Akamai Aktie

Good afternoon, and welcome to the Q3 2026 Akamai Technologies Earnings Conference Call. [Operator Instructions] Please note, this event is being recorded.

I would now like to turn the conference over to Mark Stoutenberg, Head of Investor Relations. Please go ahead.

Good afternoon, everyone, and thank you for joining Akamai's Third Quarter 2026 Earnings Call. Speaking today will be Tom Leighton, Akamai's Chief Executive Officer; and Ed McGowan, Akamai's Chief Financial Officer.

Please note that today's comments include forward-looking statements that include revenue and earnings guidance. These forward-looking statements are based on current expectations and assumptions that are subject to certain risks and uncertainties and involve a number of factors that could cause actual results to differ materially from those expressed or implied. The factors include, but are not limited to, any impact from macroeconomic trends, the integration of any acquisition, geopolitical developments, and other risk factors identified in our filings with the SEC. The statements included on today's call represent the company's views on November 5, 2026, and we assume no obligation to update any forward-looking statements.

As a reminder, we will be referring to certain non-GAAP financial metrics during today's call. A detailed GAAP to non-GAAP reconciliation is available in the Investor Relations section of akamai.com under Financials.

With that, I'll now hand the call off to our CEO, Dr. Tom Leighton.

Thanks, Mark. Akamai continued to make outstanding progress in the development of our Cloud Infrastructure Services or CIS in the third quarter. Most notably, we signed an $11.6 billion commitment over 7 years with Anthropic to support their CPU workloads with our distributed infrastructure and software. This brings the total of multiyear commits for CIS signed this year to more than $14 billion.

The total includes an earlier large contract with Anthropic, 2 large deals with leading AI companies, and numerous multimillion-dollar deals with major enterprises around the world.

I'm pleased to report that we're making good progress in delivering on these contracts. Two of the first 3 large deals are now generating revenue, and we anticipate that the third will begin generating revenue later this quarter. We did encounter some modest supply chain and third-party data center delays with 2 of the deals and so our revenue outlook for Q4 will be less than expected.

These issues have since been resolved, and they won't impact our outlook for 2027. In fact, we now anticipate that our CIS revenue will more than double in 2027 to $1.1 billion to $1.3 billion and that Akamai's overall revenue growth rate will dramatically accelerate as a result, with total 2027 revenue growing 17% to 22% in constant currency over 2026 levels. And we expect further acceleration in 2028 with our overall revenue growing beyond $7 billion.

As CIS becomes a larger portion of our revenue, we expect it to have a strong positive impact on our margins and EPS growth. In particular, we believe that the large deals we've signed this year will contribute an aggregate adjusted EBITDA margin of about 75% and a non-GAAP operating margin of approximately 30% when they're fully ramped.

There are several reasons why we think we can operate our CIS business with strong margins at scale. First, it's important to understand that we're not talking about bare metal here. In fact, we don't even offer bare metal as a service. We sell compute capacity through our Infrastructure as a Service platform, which consists of virtual machines, managed Kubernetes, storage and networking. This keeps our capacity elastic across customers and generate substantially more revenue per megawatt or dollar of CapEx than a bare metal contract.

In particular, we expect to generate about $22 million of revenue per megawatt of contracted power for the large deals we signed this year. Moreover, we're seeing that as we've grown our presence in the market, the pricing we can command has improved. In fact, the unit pricing in recent deals is much higher than in comparable deals we signed earlier in the year.

We also have several cost advantages because of our expertise and the unique nature of our cloud platform. For example, we operate one of the world's largest backbones, larger even than many Tier 1 carriers, which reduces our networking costs and allows us to leverage scarce data center space that may not be accessible to others.

Our globally distributed platform enables us to deploy compute into a multitude of smaller locations instead of just a few large data centers. This is another factor in helping us get the power and colo we need to grow and at a lower cost than using very large footprints in expensive core data centers.

We've spent many years optimizing our software and operations, so that we can more efficiently handle large volumes of customer data. We're one of the largest when it comes to moving data on the Internet since most of the world's largest brands, including several hyperscalers use Akamai to secure and deliver their content.

An even more important reason that we can sustain good margins at scale is the types of servers we use in our cloud. While we support a variety of GPUs for our customers, the vast majority of servers that we're deploying, well over 90% are CPUs. In fact, most of the same type of CPU that we use for our security and delivery services. This is important because CPUs are becoming much more widely used for AI. That's because they tend to be more versatile, more cost-effective, and longer lasting than GPUs. The CPUs we've deployed in the past are continuing to provide value across our platform even 7 to 10 years later. Indeed, the largest cloud contracts we signed this year guarantee revenue for these CPUs for 7 years. Being able to use the same servers for compute, security, and delivery further increases their useful life span and makes our platform more efficient at the same time.

