able to continue on and keep winning and prove out that we are, in fact, the model registry of choice for them, building a model as a service type offering to where they will maintain how secure and constantly train and update thousands of large language models which will then be utilized by other corporations in kind of a build versus buy scenario in which I may take a few of their models because those models work very well in writing the source code from my organization.

And those models will be trained constantly. And what's the core aspect there is JFrog Artifactory, keeping track of every change, every update, every movement, every security aspect. So when things go wrong, where am I going to go? I'm going to go directly to Artifactory to see if I'm at risk.

A few minutes we have left, the upcoming 2026 guide, recently got 4Q can you just help us understand maybe how you're thinking about with all this momentum, how you're thinking about your guidance philosophy? I know your guidance has changed over the last 6 quarters to not guide above use commitments. And then that includes some of the large cloud migrations Will that continue? Any color maybe you can provide for that coming guide?

Yes, sure. No, you did a great job on the interlude there in terms of Yes. I think that what we have found to be beneficial for us is that, and Ed kind of taking hold of the guidance philosophy and making it his own as he took the reins in '24 as the CFO of the company, is that guiding to commit is what we really have the core visibility to because in our usage-based model, you could go over, but you could say, hey, I still bought the right amount of cloud. We just had a new project or in this case, the NPM hack.

And that caused me to go over. I'll pay the overage JFrog, but then I'm going to go back to my minimum commit. And so that's why it's very hard for us to try to predict usage, not to mention that there's a disconnect between the procurement team negotiating the contract, the developer doing what he's told in just developing and consuming the consumption. There's no real correlation between all of those groups and what's going on.

So we think the best way to guide you guys now is to guide you on what we have signed or committed in contracts. And you saw us talk about a floor as it relates to our net dollar retention of 116%. We're at 118%. Why would you say 116%? Well, 116% gives you an idea of the fact of what the business looks like if we have no overusage and no real uptake of incremental security. 118% is a year of 2025, in which we benefited from strong security adoption and strong consumption trends.

Now as we guide '26, those strong consumption trends may, in fact, continue. And this is a debate I have with investors sometimes. But it's not something that we're going to forecast. We are going to continue to stay towards the commitments. And to the extent that we think that there's opportunities, then deliver upside through overages.

Jeff, thanks for being here. As Shlomi said, may the frog be with you. Thank you.

Thank you for having me. Good to see you.

Awesome. I think we'll get started. Thank you, everyone, for being here today at the UBS Global Technology and AI Conference. My name is Radi Sultan. I cover the SMID cap infrastructure software stocks here at UBS. Next up, we have JFrog, Ed, CFO; Jeff runs IR. So first of all, thank you very much for being here today.

Yes. Absolutely. Thank you for having us.

Awesome. Maybe just to kick it off. You guys have seen one of the strongest accelerations in the software group over the past few quarters, 6-point [ accel ] to 26% in the most recent quarter. So maybe just to level set what's changed? What have been the biggest growth drivers behind that acceleration?

Yes. So first of all, nothing has really changed in our strategy. We've seen tremendous execution during 2025. What we've seen though, which is different than what we saw in 2024 is usage over minimum commit. So we have very strong and robust cloud growth. 50% year-over-year in the cloud. Much of that, again, is being driven by the fact that we saw usage over the minimum commit.

Unlike 2024, where we had three of the largest deals in the history of the company, what we're seeing this year is more large deals. So we see increase in our ASPs. We see more velocity of those large deals. We're seeing expansion also in the enterprise. So we robust growth in the enterprise which is leading to higher commitments. And then lastly, security, we're seeing security being a driver of the ASPs as well. So taking all of those into consideration, we've had very robust very strong growth, and you see it in the results.

Yes. Awesome. And maybe just on that point, you reported, I believe, was the strongest top line beat ever in the most recent quarter, a little over 6% above the high end you've been vocal, your guidance excludes large deals and usage over minimum commit. So maybe just walk through those two variables, how those have been trending? And sort of what have those two sort of been the biggest driver lately?

Yes, you said something very important, variable. You said that, and that's exactly it. So the responsible guide that we do is eliminating that variable. The fact that we have usage that we can't predict, that's over the minimum commit. We derisk from that usage, and we also derisk these large deals. It's not a matter of when these deals are going to happen or if they're going to happen, but rather when they're going to happen. So it's a very responsible guidance. So understanding that first is what's very important.

What we're seeing today is strong demand. That demand is being driven by security. That demand is being driven by some of the events that are taking place like NPM where it's driving pipeline and strong opportunity to execute. We're still waiting to see if budgets will align for that demand. And we're also seeing, again, robust usage. This emerging technology that's taking place with AI models and AI-driven code. We've seen an increase, although on a very, very small base, which is driving demand. But part of that is excluding that, and letting that evolve over time. And hopefully, if we can continue to execute, should be open for a very, very good outcome for the remainder of the year.

Awesome. Maybe just on that point, as we think about the outlook for Q4, a common question I got was sort of the full year cloud guide does imply a pretty steep deceleration in Q4. So maybe just -- is there any way to talk about sort of the puts and takes on the Q4 guide, specifically understanding there probably is some conservatism in there, but maybe just anything over and above that when you think about puts and takes for Q4 specifically?

Yes, sure. Absolutely. Again, and I think there's been a lot of discussion around that is that the fact that the cloud, the way that we guide and understanding how our guidance philosophy is built, it's very responsible, excluding usage over minimum commits, only guiding on the commitments.

Secondly, excluding the largest deals, 8-figure deals that include migrations from self-hosted to cloud, security on top of that, excluding those. And again, only guiding on the commitment. And I want to remind everybody that last year at this time, we had three of the largest deals in the history of the company that closed. Those were cloud deals. They included migration from self-hosted to cloud or expansion in the cloud. We knew going into the second half of 2025 that it would be an increasingly more difficult compare and we called that out. But what we expect as we head into 2020 -- into Q4 of 2025, is that, again, we set it up for a very responsible guide, assuming that we can continue to execute on the momentum that we're seeing in the pipeline, it could be a very nice and attractive setup.

Awesome. And then maybe just on that cloud migration point specifically, is there any way to think about sort of the pace of those cloud migrations, how that's changed? And maybe when you think about like what has been the biggest catalyst behind those cloud migrations, it the functionality, features, et cetera?

Yes. So there is the secular trend to move from self-hosted to cloud, and we saw this robust migration taking place. Maybe in 2022. Yes. Before that, you had a bit of that pause as there was some optimization that's going on. And then we started to see that pick up a little bit during 2024, although not at the same pace that we saw previously.

Today, what we're seeing is customers, specifically the largest, most sophisticated customers are pausing on their decisions to migrate to the cloud. This is being driven by maybe the concerns around what AI will do in terms of cost predictability and governance around the control of the amount of code that's coming into the organization. So there's a bit of a pause right now, and we call this fit for purpose. We're hearing this more often. So we're working very closely with customers.

We're continuing to expand those customers in self-hosted and maybe taking other capabilities into the cloud like security. But for the moment, there seems to be a bit of a pause until there's this understanding around what is the cost predictability how am I going to manage the compliance and governance of the AI-driven code and large language models. And until we get past that gate, we believe that a full adoption into the cloud will probably be somewhat delayed.

Got it. Got it. And I guess -- maybe when you think about like next year, I understand you're not giving guidance, but maybe just as you think about the biggest growth drivers for next year, we talked about security, cloud migrations, usage, AI, et cetera. Like how do you think about sort of ranking the biggest growth drivers for next year?

Well, typically, migration is going to drive the biggest benefit. So when you move a workload from self-hosted to cloud, you have anywhere from 20% to 80% uplift in that same subscription type. So this is going to be the biggest driver. But if I was to look at where we head into 2026, there's two things that I'm focused on. Number one is, can I continue to see this momentum of usage over minimum commit or for that usage where I have today over the minimum commit, expanding that to a larger minimum commit, so having that expansion.

And then security. Security is going to be a continuation of what we see in terms of the performance in the cloud. If I land and see uptake in security with what we're building in the pipeline today and converting that demand into an expanding opportunity, then I think that 2026 could be a very strong year for JFrog.

Got it. And maybe just dovetailing off that, at swampUP this year, you had a heavy list of product announcements really attacking like a bunch of different growth vectors, right, AppTrust, AI catalog, Fly among others. I guess which of these are you most excited about near term and long term? And maybe why?

Yes. This is one of our best swampUP since I've been at JFrog, I've been here almost 7 years, and they've been doing this for 2 years now, but the amount of innovation that we delivered and the expansion from a technology perspective to the platform, lot of great products that we delivered. You mentioned AppTrust. AppTrust is one that there's a lot of buzz and excitement in the community right now and in the market with our customers. This really addresses a pain point for our customers, where today, it's a manual process to govern the compliance of software delivery. And so we see an opportunity with AppTrust.

We just released it. We're building the pipeline right now, and we're seeing good success in our pipeline. I don't know that it's going to drive any particular benefit during Q4. We'll see what happens during 2026. If we can execute on that demand that's being built in the pipeline. But certainly, there's buzz around AppTrust and governance, this [ DevGovOps ] category that we're creating today.

Secondly is going to be the AI catalog. AI catalog addresses the need and security for large language models. So we talked about in the public call how we're starting to see an emerging use case large language models, in particular with Hugging Face. So Hugging Face models that are being pulled into the organization. We've seen a 100% increase since Q1 and the amount of models that are being pulled into Artifactory, you have to be able to secure that AI catalog will be the way to secure your large language models. And then Fly. Fly was another product that we introduced during swampUP. Fly addresses the need for very small teams. There's a select group of customers that we're going after. This addresses the need that tend to adopt technologies quickly. Agentic capabilities, they want to use maybe a lighter version of Artifactory. We're building that group of select customers today, we will learn from that.

I think that's probably more long term. It's more of an R&D-related product. It will give us the learnings that we need in order to build the agent capabilities into the enterprise platform in the future.

Got it. Got it. I get the question a lot that like can Fly longer term, recatalyze sort of new customer growth and really start introducing that as a growth lever longer term? I mean I'm curious if you have any thoughts on that.

Yes. The intent is not necessarily to be a significant revenue growth driver. I think it's to make sure that we are bringing the small customer that is using advanced technologies or emerging technologies today. We want them to come to JFrog, and we want them to be aware of JFrog as an [ SMB ] opportunity, not go somewhere else where we then we have to wait until they become sophisticated enough to come to JFrog. So we're addressing the need. We don't see it necessarily as a revenue driver today. We see it more as an awareness and an ability to build top of funnel and ensure that all new customers, small customers that could be the largest customers in the future are coming to JFrog.

I'll just add quickly to that. I mean, I think to your point is that you're going to be working now with a new level of sophistication in startups. No longer are you going to have a team of 50 or 60 developers most likely in an Agentic world. That same team that used to be 50 or 60 developers may be three now. And so those guys are going to have a different type of profile in how they develop software, how they utilize AI, how various parts of the development pipeline are interfacing.

And I think for us, we want to be able to serve at some point down the road that type of customer. But at the same time, I think that there's a lot that can be learned to Ed's point in terms of maybe extracting R&D and knowledge out of this with Fly to take to the larger enterprises and say, "Look, we've already executed it this way. Look at this implementation of XYZ, and that may be a better choice for you".

Got it. Got it. Got it. I guess maybe just drilling onto the AI growth drivers. I mean, one thing that came through my own work was there's a lot of AI tailwinds sort of impacting your business you have sort of the AI developer tool impact on sort of usage, you have model registry, governance, the actual binaries that are used in AI application development, there's a lot of different growth drivers here. Maybe you could just sort of dissect those as you think about high level which of those sort of AI growth drivers you think will be most impactful? And then maybe how you think that sort of evolves?

Yes, it's a very good question, and we're excited about this shift in the era of AI. We think it creates a big opportunity for JFrog. Clearly, when you have code being generated by machines in the pace of code that will be generated, that means more binaries, and that's exciting for JFrog. That's what we manage. And we believe thematically, as you have more code being written and more binaries that this will be a benefit to JFrog.

But what we're really excited about too is the pace of code that's being written you have to secure that. We see a huge opportunity in terms of security, and we're doubling down on that because organizations are going to freak out with the amount of code that's coming into there. Into the -- our companies are going to be freaking out with the amount of code coming into the organization, they need to be able to secure that. And so there's a big opportunity from a security perspective. And again, we will double down on that ensure that we win that market.

Got it. And maybe just drilling down into that, I mean, how much of -- when you think about sort of the revenue benefit from those AI [ code gen ] tools is sort of that security angle versus just sort of court usage of pay, I need more binary to support 20%, 30% improve developer productivity.

Yes, it's probably too soon to call that out and to see that. There's multiple avenues. Here, you're going to have obviously, Artifactory will grow with the amount of code. You will have the AI ML direction as well as you bring models, models or binaries and you bring those into your organization. you'll need to train, secure and deploy those models. That's why we have JFrog ML. We did the acquisition of Qwak a year ago, and we already embedded that into our platform for the enterprise. So that's another avenue of opportunity for JFrog, but we're very, very early. We're probably in the first couple of innings right now, and it's really difficult for us to say definitively, it's going to be ML or it's going to be security or it's going to be AI workloads. It's still something that we want to be able to capture the market. We'll continue to invest in all of those technologies to make sure that JFrog is in the forefront of winning those opportunities.

Got it. Maybe just on the model registry side, you guys have talked about being the model registry of choice. So maybe you could just talk through the competitive landscape there, how that may be different from sort of the core artifact management business. but then sort of how that -- maybe how you're sort of mode around Artifactory translates into competitive edge there.

Well, we saw more than a year ago, maybe even 2 years ago, we started to see the data scientists bringing models into the organization and storing those in Artifactory. So we knew that there's an opportunity there because these models were starting to come into Artifactory. But what was missing was the ability to bring those models in and work them through the software supply chain. Develop those models, train those models and deploy those models. And that's why we did the acquisition of Qwak a year ago. This gave us the capability to do those three steps and ensure that models are being deployed in the most secure and compliant manner. And this is a differentiator for us.

So we know that large language model is a binary and you can store those in any repository, but what differentiates us is the ability to take those models through the development life cycle and then go through and deploy those models. So that gives us a competitive advantage at this stage.

Yes. I think at this stage, you don't see or hear yet. And I say yet, but no one is coming at the problem that you're discussing in the way that JFrog does. We're coming at it from protecting the core asset in the binary, and that's the value add that we bring to you as the organization.

I think that there certainly will be probably evolving emerging competitors for the ML landscape. It's just so nascent yet that we haven't necessarily seen even in presale activities that are somewhat still nascent for us with JFrog ML. We're not seeing a lot of alternatives being brought up as, hey, I've got you versus company X. So it's not there today, but neither is AI and ML. And I think as that starts to explode and take off, I think we'll see a shift in competitive landscape from what we've seen in the traditional DevOps or security type offerings.