CPUs generally require much less power than GPUs. And so the number of megawatts we need to power our cloud is much less than providers like the neoclouds who rely more heavily on GPUs. This means that we spend more CapEx per megawatt than the neoclouds, but that's a good thing for Akamai shareholders.

Power is scarce and expensive, and we need less of it in relation to our CapEx spend than others need for their infrastructure. There are many reasons why we think our rapidly growing CIS business will be very profitable for Akamai. But probably the most important reason is that we have a compelling and highly competitive solution in the cloud marketplace. Not only does this help with profitability but it helps ensure that Akamai will have enduring growth, even if resources that are in short supply today like power, equipment, and data center space, become more plentiful in the future.

There are many reasons why the world's most important companies are choosing Akamai for their AI and cloud computing needs. One of the most important is that with more than 4,000 points of presence in 130 countries, we can get our customers' compute workloads closer to users and devices wherever they are in the world.

And being closer means that we can provide lower latency and higher throughput which is important for a wide range of use cases such as interactive video, avatar rendering, conversational voice agents for customer service, speech recognition, translation, robotics, video processing, and real-time alerting. This is a key reason why even the hyperscalers use Akamai cloud for some of their latency-sensitive applications.

Another big advantage for Akamai is that we can package our compute solutions together with our market-leading solutions for security and delivery. So enterprises can use one platform to power, protect, and deliver their most important applications with great performance and vast global scale. This also helps with cross-sell. We can sell compute into our large base of security and delivery customers and vice versa.

We can sell our security and delivery solutions to new cloud customers. Being able to secure our customers' workloads is especially important in the age of AI. Not only can we help protect our customers from AI-enabled attacks, but we can also help keep our customers' agents from going rogue and causing unintended harm.

Reliability is another key advantage for Akamai when enterprises are choosing a cloud provider. As a great proof point, we helped the world's largest financial institutions achieve 99.999% of uptime, which means less than 10 minutes of downtime in a span of 2 years. This is not easy to do. In fact, it's very hard and takes extraordinary expertise, dedication, and experience to achieve. Akamai also provides world-leading professional services and customer care, with thousands of incredibly knowledgeable and experienced technical experts to help our customers solve their most difficult challenges. The support we offer customers, is a key reason why we win business from customers who leave competitors.

And finally, Akamai has built a very strong reputation over 25 years as being an innovative and trustworthy company to work with. We're not a newcomer. We have decades of experience in helping the world's largest and most demanding companies navigate the rapidly changing technology landscape.

Speaking of innovation, just last week, we expanded our cloud portfolio with the launch of Akamai Inference and Akamai Agent Sandbox. These new services enable enterprises to build and tune open weight models based on their proprietary data and run agents within strictly enforced boundaries. All secured and powered on Akamai's AI optimized cloud platform.

They add the inference and execution layers that turn our AI infrastructure into a full AI stack to support training, real-time inference, and agentic workloads across the world. Akamai Inference and Agent Sandbox are also integrated with our Firewall for AI and new agent security services to make sure that our customers' AI stay safe.

Turning now to security. I'm pleased to report that security revenue in Q3 grew 9% year-over-year as reported and 10% in constant currency. Growth was led once again by strong demand for our market-leading web app firewall, API Security, and Guardicore Segmentation solutions.

We were especially pleased to see Akamai recognized as a leader in each of these 3 product areas and reports last quarter from industry analysts at IDC and Forrester. AI continues to be a tailwind for our security solutions as more AI agents act on behalf of users and enterprises as more employees put sensitive data into AI tools, as more attacks use AI to grow in scale and sophistication, and as more AI agents and bots get better at appearing to be human.

Akamai security solutions help companies protect their applications, infrastructure, and workforce from AI-related vulnerabilities and attacks. For example, our web app firewall stops malicious traffic before it can reach an application, and it patches exposure points automatically. Our Bot & Agent Control solution unmasks agents to discover their true origin and intent and then takes the appropriate action on behalf of our customers.

Our API Security solution, finds and defends shadow AI and APIs before sensitive data can be stolen. And our Guardicore Segmentation solution contains the damage if and when something bad does get in. In summary, we believe that Akamai has a unique offering in the cloud marketplace with superior performance in security, global scale, and strong financials. Our unique cloud platform enables enterprises to build and scale the next generation of AI applications, from model training in the core to fast inference at the edge, all protected by Akamai's world-leading security solutions and scaled with Akamai's world-leading delivery services.

As a result, we believe that we're ideally positioned to benefit as businesses adopt AI at scale. Already this year, we've closed deals to power and protect mission-critical applications for the world's leading frontier model companies, leading AI equipment companies, major open source model companies, companies building the next generation of AI applications, and the world's leading enterprises.