Got it. And actually, one growth driver you did call on the most recent call was sort of the supporting the binaries used in AI application development, like Docker and NPM they're sort of getting pulled along as well. So maybe you could just talk through what you're seeing there, how much usage today is actually associated with those binaries to support AI application development.

Yes. What we actually see today and Jeff, feel free to jump in after this. What we see today is there's a lot of experimentation. So what we called out specifically around the usage coming from packages like Hugging Face, Python, which we know specifically are being used for AI workloads. We see those going into this -- coming into JFrog and doing some experimentation. What we don't see is a correlation between those package types and Docker, meaning that it's going into production. So we haven't seen that yet.

Now what we're seeing in terms of the robust growth that we had in the cloud during Q3 was specifically around our conventional package types. Docker, also Maven and NPM.

Yes. And I would just say, I think that it's a two-fold answer to Ed's point. We're benefiting certainly Q1 was benefiting from an experimentation in which we believe in AI and ML. I think that some of that's probably continue to go on. In Q3, we certainly saw the benefit, albeit maybe for a few weeks, but the first NPM incident happened at our swap-up event ironically, right? And customers were obviously having to do a lot of scanning of traditional packages of NPM to do updates and find the updates from the repositories and make sure that all the packages that they were using were in fact, the correct updates for the NPM language.

Got it. Got it. Got it. I guess maybe just switching gears. You mentioned the Hugging Face integration. I mean, could you just talk to that relationship. I do get quite a bit of questions sort of how that works, what the dynamic is there and sort of how that pulls through to your business?

Yes. I mean -- so the Hugging Face partnership is very unique in the sense that we look at the hugging face organization is similar to an NPM repository. They are just a repository for large language models. Now that relationship, I think, has been around. I think we've been working with them for about 1.5 years, but it's evolved even in that time.

Initially, it was having the interface to bring in the open models like a Facebook, Llama 4 or such Hugging Face into my organization and start to make it proprietary. The problem being about a year ago, Hugging Face had some other vendors doing some security work for that repository. And that particular vendor told them, 90% of these models are vulnerable. You're going to have to shut this. We're going to have to pump the brakes here. They asked us to come in and secure that repository, and we said, no, that's not the case, everything is okay.

And so that will also correlate I would think to what Ed and I have talked about that around the time that we began to secure the repository and say, "Hey, no, guys, it's safe". You started to see an increase in the downloads of Hugging Face packages. I don't know if there's a correlation, but there's some time in there in terms of the timing and such.

And so for us, it's not a revenue generator per se. It's a community engagement in the large language community. And we allow for the securing of that repository so you know that what you're working with from that repository is secure, but then the interface allows you to move back and forth, updates of those models as you bring in and out of the organization.

I think it's a very important point that Jeff brought up. This is not a revenue-generating partnership. It is a community trust and so once you have that trust that the models are secure, then they feel comfortable to pull those into JFrog.

It's an intangible benefit. Once you pull these models in, these models are very large models. And as you move those through your software supply chain, that's generating traffic for JFrog and we should benefit off of that traffic.

I said earlier, I mean it just ties back into the two gating factors. I think we're all waiting for when the floodgates open. And I think the floodgates are still holding back right now due to cost predictability and security. And security of large language models and what I'm bringing in the organization or if I'm even advanced enough to be using [ MCP ] servers to interface with other companies and share data to help train their models as well. So I think that those are two remaining gating factors, I think, that are holding back this kind of flood of adoption of AI development. .

Got it. Got it. Maybe just a related topic on a different integration you announced was the ServiceNow integration, I kind of stood out to me as an interesting growth angle. I don't think people like investors traditionally think about you guys as being sort of partners are sort of integrated. And so maybe you could just talk through, should maybe any early wins there, the dynamic behind that integration.

Let me take that. Yes. Well, we love to use the phrase to integrated to fail, that we are integrated with almost all the tools in the software supply chain. I think as it relates to the ServiceNow announcement, and it relates primarily to AppTrust. It's us helping the ops organization do their job, as Ed alluded to earlier, something that's being done on a spreadsheet today or I'll just sign DocuSign and say it's safe. Now I can interface with JFrog Artifactory, through the ServiceNow platform and know that each of the gates that needed to be achieved has, in fact, they achieved. I have a record of that. So when things go sideways, I can go back and find that information quickly through my interface with the ServiceNow and Artifactory platforms.

And so we do think that there could be more to come from that. We're very excited about working with ServiceNow. But we think what it does is it helps to continue because where we've sat obviously, is between developers and operations. And so we're now continuing to serve the operations side and bringing more integration into the platform for the operations side of software development.

Got it. Got it. I guess maybe just shifting gears to sort of the dynamic between GitHub and GitLab and you guys sort of where you sit within the ecosystem. I mean how has that relationship evolved? I know you guys have invested heavily in the GitHub integration, deprecated pipelines, right? And I think that's kind of brought closer to GitHub. So maybe you could just talk through sort of the competitive dynamic or the relationship there.

Yes. So I think that when I got here over 4 years ago, it was -- the [ Git ] are going to walk right and just mud stomp JFrog because binaries are just something you do and everybody can do it. And what we've seen is that there's actually been a change to where we're seeing the machines move less now.

And I think that what has changed for us, and I think Shlomi would say the same thing, is that when we first announced the integration with GitHub, it started to change the dynamics, not only in the industry, but I think, amongst investors as well that there were defined lines within software development. And that GitHub was the best of breed in source code. JFrog is the best of breed and binaries, and those are two separate parts of software development. There's going to be a best of breed and observability at some point, who that may be, I don't know, but that seems likely as well. And so I think that there's always a misconception that we were competing directly with those guys.

And realistically, we really don't see them in presale activity at all. And so that wasn't a competitive alternative that customers were really using in an enterprise level deployment we're looking at, let's say, a [ Git ] versus JFrog for binary management. That wasn't the case. And so I think that there's the competitive differentiation has been illuminated since we've been working closer with GitHub and you continue to see more and more engineering coming out of that relationship even at swampUP as we announced here in early September.

Yes. No, I mean, it seems like the sort of fear that GitHub one day is in a wake up and try and build an artifact repository, just isn't -- that's been a sort of fear for a long time, has it come to form.

It hasn't come yet. And I'm not going to sit here on a stage and say that if Microsoft somehow integrates GitHub and [ Bayside ] one day wake up and say, Satya says, today is the day we're taking binaries. I mean it's a big budget to fight. But I think that they've seen -- I think if you were just a GitHub universe about a month or so ago, they're not raising a white flag and saying binaries are JFrog biting me. But I think the overall topic that was being discussed, I want to GitHub employees was that we realized that JFrog does binaries very well. and that allows us to focus on our core market source code.

Got it. Maybe just shifting gears to security. At least in my own customer conversations, curation has sort of been the biggest standout where I've heard the most traction. Maybe you could just speak to sort of where within the securities suite you're seeing the most pull-through where you think there's still room to grow and sort of the biggest growth vectors there.

Yes. So historically, we've seen about a 50-50 split between advanced security, JFrog Advanced Security and Curation. And that's kind of how the pipeline was built over the last 4 or 5 quarters since we've really gone pedal to the metal with security. But over the last quarter or so, we're starting to see this groundswell being driven by the NPM events, and it's really around curation.

There's no competitive alternative when it comes to curation. We're very well positioned, and I think customers are now recognizing that they need to protect the castle. And the best way to do that is through curation. This allows you to curate what comes into your organization. So the NPM event certainly opened the eyes of the customer, and we're starting to see demand being driven there in the pipeline.

I also want to highlight, though, that this demand, we're still trying to see if the budgets are available. So we're hearing, yes, this is a very cool product. We want this, but we're getting a little bit of the straight arm saying, I don't have the budget yet, so I need to work with the office of the CFO, procurement, make sure that I can align the budget. And assuming that we can get prioritization around budget, I think curation could be a really good growth driver for JFrog in the future.

I guess when you think about sort of the -- where that growth is actually coming from, like how much is sort of consolidation of stand-alone DevSecOp vendors? Because it does seem like at least in my own customer conversations, those guys are under a little bit of pressure to sort of consolidate away from them towards a platform like JFrog maybe how much is coming from sort of platform consolidation versus different growth driver?

Yes. So the platform consolidation or vendor consolidation is really -- it's been a discussion for the last 2 years. and we see it as an opportunity. This is the reason why we brought advanced security into the market because it's consolidating 6, 7-point solutions. But we also know that you have a budget today that you're utilizing on these point solutions. We need to take that budget and we need to consolidate that onto JFrog. And that's not always the easiest thing to do.

You have tools that are working. You have an organization that tends to be very conservative, the CISO and the CIO and to rip and replace a tool or multiple tools is not always easy. This is why we're seeing customers take 3-year opportunities in security. They build a road map to replace maybe 1 or 2-point solutions in year 1, then a few more in year 2 and hopefully replace all of them in year 3. Today, we see 1 or 2-point solutions being replaced. Typically, it's infrastructure as code. Secrets detection, those tend to be the two. And contextual analysis as well, maybe three that we see customers adopting on the JFrog Advanced Security and replacing right now on the point solutions.

Got it. Got it. Maybe just switching gears to go-to-market. When we think about sort of the usage above minimum commit, which is sort of been a key part of how you guys guide and a key part of the go-to-market strategy right is attacking those customers that have usage over the minimum commit. So maybe you could just talk through sort of the trend you're seeing there in that usage above minimum commit. What has been the biggest driver? And how are you architecting the go-to-market to go and get those guys on over term commence?

Yes. Again, I'll say that the driver of the usage over the minimum commit is conventional package types. So this is going to be your doctors, your NPM and your Maven. It's not necessarily AI workloads yet. We're all waiting for that to happen. I know everybody here in the room, and it's great to see a packed room, is excited about what's happening in AI. We're still very, very early, and we're not a company that's going to give you a bunch of fluff saying I have a huge amount of growth coming from AI we're prepared to take that market, that there's no doubt about it.

But again, we see this from conventional package types. We work very, very closely with the customer. We align their demand with -- and their needs with long-term commitments. Now we're seeing multiyear commitments and making sure that they have the opportunity to flex up and down as they need by taking a minimum commitment with JFrog. We also know that the sales organization, they're motivated to ensure that they're landing the biggest and best opportunity for both JFrog and the customer, but we also know that we don't want to put them in a package that's going to be too large, and then we start to see churn. So having usage over the minimum commit is not bad for JFrog. We actually see this as an opportunity to go back to the customer and further expand their opportunities in the future.

Our sales organization is not incentivized based on usage over the minimum commit. They're incentivized on commitments. So they have an opportunity to go there based on driving usage over the minimum commit and converting that to a higher commitment in the future, and they'll be incentivized based on doing that.

Got it. Maybe just one to wrap it up, high-level question. I think I've been surprised, and I think most investors have been surprised at the durability of the competitive moat around the core Artifactory business as the DevOps space, in general, has gotten more competitive. So maybe just longer term, as you think out over the next several years, how do you think about sort of the investments you're making to maintain that competitive moat over the next couple of years?

It's a very good question, and we're always looking around the corner and who's coming after our market because binaries several years ago, there wasn't as much interest today. Everybody is recognizing that binary is the key to software updates. So for us, it's a continuation of expanding the platform. So [ DevGovOps ] is one example. The JFrog ML opportunity and as well as security. Having all of those, we're the only company in the world today that offers not only DevOps, but DevSecOps, MLOps and now DevGovOps. And as long as you continue to widen that moat and continue to add new technology to the platform, we believe we'll have a competitive advantage over anybody.

SP1 Yes. And I know we're at time. I'll keep it tight. It's a Lego strategy, right? DevOps is your first Lego, then you start stacking value on top of that to retain the customer because you continue to add value at their infrastructure layer solution. Awesome. Thank you guys so much for being here. .

Thank you for having us.

Ladies and gentlemen, thank you for joining us, and welcome to the JFrog Third Quarter 2025 Financial Results Earnings Call.

[Operator Instructions]

I will now hand the conference over to Jeffrey Schreiner, Head of Investor Relations. Jeffrey, please go ahead.

Thank you, Nicole. Good afternoon, and thank you for joining us as we review JFrog's Third Quarter 2025 Financial Results, which were announced following the market close today via press release.

Leading the call today will be JFrog's CEO and Co-Founder, Shlomi Ben Haim; and Ed Grabscheid, JFrog's CFO. During this call, we may make statements related to our business that are forward-looking under federal securities laws and are made pursuant to the safe harbor provisions of the Private Securities Litigation Reform Act of 1995, including statements related to our future financial performance and including our outlook for the full year of 2025. 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. You are cautioned not to place undue reliance on these forward-looking statements, which reflect our views only as of today and not as of any subsequent date.

Please keep in mind that we are not obligating ourselves to revise or publicly release the results of any revision to these forward-looking statements in light of new information or future events. These statements are subject to a variety of risks and uncertainties that could cause actual results to differ materially from expectations. For a discussion of 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, 2024, which is available on the Investor Relations section of our website and the earnings press release issued earlier today.

Additional information will be made available in our Form 10-Q for the quarter ended September 30, 2025, and other filings and reports that we may file from time to time with the SEC. Additionally, non-GAAP financial measures will be discussed on this conference call. These non-GAAP financial measures, which are used as a measure of JFrog's performance, should be considered in addition to, not as a substitute for or in isolation from GAAP measures. Please refer to the tables in our earnings release for a reconciliation of those measures to their most directly comparable GAAP financial measures. A replay of this call will be available on the JFrog Investor Relations website for a limited time.

With that, I'd like to turn the call over to JFrog's CEO, Shlomi Ben Haim. Shlomi?

Thank you, Jeff. Good afternoon, and thank you all for joining the call. JFrog's focus as a foundational platform and system of record for software delivery continues to drive momentum in our business, and I'm pleased to report another solid quarter across all metrics. Our execution remains focused. Our investments remain strategic, and our customers continue to tell us we are helping them to meet the demands of the fast-changing technology market.

In Q3, JFrog's total revenue was $136.9 million, up 26% year-over-year. Our operating margin was 18.7% in the quarter, demonstrating our ongoing discipline between expenses and strategic investments. Cloud revenue for Q3 equaled $63.4 million, representing 50% year-over-year growth. We experienced another strong quarter of cloud revenue growth driven by increased usage of conventional software packages and ongoing increases in usage of artifacts such as PyPI, Docker containers, NPM and models coming from Hugging Face for AI and machine learning.

Our go-to-market teams are executing on a clear strategy of guiding cloud customers with usage overages toward higher annual commitments in order to build stronger partnerships and drive predictable long-term value for customers and for JFrog. Our enterprise sales motions continue to bear fruit with greater than $1 million customers growing to 71 compared to 46 in the year ago period, equaling 54% growth year-over-year. Customers spending more than $100,000 annually grew to 1,121 compared to 966 in the year ago period, equaling 16% year-over-year growth.

In Q3, our 4 trailing quarter net dollar retention was 118%, demonstrating sustained growth among our customer base driven by strong cloud usage and fueled by our holistic software supply chain security offering. Ed will further discuss net dollar retention later in the call.

Now let me spotlight Q3's wins, including cloud and security and discuss JFrog's AI and machine learning developments. First, addressing our cloud growth in Q3. Our ongoing growth in the cloud is supported by 2 vectors: our expertise in managing customers' binaries as they scale alongside AI-generated artifacts and our observation of emerging trends in AI software package volume. We believe Q3's cloud performance reinforces how customers view the JFrog platform and Artifactory at the core of their operations. JFrog is already positioned as the universal binary repository to manage all software artifacts and is becoming the model registry of their software supply chain.

As software continues to be created at an ever-growing pace by both humans and machines, the volume of artifact rises, demanding not just intelligent storage but a robust binary delivery system and a single reliable system of record. As their delivery pace increases, our customers are adopting hybrid fit-for-purpose cloud strategies for their emerging AI workloads. They tell us they value cloud elasticity for AI adoption and deployment but remain flexible in their approach due to the unpredictable compute costs, advising us that it's still too early for them to go all in on the public cloud.

Meanwhile, the volume of AI-driven packages is rising, fueled by our hug-and-face integration and native support for ML models, Docker, NPM, PyPI and other key components demanded by AI. We continue to monitor cloud adoption and AI-driven usage closely and believe it is still too early to bet on significant cloud usage growth. Our hybrid and multi-cloud offerings differentiate JFrog and uniquely position us to capture expansion due to AI, whether in the cloud or on-prem, delivering unmatched software supply chain, holistic platform capabilities and deployment flexibility.

Next, to security. In Q3, again, we saw some of JFrog's largest customers wins, including new logos and significant multiyear contracts. Many of these deals included JFrog's holistic security solutions such as JFrog Curation and JFrog Advanced Security. We experienced this customers' purchasing behavior across multiple verticals and geographies. For example, we closed a 3-year deal with the United Kingdom's Customs and Revenue Agency with a TCV of $9 million.

In a similar move, a key U.S. federal agency chose JFrog to shift left with JFrog Security Solutions and protect their software supply chain and 2,000 developers with an Enterprise+ subscription, JFrog Advanced Security and JFrog Curation as their open source software package firewall. In North America, our enterprise focus drove the closing of a net new customer deal with a TCV of more than $4 million for one of the world's top energy corporations. They were seeking to modernize and standardize software supply chain security and processes for their 3,500 developers choosing the core JFrog platform with JFrog Security Solutions in the cloud.

The example wins give us confidence that the future of the software supply chain security and software governance market is being written not by point solutions but by holistic system of record that incorporate binary management and end-to-end security consolidated into a single platform. A unified platform is a growing requirement as companies have seen software supply chain attacks increases significantly over the past year. These attacks have become more sophisticated, targeting build pipelines, AI and machine learning software supply chains and exploiting vulnerabilities in software binaries.

Recent NPM, PyPI and MCP attacks show hackers are keeping pace with technology but our security research team and the JFrog Curation solution have been praised by our customers for protecting them from this recent potentially devastating attacks. A cybersecurity lead at a Fortune 50 American company noted to us following the NPM attack. "Our JFrog Curation deployment provides a very effective and efficient supply chain protection. We were able to shut down recent provider attacks minute once discovered, and the control has proven 100% successful since." Nothing fuels us more than our customers' feedback, and we are committed to safeguarding their software supply chains and aiming to ensure digital trust across their software package life cycle.

Hackers are targeting software supply chains. They know that binaries, software packages, containers and AI models are gateways into our customers' runtime environment. This is now a real threat to every organization. Without strong protection for your software supply chain and binaries, you remain exposed. Attacks are not a question of if but when. While one quarter doesn't define a consumption trend, Q3's adoption of JFrog Security solutions give us cautious optimism for broader long-term momentum.

Now to AI and machine learning. We continue to gain traction in the market as the model registry providers and are being positioned as a system of record for AI delivery validated by some of the world's leading AI companies. In fact, Justin Boitano, VP of Enterprise AI at NVIDIA, noted at the JFrog Annual User Conference swampUP in mid-September that JFrog enables enterprises to securely build, manage and scale AI agents, the next generation of intelligent software from development to production. "AI agents are the next wave of enterprise innovation but they are also the newest and most critical software artifact to secure. This is no longer just about models. It's about industrializing intelligent autonomous agents. By integrating with NVIDIA AI Enterprise and streamlining deployment onto AI factories, JFrog is delivering the essential pipeline to rapidly secure, manage and scale these AI agents from development straight into production."

Moving throughout 2025, JFrog has embraced the AI revolution, focusing on what we believe we do best, driving the next standards in the market as a single source of truth for AI software packages. We integrated JFrog ML into the JFrog platform, empowering data scientists and developers to build, test, experiment and deliver trusted models into production. We launched our MCP server alongside groundbreaking MCP security research. redefining how developers and AI agents interact with their software delivery platform. In addition, we introduced AI catalog, a new add-on to JFrog Curation designed to secure and govern both third-party and internally developed AI models, ensuring they are safe, compliant and aligned with company policies for responsible use across the organization.

Our next generation of offerings in ML and AI as well as our DevOps and security products were highlighted for customers and analysts at swampUP in Q3. Every year, JFrog welcomes customers, prospects and partners to our swampUP conference, a key industry gathering for DevOps, DevSecOps and MLOps users. This year, JFrog innovations were front and center with pain solving solutions for customers.

On stage, we announced a new way for companies to govern and trust their application with an innovative product, JFrog AppTrust. We were joined for the opening keynote by NVIDIA, ServiceNow, GitHub and Sona to showcase our partnerships in an industry-first DevGovOps solution. DevGovOps brings software development, security and GRC groups together, focusing on once again the key asset of any company, the software binary.

As the complexity in the world of software increases, the requirements of security compliance and governance put pressure on development teams. The need to release fast and often, combined with high regulation requires automation and collection of evidence such as quality testing or security results to keep software flowing with confidence. With the new JFrog AppTrust product, companies will now have an automated way across platforms to manage governance as they deliver software faster than ever before. True to our vision of cross-industry collaboration and our commitment to build solutions that are too integrated to fail, AppTrust was launched in partnership with ServiceNow as the IT ops system of record with JFrog as the system of record for the software supply chain in the worlds of Rahul Tripathi, General Manager of ITSM at ServiceNow. The future of software built by developers and AI agents must include automated governance, achievable only through evidence-based quality gates and systems. JFrog AppTrust is paving the way to make that vision a reality.

In security, we further showcased new abilities to secure developer extensions or the plug-ins developers use in their environments. This is a different type of supply chain attack and JFrog customers can now protect developer extensions in addition to their software assets. We also launched agentic remediation capabilities and demonstrated on stage how JFrog solutions use AI agents to detect vulnerabilities in source code, identify remediation methods, prioritize contextual path to fix problems and even protect against similar vulnerability in the future, first through our GitHub Copilot integration and soon available for other code assistance.

Finally, we introduced JFrog Fly, the world's first agentic repository, reimagining software supply chain management in the era of AI. With JFrog Fly, AI agents become co-builders alongside developers, orchestrating artifacts seamlessly across the entire software life cycle. This allows developers to focus on what matters most, delivering software to production faster at scale and without friction. Small teams now enjoy a zero hustle AI-driven experience, tightly integrated with GitHub and native AI tools like Cursor, Cloud Code and GitHub Copilot.

JFrog Fly's Agentic capabilities will be shared with selected users, giving developer teams the opportunity to experience AI-assisted software creation and delivery. These capabilities are planned for full integration into the JFrog platform, powering our customers in the new era of intelligent automated software supply chain practices built on an agentic binary repository. We believe that these product innovations share that swampUP are in the words of our customers, AT&T, a home run for our users.

With that, I'll turn the call over to our CFO, Ed Grabscheid, with an in-depth recap of Q3 financial results and our updated outlook for the full fiscal year of 2025. Ed?

Thank you, Shlomi, and good afternoon, everyone. During the third quarter of 2025, total revenues equaled $136.9 million, up 26% year-over-year. Our overachievement during the quarter was the result of strong go-to-market and operational execution, continued momentum in our cloud revenues, growing adoption in JFrog security products and expansion by customers within our enterprise-level subscriptions. As noted by Shlomi, third quarter cloud revenues grew to $63.4 million, up 50% year-over-year and represented 46% of total revenues versus 39% in the prior year.

Our growth in the cloud was driven by increased usage across a broad set of package types, demand for JFrog Advanced Security and Curation and conversion of customers with usage above minimum commitments into higher annual contracts. During the third quarter, our self-managed or on-prem revenues were $73.5 million, up 10% year-over-year. We continue to proactively engage our on-prem customers to migrate DevSecOps workloads to our cloud or explore solutions better aligned with their specific use cases, including hybrid and fit-for-purpose deployments.

In Q3, 56% of total revenues came from Enterprise Plus subscriptions, up from 50% in the prior year, driven by ongoing execution of our enterprise go-to-market strategy and broader customer adoption of the JFrog platform, revenue contribution from Enterprise Plus subscriptions grew 39% year-over-year. Net dollar retention for the 4 trailing quarters was 118%, consistent with the prior quarter, highlighting the continued adoption of our security core products and increased cloud data consumption, resulting in higher customer commitments. We continue to demonstrate that our customers view JFrog solutions as mission-critical to their software supply chain with gross retention that equaled 97% as of the third quarter 2025.

Now I'll review the income statement in more detail. Gross profit in the quarter was $114.9 million, representing a gross margin of 83.9% versus 82.8% in the year ago period. We remain focused on cost optimization with the cloud service providers and continue to expect annual gross margins to remain between 82.5% and 83.5% for 2025. Operating expenses in the third quarter were $89.3 million, equaling 65% of revenues. This compares to $75.5 million or 69% of revenues in the year ago period. Our operating profit in Q3 increased to $25.6 million or an operating margin of 18.7% compared to $14.7 million and 13.5% operating margin in the third quarter of 2024.

The continued balance between strategic investments and operational efficiency demonstrates our commitment to profitable growth. Cash flow from operations equaled $30.2 million in the third quarter. After taking into consideration CapEx requirements, our free cash flow reached $28.8 million or 21% margin compared to $26.7 million or 24% margin in the year ago period. During the third quarter, we completed payments totaling $5.7 million under a holdback agreement related to the acquisition of Qwak AI, which was completed in July 2024.

Now turning to the balance sheet. We ended the third quarter of 2025 at $651.1 million in cash and short-term investments compared to $522 million at the end of 2024. As of September 30, 2025, our RPO totaled $508 million, a 47% increase year-over-year. This performance highlights the successful execution of our go-to-market strategy as customers continue to make larger multiyear commitments to our DevSecOps offerings.

And now let's turn to the outlook and guidance for Q4 and the full year 2025. While we're encouraged by our strong year-to-date performance amid geopolitical uncertainty and ongoing macroeconomic volatility, we believe it remains prudent to continue to approach our forward outlook with measured caution. Our updated 2025 guidance range suggests sustained contributions from the JFrog Security core, steady expansion of customer commitments and adoption of the full JFrog platform.

We continue to derisk our outlook by excluding our largest opportunities given the uncertainty regarding the timing of certain large customer deployments. We estimate our full year 2025 baseline cloud growth to now be in the range of 40% to 42%. Cloud revenue guidance continues to exclude any contribution from usage above annual customers' minimum commitments. Taking into account our strong year-to-date results and fourth quarter guidance, we are raising our expectation for net dollar retention to above 116% for 2025.

For Q4, we expect revenues to be in the range of $136.5 million and $138.5 million, with non-GAAP operating profit anticipated to be between $21 million and $22 million and non-GAAP earnings per diluted share of $0.18 to $0.20, assuming a share count of approximately 125 million shares. For the full year of 2025, we anticipate a revenue range of $523 million to $525 million, representing approximately 22.3% year-over-year growth at the midpoint.

Non-GAAP operating income is expected to be between $87.3 million and $88.3 million and non-GAAP diluted earnings per share of $0.78 to $0.80, assuming a share count of approximately 122 million shares.

Now I'll turn the call back to Shlomi for some closing remarks before we take your questions.

Thank you, Ed. The third quarter of 2025 is behind us. We committed and we delivered on innovation, on product execution and across every financial metric. Despite macro challenges, the JFrog team sold, over the past 3 quarters, we not only earned the trust of the enterprise, we reinforced our position as the definitive system of record for the software supply chain. As the world becomes powered by AI-driven software and every aspect of software creation and delivery is being transformed, one fact stands out, more software means more packages, more artifacts, more binaries, and that's exactly where JFrog stands front and center now and in the future.

We're becoming the model registry of choice, securing the entire software supply chain of our customers from passport control at the gate to safe and trusted delivery. We manage the full life cycle of AI models, and we've introduced the world's first DevGovOps solution, preparing our customers for the ongoing tsunami of AI regulation and compliance. We continue to run our business with efficiency, focus and discipline while staying deeply energized by the opportunities ahead.

This quarter was also a deeply meaningful one for our Israeli team as the war in the region came to an end and hostages were finally returned safely to their families after 2 long years. We stand in solidarity with the families of the remaining hostages still held by the terrorist Organization, Hamas, and we pray for their safe return home for proper burial.

As we prepare for 2026, we are ready to leap forward in product, people and business. We remain determined to continue building on what we have created by developers for the developers of the modern software world.

With that, thank you for joining our call and may the frog be with you. Operator, we are now open to take questions.

[Operator Instructions] Your first question comes from the line of Michael Cikos with Needham.

Jeff, great to hear you on the call and for Shlomi, congrats on the strong quarter. I wanted to tap into cloud. And first, just to sanity check, the results super strong here. Was there anything one time? Or was this really just a confluence of a number of different pieces of the platform coming alongside the execution? Can you help us unpack that?

Mike, this is Ed. I'll go ahead and start with that. There is nothing in the cloud revenues that is one time. This is a convergence of strong usage across multiple package types, geos and different verticals. In addition to that, we saw strong security. Security also is a leading driver of our cloud growth. So you had a combination of those 2 things happening but no one time. It was a very strong quarter.

Excellent. And I did just want to tap in. I guess this is probably more for Shlomi but again, on the Cloud, we have a couple of quarters now of really strong growth. What is different, if anything has changed. But from an execution standpoint on the go-to-market, what have you guys implemented that continues to bear out here? How should we be thinking about that?

Mike, this is Shlomi, and thank you for the kind words. I think that what you see in the cloud is a strong and consistent execution, not only of our technology but also the go-to-market philosophy of working with our customers that have over usage, converting them to higher commitments, and that's translated to consistent cloud growth and less volatility. As you know, even when we guide, we provide you with the commitments-based guidance and consumption might come and go but we wanted to have an alignment between the go-to-market philosophy and the execution of our cloud business.