Regardless of who ultimately become the dominant AI players and how the AI landscape evolves, we believe that the breadth of our AI customer base combined with our very strong enterprise relationships and our unique cloud platform will drive substantial growth for Akamai well into the future.

I've never been more confident in Akamai's ability to power and protect an AI-driven world.

Now I'll turn the call over to Ed, who will go through our financial model and outlook. Ed?

Thanks, Tom. Before I begin reviewing Q3 results and covering Q4 guidance, I want to add to some of Tom's commentary regarding Cloud Infrastructure Service, and more specifically, the $14 billion of multiyear CIS commitments signed and announced this year.

First, I'll start by addressing some questions we have heard from investors regarding the profitability and return on investment related to these customer wins. Today, we posted a slide on the Quarterly Results section of our Investor Relations website that illustrates the expected aggregate returns when the full $2.2 billion of annual revenue from these $14 billion of signed contracts is fully ramped and being recorded in our results.

In addition, the slide shows our calculations for the expected after-tax internal rate of return of approximately 13% and expected non-GAAP average annual adjusted return on invested capital of approximately 15% from these deals. As you can see from the slide, we expect the aggregate total revenue once fully ramped to be approximately $2.2 billion a year and to deliver an aggregate of approximately 75% adjusted EBITDA margin and approximately 30% non-GAAP operating margin.

As stated previously, we expect this revenue to begin ramping in Q4 2026 and to be fully ramped by the end of 2028. Second, as Tom mentioned, we have begun recognizing revenue from 2 of the first 3 major CIS contracts, representing $2.8 billion in total contract value. We are in the final testing phase on the third contract, and we expect it to start generating revenue soon. As Tom mentioned, revenue for 2 of these deals has been delayed due to some supply chain and third-party data center delays. I'm happy to report these issues have been resolved.

As a result of these delays, we expect approximately $30 million of revenue will shift from Q4 into 2027. I want to be very clear, any timing delays do not impact the total revenue over the contract period or result in any penalties, they simply adjust for revenue recognition timing across periods.

Next, regarding the recent $11.6 billion deal announced a few weeks ago, we are pleased to confirm that we have secured and contracted for all the necessary space and power or colo required to deliver the full contract value with these sites expected to be available to us throughout the revenue ramp period previously outlined.

In addition, we've secured approximately 70% of all the necessary hardware, memory, and other components needed to deliver on this contract, including $1.7 billion of memory and other critical items in Q4 2026. As a reminder, this $11.6 billion contract has a pricing adjustment mechanism that is designed to address any possible price changes in the hardware, memory, and other components not ordered, therefore, our margin is protected if prices happen to rise between now and when we place the orders for the remaining hardware and other components.

Finally, looking ahead, we expect CIS revenue of $1.1 billion to $1.3 billion in 2027, which will help drive total company revenue growth in the range of 17% to 22% in constant currency year-over-year and we expect total company revenue growth to further accelerate the following year, surpassing $7 billion in 2028.

It's important to point out the $7 billion expectation in total company revenue in 2028 assumes no additional orders from Anthropic beyond the $11.6 billion we announced in late September, meaning we do not assume any revenue from the $9 billion option related to the Anthropic relationship. In conclusion, we're very confident in our ability to deliver on the revenue we outlined today, while quarterly revenue may fluctuate slightly against the initial schedule for some of these large contracts, any shifts in timing will have no impact on the total revenue or profitability over the life of these agreements.

Now moving on to Q3. Total revenue in the third quarter was $1.122 billion, up 6% year-over-year as reported and up 7% in constant currency. Cloud Infrastructure Services or CIS revenue was $109 million, representing 34% year-over-year growth as reported and 35% in constant currency.

Security revenue maintained strong momentum with revenue of $622 million, up 9% year-over-year as reported and 10% in constant currency. The strength in the third quarter continued to be driven by our fast-growing API Security and Guardicore Segmentation solutions.

Moving to delivery and other cloud applications. Revenue was $391 million, down 3% year-over-year as reported and in constant currency. International revenue was $563 million, up 7% year-over-year or up 9% in constant currency, representing approximately 50% of total revenue in Q3.

U.S. foreign exchange fluctuations had a negative impact on revenue of $1 million on a sequential basis and negative $7 million on a year-over-year basis.

Moving to profitability. In Q3, we generated non-GAAP net income of $262 million or $1.77 of earnings per diluted share, down 5% year-over-year as reported and down 2% in constant currency. Non-GAAP operating margin for Q3 was 25%, in line with our expectations. And Q3 CapEx was $464 million or 41% of revenue. Third quarter CapEx came in slightly below guidance, primarily due to the timing of equipment deliveries shifting into the fourth quarter.