That's on top, of course, to what Ed said that our cloud customers are now also embracing our security solution, which obviously gives us more room to grow and to expand.

Your next question comes from the line of Sanjit Singh with Morgan Stanley.

Congrats, my congrats on the strong results this quarter. Shlomi, I was wondering if you could expand a little bit more on the theme on your script around the broadening of the types of artifacts that Artifactory is managing and storing and serving as a system of record. What does that mix start to look like between kind of traditional software artifacts, containers registries and some of the AI artifacts that are starting to come through? How -- can you speak to some of the underlying trends in terms of what Artifactory is storing and managing these days?

Yes, Sanjit. What we see, and we're obviously monitoring it closely is that artifacts or different type of artifacts that are heavily used by AI creators are seeing kind of a growth in usage in our cloud and on-prem customers. We see that obviously, Hugging Face is scaling up. This is the open source models hub for AI models but also languages that support AI development like Python with PyPI, NPM, Docker containers for distribution of models, all of them are aligned and correlated and growing together.

It's still too early for us to say that this trend is actually going to lead to a higher consumption in the future. But it's very encouraging to see that our customers are using JFrog as the system of record for all packages, including AI models, and that obviously drives the higher consumption as we reported.

Your next question comes from the line of Kingsley Crane with Canaccord.

Really encouraging results. For Shlomi, the demo of JFrog Fly was really impressive at swampUP. The MCP capabilities are nice to see as well. It got me thinking AI is changing how consumers interact with technology. It's leading many companies to rethink their UI, their UX. How relevant is that for JFrog? And is that something that you're thinking about?

Yes. Thank you, Kingsley. The JFrog Fly release is a disruption to the entire software supply chain system of record. We understand that software in the future will be created not just by developers but also by agents. And therefore, whatever repository you want to build must come with a agentic capabilities. This is what we wanted first to have a JFrog Fly. The second thing based on our research with thousands of developers was that developers want to be focused on what they need to be focused on, create software, deliver software with trust fast and rapidly. JFrog Fly provide that.

So instead of taking it feature by feature within Artifactory, we thought that what we will do better is to build a full agentic experience for our users, reimagining how software supply chain will look like and then slowly bring those capabilities into the JFrog platform, obviously, the enterprise market will adopt it one by one, feature by feature. The community and small team can run faster. So that's the idea behind Fly, and we are very excited to see how the market react to it.

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

I wanted to ask on NRR and maybe pick on the metric just a little bit because it was flat at 118%. But when I look at the cloud growth, that's really, really strong and just knowing that NRR is calculated on the trailing 12-month metric, on a 12-month basis, that just really means that there's some new customers that are growing very fast. Is that the right interpretation on kind of looking at NRR versus the cloud performance? And if that is true, how sustainable is that growth from these new customers going forward?

Thank you, Koji. I'll start and Ed, feel free to chime in. NRR represent a tier of customers that are not committed to an annual NDR...

NRR. Shlomi, let me go ahead and jump in here on the NDR. So first off, let me remind you that we had 3 of the largest customers closed during Q3 of last year. So we're building off of that. It's a trailing 12-month metric. But what I said in the beginning of the year, Koji, if you recall, if there was going to be an acceleration in net dollar retention, 2 things would have to happen. Number one, I want to see an uptick in my security. And number two, I want to see usage over minimum commit in the cloud, and both of those are happening.

So you start on a trailing 12-month metric with a very strong base off of those 3 very large deals that were closed at the end of Q3, and we've continued to build on that. So we've actually expanded our net dollar retention rate, and we're very happy with where it is as it's stabilized quarter-over-quarter at 118%.

Yes. And Koji, sorry, I heard NRR and answered about the cloud monthly usage. Regarding net dollar retention, we also have to remember that part of the strategy of embedding our security solution into the platform will be a way to also expand our customers with a different persona. Getting our customers expanding with security actually addresses a completely different addressable market.

Your next question comes from the line of Mark Cash with Raymond James.

Shlomi, if I could start with you. I also want to ask around Fly but more around if you see Fly opening up new customers or buying centers that you haven't serviced previously? And then what kind of go-to-market plans or tweaks would be required to drive market awareness versus maybe the market mostly thinking about JFrog more large-scale artifacts.

And then if I could sneak one in for Ed. I'm just -- I mean, you beat by $9 million, raised by additional $6.5 million. So I guess maybe some of that comes with better visibility from RPO, cRPO strength but you're still taking a prudent approach you laid out. So what are the key drivers that give you confidence to raise by such an amount maybe compared to 90 days ago?

Yes, I'll start with Fly. This is obviously an opportunity but it's not our goal. Our goal is to make sure that everything that you have at the JFrog platform in the future will support both agents and developers as they manage their software supply chain. And why is that important? Because we know that in the future, the way that engineering team will be structured will be a combination of developers and agents. It will not be just developers. So software will be made by agents. And if you don't provide them, these agents will not have access to your repository, then they will have another system of record. So with Fly success, basically, what we see is how we fuel and power the entire JFrog platform.

Now can small team use FLY and this will be another avenue of revenue generation? Yes, maybe. But our main goal is to make sure that we adopt this agentic experience across the platform and Fly is the North Star of adopting AI experience.

Yes. And regarding the guidance and the confidence that I had moving into Q4, first of all, I only guide on the commitments. So I'm confident that what I'm providing to you 90 days after the last time I provided to you a guidance is the fact that I have the strong commitment. We saw a strong performance during the quarter. That's built into my forecast based on those commitment levels. And I've removed anything that's related to usage over the minimum commitments and feel very confident with those numbers provided.

Your next question comes from the line of Miller Jump with Truist.

Congratulations on the strong results. Yes. So for Shlomi, maybe for a couple of quarters now, we've heard about the demand for hybrid solutions correlated with AI. I'm just trying to reconcile this with the clear momentum that you're seeing in cloud. Like is this something that you're seeing shift the pipeline composition right now?

And then if I could ask one for Ed as well, just throughout this year, maybe the last 2 quarters really, you've talked about the conversion of customers that are going over commit. And I'm just curious, like what are you seeing from the customers that do move on to new contracts? Like are they continuing to ramp up their consumption? How has that played out versus your expectations?

Yes. Thank you, Miller. I'll start. Well, first, what we see is what we reported in the past, and we see it still. In our pipeline, there are some big deals that part of them include cloud migration. So we hear from our customers that they need more certainty before they double down on the cloud and go there with the entire AI workload that is currently being built. So for sure, the fit for purpose that we reported in the past is still relevant. But you also know that part of our guidance philosophy, these deals are being derisked from our pipeline. And we are being very cautious with kind of hanging the expectations high when it comes to cloud migration and cloud consumption.

In terms of the consumption, as I answered before to Sanjit's question, we see more software packages that are related to the AI world being used. And still, we want to make sure that this is a new behavior and not just a trend that will pass when people will decide whether they go with cloud or on-prem.

Regarding what we see from the behavior of those customers that are above minimum commitments that go to a higher annual commitment, it's still not easy to do this. And the sales team has done a good job. That's strategically aligned with our philosophy. Budgets are still not fully aligned there yet. But for those customers, they're now secure and it gives them that confidence to flex up and down. So we continue to see strong usage across the board, especially in Q3. But we also know heading into Q4, it's a seasonably slower quarter due to the holidays. So that's also something that we take into consideration.

Your next question comes from the line of Brian Essex with JPMorgan.

Congrats from me as well on these results. They are really nice to see the acceleration. Maybe for Shlomi, could you -- any way to quantify what kind of lift you get in the quarter or you had in the quarter from conversions to cloud? And maybe what percentage of your customer base is kind of remaining to convert?

Yes. So we are not disclosing this number, Brian. But what I can tell you that if you look at the report of the over $1 million customers, we have now 71. That's a growth of 54% year-over-year. I can say that the majority of them were not just using of more platform capabilities but also use it in the cloud. So while we are not disclosing it, the cloud for sure is a strong growth engine for the company.

Your next question comes from the line of Shrenik Kothari with Baird.

So Shlomi, definitely securing the software supply chain for AI models, continuous packages sounds like a critical pain point right now. I just wanted to hear your thoughts on what do you think about customers of truly allocating new budgets towards this and just the sustainability of these budgets? Or do you think it's getting bundled into existing DevSecOps and you are gaining from consolidation. You did note many attacks like the recent NPM and PyPI threats that are thwarted by Curation. Just curious how customers are thinking about the budgets going forward? And then I have a follow-up for Ed.

Yes, Fredrik, that's actually 2 different questions. First of all, there is a higher threat on software supply chain because hackers are after your software packages. And if you don't cover your software, it's -- as I said, it's a matter of when and not a matter of if you will be attacked. NPM, PyPI, MCP, in the past, you remember Log4j, these are all software packages and attackers are going there because this is binaries and binaries are also the asset that our customers have in the run time. So basically, if you are attacked from run time for the binaries, you have full access to the software supply chain. That's front and center for every CISO now.

And the other question regarding new budget allocated to security in the world of AI, I believe the answer is yes. And it's not only the traditional security but also GRC budgets. So cyber risk organizations are also looking at what happened in terms of governance and regulation, a moment before AI is fully adopted by the biggest bank and the biggest car manufacturers and the biggest retail companies. And the gap between full adoption of AI to what we see now is basically trust, control, security and governance. So for sure, there is more budget there and more attention of CISOs and CIOs to make sure that software delivery driven by AI is not putting the company in a risk.

Just quickly, Ed, a follow-up, Federal wins were a highlight. You mentioned meaningful TCV. So just in light of the strong Q3 and traction around security AI, the guide for Q4 does appear measured. You did touch upon it. It's coming from place of prudence but just implied is more kind of 25% to 30%. Is there more prudence coming from the Fed side? Anything that you can comment on deal timing, variability just given the prolonged shutdown?

Yes. Thanks for the question, Shrenik. There's nothing related to a government shutdown that would change anything in terms of our sentiment going forward. We've managed this year with prudence. We're continuing to manage the year with prudence. There's a lot of factors that we take into consideration. I want to remind you, there is no usage over minimum commitments in our guide. We derisk for our largest deals as the timing of those deals are still uncertain but nothing has changed from our previous philosophy, and we're continuing to use that same philosophy in our Q4 and fiscal year guidance.

Your next question comes from the line of Ittai Kidron with Oppenheimer.

Congrats, guys, really impressive. Maybe a couple for me. And just on this last comment that you made about that you don't give guide over the minimum commits. Can you tell us in the quarter itself this quarter, what was the upside from kind of spillover over the minimum commits?

Yes. We didn't break that out, and we're not willing to provide that information. But if you think about what we've carried over from Q3 into Q4 sequentially, that is the commitment levels. Anything above that, I would consider to be usage over minimum commits.

Okay. And then Shlomi, for you. Clearly, you're doing very well. I'm kind of wondering internally, what percent of your sales force is running above quota for the year? And if that's running at a level that's higher than your normal internal averages. I'm just trying to think about, you're soon to finish the year, you have to start thinking about how you make tweaks and changes. I was wondering if you have any initial thoughts given the change in the landscape, AI, the breadth of your portfolio, do you have any initial thoughts on how you're going to tinker with comp as you go into next year?

Yes, Ittai. So obviously, we will not share some operational metrics but of course, we are tracking it. There are very important takers for the go-to-market team when it comes to the growth engine, when it's security, when it's cloud, cloud migration, AI and MLOps adoption, now DevGovOps and compliance. What you can see in the numbers, and this is not the first quarter, it's, I think, the fifth quarter in a row that we are consistently deliver what we are committed to is that not only the sales team is focused but also they use the methodologies of top-down enterprise sales, things that in the past, we didn't practice like 3 years ago or 4 years ago, when we used to sell to developers, there were no expansion.

And now what you can see not only by the reports of the big numbers or the big companies but also the entire revenue growth and the consistent growth is that the team is focused on enterprise sales upmarket, know exactly how to expand the platform, and we are very pleased with the results.

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

Shlomi, I'll take the 5 quarters one step further and say the magnitude of the SaaS speed over those 5 quarters continues to get bigger. So maybe building on Ittai's question, is there any reason why you wouldn't accelerate sales and marketing headcount spend as we head into next year to take advantage of all this opportunity you see?

No reason. We are investing in go-to-market. We are investing in go-to-market in a responsible way. We want to see that what we deploy also comes out as the right ROI for our investment. And it's on all fronts. You can see the amount of releases from the product side, but you can also see the alignment on the go-to-market side when we deliver the numbers with such a strong bid.

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

Great. Shlomi, I wanted to ask about the security wins and pipeline, some good color there in the quarter. But for the Q4 pipeline and beyond that, how is that tracking? How much is that pipeline up? What's your confidence in closing some of these big deals? Have the sales cycles changed at all? How is the market kind of shaping up competitively for these deals, too?

Yes. Great question. We see a lot of opportunities in our pipeline, including security. Obviously, JFrog Curation following the incident of NPM is something that a lot of our customers are asking and inquiring about. The sales cycles are longer when it comes to security because no customer comes with 0 security coverage. So obviously, it's not just a matter of upgrading themselves but also displacing something that they used in the past. So the answer regarding the pipeline, it's growing, and we are very pleased with what we see there. It's also -- to remind you, it's only part of our Enterprise X and Enterprise+ subscription. So sometimes it's also drive an upgrade in the subscription.

And through to our methodology, when it comes to mega deals, in the world of platform that also includes security, we are derisking it to make sure that we are not missing a quarter because the customer needs a bit more time to conclude the proof of concept. Overall, the sales cycles are a bit longer but not massively longer. I would say 3 quarters in average, sometimes it's 4 quarter really depends on how many persona are involved in what you need to adopt from JFrog.

Your next question comes from the line of Jason Celino with KeyBanc.

Really wonderful quarter. The 2 wins you talked about that I thought were interesting, so the one U.K. customer and then the U.S. Federal Agency, did these deals directly benefit from the launch of AppTrust? I'm just trying to understand how AppTrust and the government opportunity might play as a catalyst.

Yes. So that was really a big win. And obviously, it was an RFP. So not only JFrog compete over this opportunity, and we are very proud of the team that delivered that, by the way, a very long process, obviously, in front of the government. AppTrust is not yet included. AppTrust was just announced last month, a few weeks, a baby product. We announced that together with ServiceNow. We are building the pipeline for AppTrust. And the DevOps landscape, governance is something that we know that will get even more demanding when AI will fit in. So not yet. But I will answer something that you didn't ask. There is a room to grow with all of our customers when AppTrust is in.

Your next question comes from the line of Eamon Coughlin with Barclays.

I would reiterate congrats on a great quarter. The strong enterprise customer adds really stood out in the quarter. Can you help us understand some of the key drivers there? Like are you spending more time engaging with your partner ecosystem? Is it a function of a more mature sales team? Just curious if there's anything you're doing to adjust a little bit of your sales motion or go-to-market motion.

Thanks for the question. Yes, we're seeing great traction in the enterprise, especially those customers over $1 million. What we see, and Shlomi talked about this previously, security is becoming a driver of many of those large deals, and it's also in the cloud. So we see this expansion of the usage over the minimum commit and expanding to a bigger commitment, taking security on top of that. That drove many of the deals that we saw 10 customers that we had over $1 million during the quarter, 54% growth. It's the efforts of the go-to-market team, and it's really a 3- or 4-year investment that we made, and we're starting to see the fruits of those labor.

Your next question comes from the line of Jonathan Ruykhaver with Cantor.

Congrats on the performance. I'm curious, so last week, OpenAI announced a private beta of security application called, I think it's Aardvark. But it seems to differentiate with an LLM-driven approach to vulnerability detection and remediation. And I'm curious if you could just talk about how this compares to X-ray's capabilities. But maybe I think more importantly, the announcement at swampUP around agentic remediation capabilities. How do you think this maybe further differentiates you from emerging competition?

Yes, Jon. Well, this was a very important announcement coming from OpenAI. And obviously, we saw it, and we work with OpenAI on a monthly basis, if not daily basis. This is great news because it's yet another proof that human security is -- sorry, human code creation is being covered by models. That's basically supporting the philosophy that what you need to invest in is to protect your binaries. The announcement from OpenAI is a solution that replaces static code analysis, basically code that is created in human language and not binaries. If it will drive something, I think it will drive more interest in what we bring to the market, which is a complementary solution to protect your entire software supply chain. So yes, LLM can provide your source code security, but not binaries, and that was the release from OpenAI.

Your final question comes from the line of Rob Owens with Piper Sandler.

Just a couple of quick ones for me. Where you talked about cloud migration and some of those in the pipeline. Taking the other side of that, the customers that aren't moving to cloud, what are some of the governors or some of the reasons that they're not moving to cloud at this point? I understand that having a hybrid architecture is part of your value proposition. But just contemplating that move to cloud and maybe some of the barriers to get certain customers to move.

Yes. That's a good question, Rob. So assuming that our customers -- enterprise big customers moving to the cloud from an on-prem setup means that they had their own projects. They understood the scope of these projects, and they wanted to move the workload to the cloud in order to kind of provide a better elasticity for the company. Now comes AI and they see a new workload. They don't know how much they will pay for storage. They don't know how much they will pay for data transfer when you train these models, where the data to train this model will be hosted. That takes a bit more time for them to analyze.

Putting aside the cost predictability, they also have some security and governance concerns. Who is training this model, how this model is being exposed? What are we doing in order to protect the company from having any type of malicious models coming in. And this, I think, bring them to take a bit more time or at least this is what they tell us. They need a bit more time before they fully bet on the cloud like it used to be maybe in a process a year ago or 2 years ago.

JFrog is positioned to capture the opportunity both ways. Either they will stay an on-prem customer, and we will support them and grow with them there or they will move to the cloud, and we will do it in the cloud with them on a multi-cloud basis or one cloud. Basically, the hybrid solution that JFrog provided is the fit for purpose that they ask for.

There are no further questions at this time. I will now turn the call back to Shlomi for closing remarks.

Thank you, everyone, for joining us on this quarter results and earnings call. May the frog be with you. Take care.

This concludes today's call. Thank you for attending. You may now disconnect.

Ladies and gentlemen, thank you for joining us, and welcome to the JFrog Second Quarter 2025 Financial Results Earnings Call. [Operator Instructions].

I will now hand the conference over to Jeffrey Schreiner, Head of Investor Relations. Jeffrey, please go ahead.

Thank you, Nicole. Good afternoon, and thank you for joining us as we review JFrog's Second Quarter 2025 financial results, which were announced following market close today via press release. Leading the call today will be JFrog's CEO and Co-Founder, Shlomi Ben Haim; and Ed Grabscheid, JFrog's CFO.

During this call, we may make statements related to our business that are forward-looking under federal securities laws and made pursuant to the safe harbor provisions of the Private Securities Litigation Reform Act of 1995, including statements relating to our future financial performance, including our outlook for Q3 and the full year of 2025. 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.

You are cautioned not to place undue reliance on these forward-looking statements, which reflect our views only as of today and not as of any subsequent date. Please keep in mind that we are not obligating ourselves to revise or publicly release the results of any revision to these forward-looking statements in light of new information or future events.

These statements are subject to a variety of risks and uncertainties that could cause actual results to differ materially from expectations. For a discussion of material risks and other important factors that could affect our results, please refer to our Form 10-K for the year ended December 31st, 2024, which is available on the Investor Relations section of our website and the earnings press release issued earlier today.

Additional information will be made available in our Form 10-Q for the quarter ended June 30th, 2025, and other filings and reports that we may file from time to time with the SEC.

Additionally, non-GAAP financial measures will be discussed on this conference call. These non-GAAP financial measures, which are used as measures of JFrog's performance, should be considered in addition to, not as a substitute for or in isolation from GAAP measures. Please refer to the tables in our earnings release for a reconciliation of those measures to their most directly comparable GAAP financial measures. A replay of this call will be available on the JFrog Investor Relations website for a limited time.

With that, I'd like to turn the call over to JFrog's CEO, Shlomi Ben Haim. Shlomi?

Thank you, Jeff. Good afternoon, and thank you all for joining the call. As the JFrog platform becomes the system of record for all software packages, I'm pleased to report another excellent quarter for the company. We continue to execute with discipline, aiming to meet the growing market demand for a unified platform that enables trusted, scalable software delivery.

In Q2, JFrog software revenue was $127.2 million, up 23% year-over-year. Our operating margin was 15.2% in the quarter, demonstrating consistent execution, while maintaining disciplined strategic investments.

Cloud revenue for Q2 equaled $57.1 million, representing 45% year-over-year growth. We observed sustained cloud usage, while maintaining a strategic focus on converting customers with steady usage above minimum commitment into annual contracts.

Our enterprise focus, product and go-to-market bore fruit again this quarter. I'm pleased to report that our greater than $1 million customers grew to 61 compared to 42 in the year ago period, equaling 45% growth year-over-year. Customers spending more than $100,000 annually grew to 1,076 compared to 928 in the year ago period, equaling 16% year-over-year growth.

Our trailing 4 quarters net dollar retention increased sequentially, driven by continued adoption of our security product and solid growth across our customer base. Ed will discuss this in greater detail later in the call.

Entering the second half of 2025, we are confident that our focus on DevOps, security and MLOps aligns with modern software demands. This quarter's success was fueled by continued cloud growth, rising demand for our unified security solutions and a clear value proposition in the rapidly evolving world of Agentic AI and MLOps.

Now I will take a moment to spotlight what powered our Q2 success, highlighting the key drivers behind our strong performance.

First, on cloud growth. I want to address 2 items for JFrog Cloud, both our core business results and then talk about emerging market dynamics. Q2 delivered strong cloud results fueled by expanded annual commitments, helping our customers gain budget clarity and SaaS spending visibility, while growing our recurring revenue base.

Now with AI adoption exploding, CIOs are rethinking infrastructure at its core. The unpredictable cost of running AI at scale is forcing a shift from cloud first to fit for purpose. This might mean hybrid, balancing cost predictability with agility, compliance, security and control.

JFrog has been hybrid from day 1, a unified software supply chain platform that runs in the cloud, on-prem or both. We are not reacting to this hybrid demand. We've been building for it, giving our customers true freedom of choice. This strategic shift by companies might extend sales cycles of customers that were in the process of migrating workloads to public clouds, but it reflects a deeper enterprise commitment across infrastructures and setups.

As customers bet big on AI, they need a robust solution, but also cost predictability, and that's exactly where JFrog can help them build smarter for the future. We're heading into H2 confident and focused, aligned with customers' demand and build for what next, staying loyal to our disciplined derisked approach.

Next, to AI and machine learning. Over the past few years, JFrog has not only invested in making our platform the system of record for MLOps and AI-driven software delivery, but also partnered with the world's largest companies and leading organizations in the AI ecosystem.

Last quarter, we highlighted our partnership with Hugging Face to secure open source AI models, the launch of JFrog ML to enterprise customers as part of our unified platform and the growing adoption of JFrog by native AI companies using Artifactory as the centerpiece model registry in their software supply chain.

In Q2, we were honored to be included in yet another mega initiative in the world of AI when NVIDIA's CEO, Jensen Huang, announced their new enterprise AI factory at the NVIDIA GTC Conference. As part of this emerging standard for enterprise AI development, JFrog was announced as the cornerstone software artifact repository and Secure Model Registry for NVIDIA's initiatives.

Justin Boitano, NVIDIA's VP of Enterprise AI Products noted, "enterprises building AI factories need to manage the complexity of AI adoption, while ensuring performance, governance and trust. JFrog's unified software supply chain platform paired with the NVIDIA Enterprise AI factory validated design enables rapid responsible AI innovation at scale".

We also continue to gain traction with leading AI industry customers as referenced in Q1 as they rapidly adopt the JFrog platform as an infrastructure solution, a system of record for binaries.

Just as we previously became the registry and single source of truth for all software packages, containers and artifacts or in short, all type of binaries, we are on a mission to become the world's leading AI model registry, offering our customers comprehensive 360 coverage across the entire model life cycle.

To achieve this, we are deepening partnerships with AI industry leaders, expanding our support for the AI ecosystem and driving community standards for responsible ML and AI adoption.

As AI continues to transform the way we live and work, developer tool stacks are evolving rapidly, integrating code assistant tools to meet growing demands for speed and efficiency. But it's not just tools that are changing. The architecture of software products must now be designed around MCP servers that enable agentic interaction, giving AI technologies access to tools and the system that power modern application development.

JFrog is all in as an open platform, building a solution that deeply integrates into the AI ecosystem. Our commitment was also marked by the launch of the JFrog MCP server in mid-July during the AWS Summit in New York. This is not all.

As responsible members of the AI community and as a vendor committed to building trust across our customers' software supply chain, we recognize that as MCP adoption accelerates in the development world, so does interest from threat actors looking to exploit evolving MCP standards.

Our security research team recently uncovered and published significant exploitation risks tied to MCP usage, and we're proud to lead the charge in securing the AI community from these emerging threats.

Finally, I want to highlight our DevSecOps solutions. In Q2, we saw multiple customers wins continue to be driven by security with companies focused on the consolidation of tools, a model security and software package curation.

For example, during the second quarter, one of the world's largest telecommunication companies expanded its use of JFrog through JFrog Curation in a 7-figure deal. As part of their standardization and consolidation efforts across their DevOps and DevSecOps tool sets, they placed a strategic investment in our solution to enforce policies and act as a firewall for their software supply chain.

JFrog security solution are boldly transforming the industry, replacing legacy point solution tools with a unified software supply chain platform and blazing the trail with world-class research and cutting-edge innovations to deliver trusted AI.

Some of these innovations will be announced at swampUP. I'm excited to remind you that JFrog will be holding our Annual User Conference on September 9th and 10th in Napa Valley, California. We'll be announcing new products and strategic partnerships, highlighting new innovations and delivering new solutions to the market. We look forward to welcoming our community to swampUP, where the world of every of op standards is crafted.

With that, I'll turn the call over to our CFO, Ed Grabscheid, with an in-depth recap of Q2 financial results and our updated outlook for Q3 and the full fiscal year of 2025. Ed?

Thank you, Shlomi, and good afternoon, everyone. During the second quarter of 2025, total revenues were $127.2 million, up 23% year-over-year. Our strong performance during the quarter was a result of continued operational execution, driven by strength in our cloud revenues, accelerating adoption in security core products and ongoing demand for our enterprise-level subscriptions.

Second quarter cloud revenues grew to $57.1 million, up 45% year-over-year and represented 45% of total revenues versus 38% in the prior year. As Shlomi noted, our strategy remains focused on converting customers with usage above minimum commitments into higher annual contracts.

Our growth in the cloud was driven by momentum in the JFrog Security Core and conversion of customers with usage above minimum commitments into higher annual contracts.

During the second quarter, our self-managed or on-prem revenues were $70.1 million, up 10% year-over-year. Aligned with our cloud-first approach, we proactively engage our on-prem customers to migrate DevSecOps workloads to our cloud and enable them to capture even greater long-term value.

In Q2, 55% of total revenues came from Enterprise Plus subscriptions, up from 50% in the prior year. Driven by the ongoing execution of our enterprise go-to-market strategy and broader customer adoption of the JFrog platform, revenue contribution from Enterprise Plus subscriptions grew 36% year-over-year.

Net dollar retention for the 4 trailing quarters was 118%, up 2 points sequentially, driven by the adoption of our security core products and increased data consumption, resulting in higher customer commitments.

We continue to demonstrate that our customers view JFrog solutions as mission-critical to their software supply chain with gross retention that equaled 97% as of the second quarter 2025.

Now I'll review the income statement in more detail. Gross profits in the quarter were $105.7 million, representing a gross margin of 83.1%, in line with our guidance range versus 84.4% in the year ago period. The change in gross margin relative to the year ago period was primarily driven by the increased mix of our cloud revenues. We expect annual gross margins to remain between 82.5% and 83.5% in 2025 due to continued focus on cost optimization with the cloud service providers.

Operating expenses in the second quarter were $86.4 million, equaling 68% of revenues. This compares to $73.3 million or 71% of revenues in the year ago period. Our operating profit in Q2 increased to $19.4 million or an operating margin of 15.2% compared to $13.6 million and 13.2% operating margin in the second quarter of 2024. The continued balance between strategic investments and operational efficiency demonstrates our commitment to profitable growth.

Cash flow from operations equaled $36.1 million in the second quarter. After taking into consideration CapEx requirements, our free cash flow reached $35.5 million or 28% margin compared to $16 million or 15% margin in the year ago period.

Now turning to the balance sheet, wended the second quarter of 2025 at $611.7 million in cash and short-term investments compared to $522 million at the end of 2024. As of June 30th, 2025, our RPO totaled $476.7 million, a 75% increase year-over-year, benefiting from customers' multiyear commitments to JFrog's DevOps and security offerings.

And now let me turn to our outlook and guidance for Q3 and the full year 2025. While we are pleased with our strong performance in the first half of the year and see pipeline opportunities continuing to build, given the current macro uncertainties, we believe it is prudent to continue to exercise caution in our forward outlook.

Our updated guidance range suggests growing contributions from the JFrog Security Core, increases in cloud commitments and continued adoption of the full JFrog platform. We continue to derisk our outlook by excluding our largest opportunities given the uncertainty regarding the timing of customer deployments.

We estimate full year 2025 baseline cloud growth to now be in the range of 34% to 36%. Cloud revenue guidance continues to exclude any contribution from usage above our annual customers' minimum commitments. We continue to expect our net dollar retention rate to remain in the mid-teens during 2025.

For Q3, we expect revenues to be in the range of $127 million and $129 million, with non-GAAP operating profit anticipated to be between $16.5 million and $18.5 million and non-GAAP earnings per diluted share of $0.15 to $0.17, assuming a share count of approximately 122 million shares.

For the full year 2025, we anticipate a revenue range of $507 million to $510 million, representing approximately 18.7% year-over-year growth at the midpoint.

Non-GAAP operating income is expected to be between $75 million and $78 million and non-GAAP diluted earnings per share of $0.68 to $0.70, assuming a share count of approximately 121 million shares.

Now I'll turn the call back to Shlomi for some closing remarks before we take your questions.

Thank you, Ed. This quarter demonstrated yet another powerful example of JFrog's strong execution and even more remarkable given the adversity we faced. Our team in Israel, supported by the global JFrog team, worked under unimaginable threat, literally under fire during the recent conflict with Iran. This resilience and brotherhood are woven into our DNA, and I'm truly grateful and honored to stand alongside such a courageous and dedicated team. Together, we continue to live stronger no matter the challenge.

We continue to hold on to hope and pray for peace in the world and for the fast release of the 50 hostages held captive by the brutal terror organization from us. We deeply wish for their safe return home well before reaching 2 years in captivity in Gaza's underground tunnels.

JFrog's business remains strong. Our technology continues to lead and innovation is growing. We're making meaningful progress towards becoming the system of record for all software. H1 of 2025 was solid, and we're focused on building even more success ahead.

With that, thank you for joining our call and may the frog be with you. Operator, we are now open to take questions.

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

Can you hear me? Awesome. Congrats on a strong Q2. Shlomi, I wanted to talk a little bit about the evolutions, cloud first fit for purpose. If I rewind it back a couple of years ago, there was this sort of customer hesitation on self-managed data center offering and then sort of looking to secure budget for their cloud initiatives. Could you just talk a little bit more about the evolution that you're seeing now? I just want to sort of unpack your comments in terms of how it would translate into continued cloud growth versus potential self-managed growth going forward given the evolution that you were speaking to in your earnings script?

Absolutely. Thank you for the question, Sanjit. What we are hearing from our customers, especially those that were in the process of migrating workloads to the cloud, to the public cloud, is that there is a new type of uncertainty coming from the predictability or the ability to predict costs in the world of AI.

AI is being adopted rapidly. And therefore, you need to think or rethink, where your models will be, where the data will be, how you will train it. And therefore, just going blindly to the cloud is not a responsible move for them. So they take a bit more time to consider that. Obviously, this is more relevant to the high-scale companies. a small developer shop that are practicing AI will not start its own data center, but big organizations that are betting heavy on AI will think and rethink, where their data centers will be, whether it would be in the cloud, on-prem or hybrid as most of them are now talking about.

Understood. My follow-up question goes to a little bit about how some of the product portfolio decisions the team is making is translating to your -- both your growth and your partnerships within the ecosystem. And specifically, what I'm referring to is like the sunsetting of the JFrog Pipelines product, I was wondering to what extent has that improved your go-to-market collaboration and initiatives with some of the CI/CD players, including Microsoft GitHub? And is there any way where that's actually helping put more focus on the security part of the portfolio, which seems like it's also seeing some rising adoption. I'm trying to connect the dots there, but tell me if I'm over extrapolating.

Yes. So as you know, our growth is based on a very diverse portfolio, all sectors, all industries, all size of companies, cloud, on-prem, so different setups, but also different practices. JFrog is the only unified platform that provides DevOps practices, MLOps practices and DevSecOps practices. This means that we are active in 3 different [ course ] to accelerate our growth.

Specifically regarding JFrog Pipeline, the moment we found out that most of the world is betting on GitHub action and there are new CI practices coming with the trend of AI, we decided to be #1 in what we know how to be #1. Pipeline was not one of them. Therefore, we were focused on the execution and the growth that we established with the new core, mainly security and mainly the migration to the cloud. And I think it was a very smart decision because now, as you can see by the numbers, we are even more focused on the execution.

The other thing around it is that it opened door to better integration and better partnership with companies like GitHub, with companies that are doing CI/CD on top of the source code. And I think it's also accelerated the partnership and the go together to the market without confusing our customers.

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

Congrats. Shlomi, it would be great to hear more about the pipeline of large enterprise deals. It sounds like you had some in the quarter. It would be great to hear more about those, maybe about this Curation telco big deal. What do they find so compelling? And can that be a good indicator of deals to see in the second half?

Yes. Well, thank you, Andrew. Our pipeline is obviously being focused on executing big deals that are combining 3 factors. Factor number one, migrating to the cloud, some workloads and making sure that we are working with our customers in full partnership to have the right commitment. As you know, our guidance are based on commitments and not usage.

Second thing, the addition of security, holistic software solution for all the software supply chain in terms of security. One of the things that we noted is this big telecommunication company that added on top of JFrog Advanced Security added JFrog Curation as their firewall for binaries between the public hub and their internal software supply chain.

And this new emerged solution coming from JFrog -- and based on the same practices of every Ops is the MLOps, managing your models, 360 all the way to deployment as part of one unified platform. This all goes to the differentiator, super strong differentiator that we bring, which is Artifactory as a centerpiece, the model registry, the system of record for the platform. And on top of that, we build capabilities.

So in terms of the pipeline moving forward in the second half of the year, we are very positive. We keep the conservative way of looking at the pipeline and derisking the big deals that might move a week or a month here or there. And we are very positive and based our guidance based on what we've seen.

That's great, Shlomi. The security sounds like it had a big impact in the quarter. That's great to see. Was there some -- was that expected? Or I thought there were a lot of the renewals coming up in the second half. Did any of those pull into this quarter? And how are those renewals going so far, especially for those that were kind of on trial intro pricing and now you're trying to convert to bigger paying customers?

Yes. So we are very excited about what we see in the security landscape, not only that our technology is addressing the real pain and the real threat of today's software supply chain, but also the narrative of consolidating security around the platform and [ endpoint ] solution is catching up really, really fast.

In terms of renewal, I think that if you look at our retention rate, it's still very, very high. So it says something about all the renewals that happened in this year. And also in terms of the adoption of more and more security tools with more partners that are reaching out to JFrog and ask to partner with our security tools, we are very positive about 2025 as we were in 2024.

Your next question comes from the line of Miller Jump with Truist.

Congratulations on the strong results. I just want to stay with this idea of Artifactory getting used as the centerpiece and as a model registry for customers. Can you just talk about the consumption trends that you see when customers make this decision? Is there like a notable step-up or acceleration that's being driven there?

Absolutely, Miller. Listen, you guys remember that JFrog started as a software package and artifact repository company. And then 2014 to 2016, Docker changed every developer's life with bringing containers in. And JFrog became the biggest container registry of the world, the biggest Docker registry of the world.

What we are now facing is an amazing opportunity because AI models are yet another binary. So part of our strategy is to become the model registry of the world. And by that, we are not only supporting software packages, containers and all type of artifacts, but also models as the infrastructure of the primary asset of AI. That means that most of our customers, when they are now looking at the need for model registry because everybody is implementing AI.

When they are looking at the need for a model registry, they actually have 2 choices: a, to consolidate it around one system of record, one single source of tools, which is Artifactory with all other 30 packages or to have a stand-alone model registry, which will break the system of record rule to keep your company safe, secure and efficient when you build software and release software. That we estimate -- AI is a big, big world today.

But we estimate that, that will not only contribute to the stickiness of the platform and the expansion of the platform with all the add-ons that we added on top of Artifactory, it will also position JFrog as a centerpiece of your software supply chain because there is no primary asset, no other primary asset, but models in the world of AI.

And it doesn't matter if software is being created by human beings or by agents, it will still require a system of record. So to your question, we assume that the need for Artifactory and the fact that it's playing a centerpiece will only grow.

Makes sense. I want to stay with AI. Last quarter, you talked about a key AI technology leader that you landed as a customer. I'm just curious if there's any update on the work that you're doing with them and then how that ramped versus your expectations in the quarter.

We are extremely, extremely excited about the partnership with them. It was not included in the call, but I can tell you that they upgraded their subscription and doubled their bet on the annual basis with JFrog in just 1 quarter. So we are not just operating in the world of AI, but also security and DevOps, but AI is growing fast.

Your next question comes from the line of Kingsley Crane with Canaccord.

Congrats on the quarter. First question, so with AI enabling smaller teams to ship code faster, do you see a world JFrog becomes even more critical to smaller orgs and teams? And are you already starting to see that with some new customer interest?

What a great question, Kingsley. And I think that what we see today is that something fundamental has been changed. First of all, if you were a developer yesterday, you had your IDE, you had your CI/CD source code and JFrog, fine. Now there is a new piece. It's called code assistance. So it's not only you. There is a code assistance next to you.

And the second piece is the architecture that was changed as well with MCP. This is how you build platform today. In both cases, and with every environment that you will have, cloud, on-prem, this AI or another, this code assistance or another, you will still need a model registry, it's fundamental. And you will still need to manage your security around one flow, one pipeline. Therefore, we see JFrog stepping in, in a very kind of strong infrastructure role -- and we are betting on that and with the expertise that we built through our deals.

Great. That's really encouraging to hear. And Ed, really impressive cash flow in the front half of the year. Is there any reason why this year would be more front-end loaded than last year?

Yes. That's a great question, Kingsley. As you remember, we have many multiyear deals that we've closed. It's reflected in our RPO that happened in the second half of 2024, and that continued in the first half of 2025. And this is what's really driving the free cash flow.

In the second half of the year, as Shlomi had noted, we are derisking our largest deals. This is a big driver of our cash flow. Assuming that those deals come through, I don't see a change in our free cash flow from what we would anticipate going forward, but it's all contingent, of course, on the multiyear deals as well as these large deals.

But we remain very focused on profitability. We remain very focused on our free cash flow, and we've delivered strong cash flow in the past, and we continue to focus on that going forward.

Your next question comes from the line of Mark Cash with Raymond James.

Shlomi, if I could start with you. I wanted to ask on the overconsumption happening in the cloud. I think last quarter, you talked about how you didn't really see the budgets that may have been developers experimenting here and there. And you have another really strong performance this quarter and seeing success with converting to larger commits. So has there been a change in budgets? And are you seeing experimentation move to actual programs now?

Yes, Mark, so you remember very well. First of all, we see more usage, mainly around -- when you use JFrog infrastructure for your models, so there will be more data consumption, there will be more storage consumption. Same thing for containers and everything that has to do with AI, if you refer to this specifically.

You have to remember that our strategy is that every time that we see an over usage on top of the commitment of the customer, our team is being sent to offer a better deal for the customer and to get an annual commitment. This is how we guide you. This is how we preserve our conservatism and keep ourselves in line with the plans. But overall, we see a healthy consumption, not yet in the days of 2022, but we see a healthy growing consumption.

Okay. And Ed, if I could ask you one, just absolutely another massive RPO quarter that's actually accelerating, going to be lapping 3 very large deals this quarter in 3Q. Is there anything we should be considering from an RPO or cRPO dynamic when looking at quarter-over-quarter or year-over-year compares to make sure we're not getting too far ahead of ourselves in like a booking perspective for 3Q or the second half?

Yes. Thank you. That's a very smart question, Mark. And we don't always see a correlation between our cRPO or RPO to revenue. So we think revenue is a good indicator and the guidance that I've given you in the revenue is what I would use to forecast going forward.

But you have to remember, our RPO takes into consideration factors of multiyear, takes into consideration the timing of when those bookings happened. We had 3 of the largest deals during the second half of 2024. So RPO may be impacted if larger deals do not transpire, but by the way, derisked out of our guidance going forward. So RPO is a great indicator, but it's not the indicator that we would lead you to, continue to look at the guidance that we provide on our revenues.

Your next question comes from the line of Shrenik Kothari with Baird.

This is Zach Schneider on for Shrenik. Congrats on the strong results. So following up on a previous question with cloud adoption continuing and DevSecOps needs shifting earlier in the life cycle, how are you leveraging your hyperscaler partnerships for co-sell marketplace attach? Maybe what percent of new security wins in the quarter were sourced via these marketplaces or just driven by this channel influence?

Yes, Shrenik (sic) [ Zach ], this is Shlomi. I'll take the call and Ed, feel free to chime in. The collaboration with the cloud providers is very important because of 2 reasons: a, it helps us to accelerate deals, especially mega deals that are coming in, that mainly through the marketplace and with a lot of collaboration from AWS, GCP and Microsoft Azure.

We have strong relationship with them. Some of the deals we are accelerating together. This is not just co-sell and partnership. This is also kind of co-services even when we are going with the customer, scaling with the customer.

The second side of it is obviously optimizing the cost. As Ed mentioned on the call, we are optimizing the contracts with this hyperscaler. We are looking at the gross margin, being very responsible to how we grow in the cloud. So not just pushing the pedal all the way down, but also to do it in a responsible, smart way.

Great. Makes total sense. And then I guess switching gears a little. Obviously, it sounds like strong sequential growth in AI/ML package usage across the platform. Are there any new usage tiers, ingestion thresholds? Should we expect any pricing model changes in 2026 to really capitalize on this trend?

Yes. So what we saw actually in terms of the usage for AI packages, and I think you're probably calling out the Hugging Face and PyPI that we called out in the first quarter, it was more of stabilization or sustained usage on a quarter-over-quarter basis. It's still very early, and we're evaluating the monetization of AI is still in the infancy stage. But once that starts to mature, we'll certainly capture value.

Your next question comes from the line of Jason Ader with William Blair.

Yes. Can you hear me okay?

We hear you well, Jason.

Okay. Good to talk to you guys. Shlomi, I just want to take a step back. Can you talk about the impact of all the AI coding tools? I mean there's just so much happening there. Every week, it seems like there's a new announcement, and it's pretty powerful stuff, obviously. What impact has that had on the DevOps tool chain in general and on your business in specific? And if it hasn't had an impact yet, can you just talk about how you think it might play out over time?

Yes. As I mentioned before, Jason, there is a new creature in the tool stack of every developer now. It's called code assistance. If you're a junior developer or experienced developer, it doesn't really matter, no matter what language you use and no matter what code you are writing. This is part of your tool chain. And this item in your tool chain is now helping you to build faster and create more.

What you create more is more binaries. And therefore, we are happy about this change as long as these binaries are hosted in JFrog. So we are happy about that. It also means that there are all kind of new threats that are coming. We mentioned the MCP server. MCP is the #1 change in every company today. If you refresh your browser, you will see 5 more companies that added MCP to their tool stack.

The fact that it is being adopted so fast is also the biggest risk that come with it. The threat and the hackers are there. They know how fast this is happening. And every new company that adopt MCP architecture and the new code assistant tools, they are also subject to maybe a threat of a security solution. Therefore -- a security threat, sorry.

Therefore, we see how JFrog become more and more relevant, not only with the technology, but also with the holistic approach of building these layers of security and automation in -- on top of Artifactory.

Now remember, this is not just a one kind of partnership with one tool in the market. JFrog tools are agnostic to all integrations. So whether it's Cursor or Windsurf or whether it's MCP from this company or another company, JFrog is still the system of record for you.

Okay. Great. So I mean, just to round out the question, do you feel like it's benefiting you guys yet, all of this kind of front-end AI coding or it's still not moving the needle yet?

Well, for sure, there are new threats that are coming. And when you speak with the customer about security and what you need to protect, model security are coming on every discussion. So the answer is yes. It's still very early in the process. People are still trying to understand how this new environment looks like, but all of them are aware of the threat and therefore, open to have discussion with modern solution and not yesterday solution.

The second thing, as I mentioned before, if you have your Docker registry in Artifactory, your Python Registry in Artifactory, your Java repository in Artifactory, then why you would go with a model registry as a stand-alone. So my answer is yes. And still, I will stay humble and say it's still too early to say it will completely change everything we guided for in the next years.

Okay. Great. And then just if I could sneak one in for Ed. Ed, you guys have talked about security being a material part of the business in 2025. Can you just update us on that? And is -- will you be able at some point to give us some more granularity around the contribution of security to the JFrog business?

Yes. What we've committed to you, Jason, and we did this at the end of 2024 was we gave the metrics around our attach rates with security, and we plan to do the same thing during 2025. We see nice momentum. We're excited about what security can bring, but we'll give you an update at the end of the year.

Your next question comes from the line of Jonathan Ruykhaver with Cantor.

Can you hear me?

We can hear you now, Jonathan.

Okay. Yes. So the MCP opportunity, I'd just like you to talk a little bit more about that. I think it was 2 weeks ago that GitHub announced a critical flaw in its MCP server. And I think some attackers were actually able to manipulate AI agents into leaking sensitive data. So how do you see that solution not only today, but when you look at securing AI agent behavior relative to repositories, what is that opportunity for you? And will you see additional capabilities? It seems like from what I understand, it's a basic kind of entry-level offering, but there are a lot of other threats around protocols, around access for these MCP servers that still need to be addressed. So just talk about that position and what it looks like 12 months from now.

Yes, [ Jon ], MCP is the new way in the world of software products to interact with your product, to build the integration. JFrog philosophy from day 1 was the philosophy of too integrated to fail. Basically, there is no tool on the planet, on our planet, which is software supply chain that doesn't integrate with JFrog. CI/CD, security, databases, storage, because we became the system of record for all binaries. So it was based on APIs, and it was based on rest technology.

And now in the world of AI, when it's not only a human being that need to integrate with your tool, but also agents and machines, MCP is the protocol that will allow them to come in and integrate with JFrog platform or with any other platform. So therefore, MCP became a very, very important for every product provider. If you want the world of AI to interact with your tool or with your platform, you have to enable MCP at the front.

In terms of the other note, was everybody is using MCP and everybody is installing MCP and everybody is having an MCP user -- and that by itself is a threat that is addressed by hackers. So JFrog Research team also unveiled a threat and a vulnerability in the world of MCP, and we shared it obviously immediately with the public. That obviously will give us points when we come to AI security and trustable AI in your software supply chain.

Yes. And Shlomi, as a quick follow-on, where do you see this opportunity? Will it be aimed more at the foundational model providers like OpenAI and Anthropic or large enterprise or both?

No, it's not at all one company or another. Every agent you can assume that in the future or not in the future, it's already in the present, but it will exceed in the future. Not only developers will write code, but also agents. So if I want this new persona to see JFrog as a system of record, I have to open the door for it. So every type of Agentic AI that will interact with JFrog will come through an MCP integration.

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

This is Eamon Coughlin on for Raimo, A great quarter. Can we dig a little deeper into linearity in the quarter? Should we view 2Q's performance broadly as a continuation of the customer usage from Q1? And did you see a continuation of these trends in July?

So first, let me just start by saying I will not comment on anything in the third quarter. I'll go back to the second quarter and give you an update on what happened. There was 3 dynamics with our cloud. First was security wins. We're starting to see customers uptake in their security and big security wins, those tend to land in the cloud. So that was dynamic number one.

The second piece was the usage. We had customers that saw benefit of taking a larger agreement, annual agreement with JFrog. These were users that were using over minimum commitments, and we secured those. The confidence that gave me is reflected in my guidance that I gave to you in the 34% to 36% on the cloud.

The third piece is the sustained usage. So for those customers that were using over minimum commitments, that has been sustained on a quarter-over-quarter basis.

Got it. That's super helpful. And then great to see continued cloud growth acceleration in the quarter. But can you help us understand the key drivers of your subscription self-managed portfolio?

Yes. Why don't I answer here? Self-managed, we addressed that on the call as well. There is a new trend now that AI might need a hybrid solution. Now what can be better than an identical product, hybrid means that it's identical in the cloud and on-prem. So that's a new trend, mainly driven by the AI adoption and the seek for cost predictability and stability.

The second thing that we see in the on-prem is just like what Ed mentioned in the cloud. Our on-prem customers are also looking for security. So the fact that we have our security available for you in the cloud, but also as a self-hosted or private cloud is also a benefit, and we see that this growth generated by these trends as well.

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

This is George McGreehan on for Koji. I just wanted to ask with the strong momentum we're seeing in security deals and how that's contributing to forward-looking metrics like the acceleration in RPO. What would you say -- is there any change in the competitive landscape for security? Any notable changes in win rates? And how really are the tone of conversations with customers now when you're talking about security deals versus maybe in the past a year or 2 years ago?

Yes. Thank you, George, for this question. It comes in 2 different ways, as I mentioned, and as you know, on the JFrog go-to-market strategy with security. First of all, our customers, whether it's the CISO or the CIO or both, they are asking to look for consolidation, and they will push very hard on it, not only because of the cost benefit from it, but also from the enormous number of scanners that they used to have just 2 years ago, 3 years ago.

Every developer bought another tool, every compliance group brought another tool. It became a zoo of scanners, and they are looking for consolidation and obviously also benefiting from the efficiency and the cost perspective. This is more business side and management side.

What happened on the technology side is that there are a lot of new threats that cannot be identified by the previous software that covered software supply chain security. If you remember what we spoke about in the previous quarter, the fact that we scanned the entire Hugging Face 1.5 models and provided not only vulnerability findings, but also contextual analysis, telling you not only that we found this vulnerability that maybe every scanner can find, but also how exposed are you to this risk and if you should waste time on doing it.

And the third thing is that modern security is a new thing. Nobody knows really how to cover it. And since models are hosted at JFrog and also the data, the metadata that you train models with, we have better access to this asset and better visibility to this asset. And I would welcome you to swampUP when we will speak about it even more and even share some great innovations with you to address this pain.

There is no CISO in today's world that is not waking up at night several times because of model security. And I think that this is another driver that pushes JFrog to be a relevant player, not only in the world of DevSecOps, but in the world of MLSecOps as well.

Your next question comes from the line of William Mandl with KeyBanc Capital Markets.

This is Billy on for Jason Celino. Shlomi, it sounds like there's strong demand for the security core, but curious what customer reception has been around runtime security, maybe how that is contributing to the strength you're seeing in your security business?

Yes. Billy, I think that cyber is important. Run time is important. And DevSecOps and software supply chain is important. This is a complementary solution. We were all very excited to see the acquisition of Palo Alto, which emphasized their advantage in the world of cyber and run time. And if you look at the growing demand for software supply chain security, it's coming from a different threat and a different need. And I think that they will coexist together.

Your final question comes from the line of Rob Owens with Piper Sandler.

Thanks for squeezing me in. Obviously, very exciting about the security success that you guys are seeing here. Curious if you look at those deals, I know you're going to give us more information around attach rates at year's end, but any sense that you can give us in terms of what this is doing to pricing, especially as you're seeing certain customers maybe take that end-to-end portfolio?

Rob, so yes, as you've mentioned, we will provide more details at the end of the year as we did in 2024. Pricing and subscription model is something that we are looking at, especially now when there are some innovations and new offerings that we will announce at swampUP. There might be new packages for security and new packages for advanced adopters of our security solution.

And we also have to think about the fact that JFrog started to bring the software supply chain security not too long ago to the market. So we have to be focused on the adoption. We have to be focused on growing within our portfolio with a number of logos. So the entry point is also important. But stay tuned. There are some announcement that will change some of the subscription package on the security layer as well.

There are no further questions at this time. I will now turn the call back to Shlomi for closing remarks.

Thank you, everyone, for joining the call. We are very, very excited about the momentum. We also appreciate all of your questions and may the frog be with you.

This concludes today's call. Thank you for attending. You may now disconnect.

Everybody. My name is Koji Ikeda. I am one of the software analysts here at Bank of America. Welcome to day 3 of our technology conference. I am absolutely thrilled to have JFrog doing a fireside chat with us. We have the CFO, Ed Grabscheid. And we also have IR, Jeff Schreiner here. So thank you so much for being here. We appreciate it.

I always ask the obligatory introductory comments or a question of what is JFrog? What do you guys do? What problems are you addressing today? And what problems are you addressing for the future?

Well, first of all, thank you for having us. It's great to be back here in San Francisco. I see some new faces in the audience here. So I'll give a background at JFrog. JFrog was brought to this world to make developers more efficient. So as code, source code becomes a machine language, that's a binary. JFrog manages the binary. So today, we are the only company in the world that has binary and Artifactory as a platform.

We also have DevSecOps, which is our security product and now JFrog ML. So 3 applications in a platform, only company in the world to be able to do that. We're focused today primarily on managing those assets, ensuring that they are brought into your organization, they are secure and then they are delivered in the form of an update. We also will start working in the future with large language models. It's very nascent technology today, but the focus is on delivering value for the AI and large language model world in the future.

You guys dub yourselves as the one of the disruptors and protectors of the supply chain of software. What exactly does that mean for being the supply chain of software?

Yes. The supply chain of software is really the motion of taking a language that's written in a source code in English or Spanish or German and putting that into machine language. And what is different today than what we saw maybe a decade or 2 decades ago was the pace of updates. So if you recall, maybe if you use Microsoft and you did an update in Microsoft, you would do that once a year or every other year, they would do an update. They can manage that update. Somebody would write the code, they would turn it into a language, they would test it. They'd bring it to market.

There might be an error that's found, they would fix that and maybe 6 months down the road, they would update that again. Today, updates are being done 10, 20, even 100 times per day. The pace of those updates because of open source packages coming into your organization today source code, let's say, 80% to 90% of that is an open source package. You're only writing 10% to 20% of your code, but it has to convert into a binary. Software supply chain and the management of that binary was brought to this world because of these open source packages, and this is what JFrog does. It manages the process of taking those binaries, those updates and bringing them through your organization, testing them, securing them and distributing those binaries, those updates.

When I was preparing for the conference, and I was kind of looking at the first quarter -- my first quarter result notes, I was kind of reviewing them, I'm like, man, you guys had a pretty good quarter. And so tell me a little bit about the quarter. What was kind of the highlights of the quarters? And what have investors have been kind of asking about the quarter that they've been pointing to from a bull case.

Yes. So Q1 was a culmination of not just one quarter. It was a culmination of many quarters that we built to the results that you saw during Q1. So we really saw this transformation that was happening early on where we wanted to penetrate at the C-suite. We saw there was an opportunity to move from a developer sale into the enterprise. We invested heavily 2 to 3 years ago with secure -- I'm sorry, with enterprise, with strategic sales, with building the infrastructure around that architects and solution engineers.

Now Q1 is a result of those efforts, the efforts of the acquisition of VDoo, where we have security and security going from a point solution to a platform and adding that on top of Artifactory. And then in Q3 of last year, closing 3 of the largest deals in the history of the company. 3 or 4 years ago, if we did a $0.5 million deal, that was considered a mega deal within JFrog. Today, we've closed deals that are 8 figures. We have a customer that's over $30 million on an ACV value.

So when you say Q1 was a great result, that wasn't because it was just one quarter. These were multiple quarter efforts, and we really started to see that momentum happening in Q3 of last year with these large wins with security. If you recall at the end of 2024, we disclosed the metrics in security. We delivered more than -- or we delivered 3% revenue, 5% of ARR and 12% of our RPO, which is coming from security, where we had essentially 0 in the year before.

So we've really started to penetrate the security budgets as well. And then the last piece of what we saw during Q1, which was a bit of a surprise to us was the amount of usage in our cloud. So we have customers that commit to JFrog minimum commitments in data consumption. So they may commit to 2 petabytes, 10 petabytes, 15 petabytes, whatever that usage is, and we take that revenue ratably. When the customer exceeds those minimum commitments of usage, then we have an overage.

In 2024, the vast majority of our customers were not spending or using above the minimum commitments. In Q1, we had this robust usage across a diverse set of our customers in our portfolio across multiple geographies and distributed linearly across all 3 months of the quarter. So this was something that was not expected. It's historically a slow quarter, Q1 as many budgets are being built, as sales kickoffs are happening. So it was a bit of a surprise to us, but it ended up being a great result for JFrog.

I want to focus on the large deals were signed in. Just a point of clarification, make sure I heard it right? You said $30, you have customers that are $30 million in size on an ACV basis. That's...

We have one customer that's $30 million in size on an ACV basis.

Wow. Okay. Okay. And then on the cloud usage front, got asked the question, right? I'm sure you've been getting the question a lot. AI coming into play with that? I mean, is there any sort of -- you guys should be a beneficiary of AI driving more binaries good for JFrog? I mean I don't want to get ahead of my skis here, but it does feel like is AI playing into that just a little bit.

Well, we just -- we're going through the circuit. This -- we were telling you before the fireside chat, how many of the fireside chats we've had. So you're not getting ahead of your skis because this is the third question. We typically get it on the first or the second around AI.

So waited a little bit.

Yes. Well, let me tell you a little bit about the usage. We -- well, first off, AI. What we believe and why we're so excited about this opportunity. When you bring a large language model into your organization, it's a binary. And we are the company that manages the binaries. So we believe we should win this market. We made an acquisition at the end of Q2, we acquired a company called Qwak AI. That is the platform that manages large language models to bring those models into your organization to train, to secure and distribute large language models.

We released the first phase of the product, which is on the cloud in Q1. And we're planning to release the self-hosted version at the end of this quarter. Now what are we seeing? I'll be honest, we're not a company that's going to deliver a message of fluffiness. It's very early.

If we talk about a baseball game, we're singing the national Anthem at this point. We're not even exiting the dug out to start playing the game. We see some things that would believe -- lead us to believe that there could be a tailwind. And the thesis around it is that more binaries means more code means more binaries and more binaries is going to be beneficial to Jfrog.

Now in Q1, during this overusage period, we did track the packages that were being used by the customers. And we saw 3 packages that seem to have the most significant usage on a sequential basis. That was Docker, Hugging Face and Python, PyFi. So when you think about what the developers are using, you could make some assumptions that there's experimentation that's going on right now with AI.

You mentioned Hugging Face. I believe that's a customer or a relationship that you highlighted on the last quarter call, you mentioned it right there. What's going on with Hugging Face? Why did you guys talk about it so much on the call?

Yes. So, I'll take that question, Koji. I think the Hugging Face relationship is unique in the sense that it's a repository for large language models, much in the same fashion that you would garner a repository for NPM Maven on the package side that we've already been doing with Artifactory today. And so when you look at what we've been doing here and what was the -- I apologize, what was the topic again?

It was the Hugging Face...

I apologize guys, a long week. I apologize. Hugging Face is a relationship in which they came to us, Koji. They had another company that was reviewing the repository and securing that repository, but that company had told them that 80-plus percent of the models were malicious, okay? And so they asked JFrog to come in and scan those models. And we said that those were false positives. So in essence, maybe some of that Hugging Face traffic that Ed alluded to could be that JFrog customers felt that this was a more secure repository from which they could pull from.

So the Hugging Face relationship is us supporting the community today and supporting that repository. But obviously, there could be indirect benefits as customers start to utilize that Hugging Face repository and use the proxy function that's within JFrog Artifactory today. So we view that as kind of the first motion of what you're seeing, where the repository is more focused on an LLM than it is on an individual package technology.

And I think there -- I'm going to just jump in real quick. I think there was something very important that Jeff mentioned was that Hugging Face came to us, JFrog to scan and ensure the confidence that the models that are being stored in Hugging Face to pull into your organization are secure. So that brings a lot of confidence to the community that JFrog is the system of record in terms of the scanning and the level of confidence that these models are secure. So even if you're not a customer today of JFrog, you know that JFrog is going to be the system of record in terms of the scanning capabilities.

Therefore, if you're going to bring a large language model, and this is where the intangible benefit that Jeff talked about, and you're going to bring that as a data scientist into your organization, it makes sense to store that in JFrog. So we believe that there could be some opportunity coming from that. Today, it's not a monetized relationship. It's a matter of a proxy between similar, like Jeff said, let's say, Maven or NPM but it's a relationship of bringing large language models into your organization through a proxy, but we do believe it could create some intangible benefit for us.

Got it. Let's talk about the guidance for a second.

Sure.

You guys -- we just talked about big deals. We talked about cloud usage. But how does that I mean I know -- let's talk about it. We know it's not incorporated into the guide, right? You guys are not including big self-managed migrations to the cloud in the guide, and you're not including usage, upside usage in the guide. So why is that not included in the guide this year?

Yes. Look, first, let me explain why we chose to do this. And I talked a little bit about the transformation of the company 4 or 5 years ago when we went public to where it is today. A deal of $0.5 million, if that deal pushed out, I could pull 2 or 3 deals from an out quarter into the quarter, and it didn't have a significant impact on my revenue. Today, when I take an 8-figure deal and that deal, because of the complexity of the deal, if I forecast for that deal to happen in a quarter and it pushes out, I don't have deals tens of deals that I can pull into the quarter to backfill that push.

And so what we see with these large complex deals that you're talking about that are a migration that have security elements on top of it, it's not a matter of if. but a matter of when. And we must be patient. We saw this in Q2 of last year, where we had forecasted a deal that was 99% of the time would close. It didn't, and it had a significant impact to our ability to overachieve on the revenue side. Therefore, we took a very cautious approach by derisking those deals and allowing those deals to mature and then come in as upside potential.

The second piece is the usage. And why don't we include the usage? Well, we saw this in Q1. The usage created a tailwind for us. It created a significant outperformance in our revenue, but they can turn it off as well very quickly because it is not a commitment. So today, we have a bit of a disconnect between the developer that's been pushed to innovate to bring in new technology. And the other side, my office of the CFO and procurement that say, hold on, I don't necessarily have the budget yet. So let's maybe ratchet down the usage until we can secure the budget. And so those discussions right now are a bit of a tug of war. Innovation that's happening, budgets and a very tight rigid purchasing environment that's pushing back. So it doesn't make sense for us to put that type of assumption into our guidance. So we exclude that.

The last piece is, as we stepped into Q2 coming off of our guidance in February, it became incrementally uncertain. The market changed. You had tariffs on, tariffs not, the liberation date. There was all kinds of things that were happening. We had to take a more cautious approach, and we balanced the overperformance in Q1 with our prudence, and this is why we took a more conservative approach going forward. And we believe that, that's the best approach, not only for us but for investors and shareholders.

I'm going to try. It is June. So it's been a while since you guys reported. Any sort of update in the demand environment you could share with us?

It's worth the effort to try because you will never know unless you try, but no, I will provide an update after we close Q2.

That's right Yes. Competition, I wanted to ask you a bit about that. You guys are -- we consider you guys next generation. But JFrog is not new. You guys were made last year, year before, 5 year you've been around. And whenever we do our checks, it seems like the pool of competition is quite small for you guys, and you guys are one of the disruptors and leaders there. And so why hasn't another competitor emerged? What is it about the category that is difficult?

Yes. That's a great question. So let me give the competitor landscape first, and then we can talk about the differentiators. We are the only publicly traded company in that DevOps binary management tool infrastructure space. The closest competitor is a private company, a PE-backed company called Sonotype. We also compete with a very, very small start-up that is a cloud-native tool. We don't see them often not even worth mentioning the name, but that they are sub-$10 million. So we've seen them kind of pop up. We're not naive to think that somebody could come into that market.

We also have the hyperscalers. They offer some type of container registry, very basic container registry technology as well. So what is the differentiator for JFrog and how we built the moat and actually probably widen the moat is through our technology and the number of languages we support. We have a very deep technology stack in terms of the languages that we support, the security elements. At the end of the day, when you own the most critical asset, which is the binary. So JFrog owns that binary. It's being stored with Artifactory.

You are the company that can secure that asset as well. And so now we've created a differentiator with our platform in security. Point solutions today cannot operate on the security of binaries without having a proxy to JFrog. So we want to please the community and continue to delight the community. So therefore, you maintain those relationships, but essentially, you could cut them off. You could cut the oxygen off to these point solutions. We choose not to do that, but most customers see advantage to 2 things: consolidation, and to secure the most -- the critical asset, which is your binary. And so as we continue to add more technology like MLOps, that moat becomes deeper and wider.

And I'll just add quickly, Koji, I think on your question about why there haven't been more. This is a fairly new technology. And I think the time that I've been at JFrog now almost over 3 years, I think the importance of the binary has grown. And so I think part of that is that it's taken time to get to a level of sophistication and software development, where individuals that are working on this technology are realizing the importance of the binary asset. And so I think that's the unique part because I think binaries are the newest form of the software development software supply chain [indiscernible] been there, observability production environments have been there. This is kind of a new aspect to the whole development process.

You mentioned 2 things in your answer there that I wanted to touch upon. One MLOps. But two, you mentioned the hyperscalers and container registry. And so maybe good knowledge for me. I just kind of want to understand when we think about the hyperscalers do customers sometimes start with the hyperscaler option and then eventually graduate to JFrog? Or do they start with JFrog from the beginning and the ones that are on that container registry from an AWS or a hyperscaler, they just kind of stick with it over time?

Yes. We don't necessarily see migrations coming from the hyperscalers to JFrog. What we would typically see if they're sophisticated, they're coming either from a competitor like Sonatype that have scalability issues or want to migrate to the cloud and are unable to do that or they're coming from a homegrown tool where they've built something to manage their binary -- manage their binaries and distribute the binaries. They have a large organization. They're unable to scale on the homegrown tools.

So they come to an automated tool like JFrog provides. You probably have smaller organizations that are using the hyperscalers for container registry that maybe want to operate on 1 or 2 languages. Those are not necessarily the right customers for JFrog. We'd happily take them, but because of the depth of our technology, the number of languages that we support, the complexities that we support. So less sophisticated organizations typically go to the Hyperscalers. They have different metrics in us.

They're looking at trying to generate as much traffic as possible. So they -- by the way, we operate and work on the marketplaces of all 3 of the hyperscalers, and we work very closely in a partnership with those hyperscalers. They're really looking for traffic. They're not looking to take the business from us, but they would happily bring a customer into their container registry solution if it's a very simple basic case.

Okay. What's the opportunity with MLOps for you guys? Tell us a little bit about that?

Yes. I'll start, and then Jeff, you can talk about the technology piece sure. As I mentioned in the beginning, we acquired Qwak AI in Q3 of last year. We saw this as a huge opportunity for JFrog at that time because we said, as this market is shifting more towards large language models, we're creating -- there's a new buyer here in the data scientists. The data scientist is now creating these large language models.

In order to deploy those large language models, they become a binary. So they have to train them, they have to secure them, they have to deploy them as a binary. And we wanted to win that market. We saw it shifting, and we made that acquisition in Q2 of last year. We also saw that we wanted to get ahead of it because we were worried that valuations could change very, very quickly, which they have.

So when we did the acquisition, maybe there was questions around it, why are you doing the acquisition now? We did not get questioned at all around valuation, around the decision to acquire, I'm sorry, Qwak AI. And the last piece of that is we saw an opportunity here with a product that we can integrate into our platform very quickly. Unlike the acquisition we did with Vdoo on the security side, where it took a significant investment in terms of building the application to be a platform and integrate. With JFrog, this was a product that we could add to our platform very quickly. As you saw, we released the cloud native version in Q1, and we're getting ready to release the self-hosted version this quarter. So it was the right decision from us. And we believe we should win that based on the binaries.

Yes. And to Ed's point, when we talked about singing the National Anthem as a comparison to the baseball game of where we're at. What we have right now is that a lot of companies are tinkering and experimenting with ML in their organization. Perhaps we think that's some of what we saw in the overusage in Q1. And what they would really like to do is look at the various platforms that they could use within their organization. So to Ed's point, I think another key aspect there is that within 6 months, we were able to take this technology and meet the market when it's ready to start utilizing and bringing this into organizations.

I would say that we view that there are 2 gating factors right now before we see mass adoption of this technology across many enterprises. And that's within the industry determining what the monetization will be, how will JFrog be compensated for the value we provide and what is the buyer willing to pay for that value.

And on top of that, given the rapid change, the speed of innovation that you're seeing on the MLSecOps side and what AI can allow that hacker to try to do and bring into the organization, that will be another gating factor to a much more broader adoption. So we think that there's people that are working within this today, but I don't think that we're really at a mass scale yet until some of these other broader gating factors are really relieved.

Got it. Maybe a question here on the M&A strategy going forward. I mean you guys did buy Qwak last year, you bought Vdoo couple of years ago. I think there's a couple of other acquisitions in between. Keep me honest there, I can't remember. Yes, how do we think about your M&A strategy going forward?

Even before I talk about the M&A strategy, one of the strategies that we have is to focus on delivering free cash flow so that we have the capability of being able to react quickly. So we delivered 26% free cash flow margin in Q1. We continue to generate a significant amount of cash. This gives us that flexibility as when to a shift in the market that we can react very quickly and do a transformational or even a tuck-in type M&A acquisition. So we continue to focus on that to ensure that we're nimble and agile. We're always looking at the market.

I don't know that we'll do something transformational per se. But as we release JFrog ML and customers begin to use the platform, feedback will start to come back to our engineering team. We'll identify if there's holes in the platform and see how we would go and address that. That could be with the tuck-in. But most of the focus today would be on ML and AI, not so much on the security side. We believe we're well positioned in the security side. We've invested heavily over that. in that domain over the last 3-plus years and the focus will really be about delivering AI technology.

Let's talk about security for a second. So you guys have delivered good revenue from security from essentially 0, right? And so how do we think about security as a growth vector over the next 12 to 24 months?

It's a big opportunity for us. There's big budgets in security, and we want to continue to penetrate that, but we're still at the early stages of penetration. We don't have -- we have thousands of customers, over 7,000 customers today. We've got more than half of those customers using JFrog X-ray which is our Tier 0 security product. So we have an opportunity to penetrate through those customers that we know today are using X-ray. You need to use X-ray to be able to use our advanced security products. So our focus is on that. We believe we have a long runway. It should be a huge growth driver for us going forward, and we continue to see that as a big piece of our business in the future.

Okay. How do we think about how you're selling security? One of the things that we is often debated within DevOps is the ability for a dev tool to sell into ops or security, security tool to sell the other way, Ops, you get where I'm going. So how are you selling security?

Yes. It's interesting. 3 years ago, when we started to build our security team, we built a security overlay team. We brought security-based architecture and solution engineers, and we've integrated those into our process. So we have that overlay team. But we also saw a shift in the way that the customer is buying. Before you had the DevOps side, you had the CISO and the CIO, 2 separate budgets. You're now starting to see that come together as a panel. So when we sell, it's a matter of selling to a panel. You've got the developer that you're selling to. You've got the CIO and the CISO budget. Those are coming together to make a decision. So you have one point of contact in terms of your sales now that is selling an opportunity, the value proposition of JFrog, the DevOps side, the DevSecOps side brought together as one value.

Got it. Got it. got about 1.5 minutes here, and we briefly touched upon free cash flow margins. You talked about how you guys think about free cash flow and M&A. But how do we think about free cash flow overall? I mean you guys just did 26%. Your targets are 26% to 29%, right, you're kind of there. So how do we think about free cash flow generation and balancing that against growth, growth potential?

Yes. That's a great point, balance. I'm going to focus on that word. We've always had a focus on profitability and free cash flow. This was not something that was outside of this year or last year. Today the shift is going to be now on profitability and free cash flow. That's always been part of our DNA. In fact, 3 or 4 years ago, when the focus was around profitable growth and durable growth, we were being questioned about that. Today, there is no question. Companies are focused on profitability.

We continue to balance innovation with profitability. I'm not going to spend $3 to earn $1. That's never been part of the way that we operate as a company. I've been with the company for 6 years, CFO for 1.5. That will remain part of my DNA, and we will continue to look at generating value through free cash flow and profitability.

Yes. And I'd just add quickly, Koji, we think that as a Rule of 40 that is driving what Ed and I and the management team at JFrog are trying to do that now that portion of the Rule of 40, a greater portion of that is coming from the free cash flow margin. And so that was something that we thought would happen over time a few years back and the focus that we've put on that metric.

Got it. Jeff, Ed, we are all out of time. Thank you so much for doing this. This has been great, fun conversation. Thanks so much for being here. Thank you.

Thank you for having us. Thank you. Of course.

Thank you.