Moving to cash and our capital allocation strategy. In Q3, we did not repurchase any stock. Year-to-date, we have repurchased roughly 5 million shares for a total of approximately $616 million. We ended the third quarter with approximately $565 million remaining on our current repurchase authorization. As of September 30, we had approximately $4.2 billion of cash, cash equivalents and marketable securities. I'd like to touch on our current capital plan and funding outlook.

Based on our current plan, we currently expect to fund our CIS expansion and broader growth investments through 2028 as well as address our 2027 convertible note maturity, primarily from our existing cash balance and expected operating cash flows, which may be supplemented by measured incremental financing activities as appropriate.

We remain highly disciplined in our approach to capital allocation. To the extent we access external capital, we will evaluate flexible, cost-effective structures with a clear focus on preserving shareholder value. As our CIS customer commitments ramp, we expect a meaningful increase in our operating cash flow beginning in 2028, which we expect will enhance our ability to fund our operations and investments from internally generated cash and reduce leverage over time.

Before providing our fourth quarter and full year 2026 guidance, I'd like to touch on a few housekeeping items. First, as a reminder, our Q4 2025 results included an unusually high $18 million contribution from security license revenue. We expect this year we'll return to a normal level of $4 million to $5 million of license revenue in the fourth quarter.

Second, regarding Q4 operating expenses. As is typical for our business, we expect them to be higher than in Q3. The main driver is the seasonal jump in sales commissions as most of our successful reps achieve their annual [ quota ] accelerators. In addition, in Q4 this year, we expect additional margin pressure this quarter, driven by the ramping of CIS expense, mainly colocation costs ahead of revenues.

Third, we remain on track to spend approximately $1.7 billion in Q4 2026 to secure and prepurchase critical supply chain components, notably memory to support our large Anthropic contract we announced a few weeks ago. This will substantially increase our CapEx spend in Q4 and lower our interest income for the quarter. As a reminder, depreciation of this equipment will begin when we start to recognize revenue throughout 2027 and 2028.

And finally, going forward, we do not plan on announcing every large CIS transaction, unless mandated by SEC disclosure rules. We will continue to highlight notable commercial wins as part of our regular quarterly narrative and we will keep investors updated as our capital deployment priorities and revenue dynamics continue to evolve.

Moving now to guidance. For the fourth quarter, we are projecting revenue in the range of $1.145 billion to $1.170 billion, up 5% to 7% as reported and in constant currency over Q4 2025. At current spot rates, foreign exchange fluctuations are expected to have a negative $4 million impact on Q4 revenue compared to Q3 levels and a negative $5 million impact year-over-year.

At these revenue levels, we expect cash gross margin of approximately 70%. Q4 non-GAAP operating expenses are projected to be $368 million to $378 million. We anticipate Q4 adjusted EBITDA margin of approximately 38%. We expect non-GAAP depreciation expense of $165 million to $167 million. We expect non-GAAP operating margin of approximately 24%.

With the overall revenue and spend configuration I just outlined, we expect Q4 non-GAAP EPS in the range of $1.50 to $1.70. This EPS guidance assumes taxes of $48 million to $54 million based on an estimated quarterly non-GAAP tax rate of approximately 18%. It also reflects a fully diluted share count of approximately 149 million shares.

Moving to CapEx. For the reasons I highlighted earlier, we expect to spend approximately $2.445 billion to $2.645 billion in the fourth quarter. Looking ahead to the full year 2026, we expect revenue of $4.44 billion to $4.465 billion, which is up 6% as reported and up 5% to 6% in constant currency.

For Cloud Infrastructure Services, we are now expecting revenue growth of 40% to 42% year-over-year in constant currency. We continue to expect security revenue growth in the high single digits on a constant currency basis in 2026 and for delivery in other cloud apps, we continue to expect a decline in the mid-single digits year-over-year on a constant currency basis. At current spot rates, our guidance assumes foreign exchange will have a positive $5 million impact to revenue in 2026 on a year-over-year basis.

Moving to operating margins. For 2026, we are estimating a non-GAAP operating margin of approximately 25% as measured in today's FX rates.

Turning to CapEx. At this time, we anticipate that our full year capital expenditures will be in the range of $3.462 billion to $3.662 billion.

Moving to EPS. For the full year 2026, we expect non-GAAP earnings per diluted share in the range of $6.47 to $6.68. This non-GAAP earnings guidance is based on a non-GAAP effective tax rate of 18.5%, and a fully diluted share count of 148 million shares.

With that, I'll wrap things up, and Tom and I are happy to take your questions. Operator?

[Operator Instructions] The first question will come from John DiFucci with Guggenheim.

So thanks for all the information, Ed and Tom. This is great. It's really helpful. I think you made it clear last quarter not to expect acceleration of CIS in 3Q. And it was nice of you, to not to rub it in our faces, but somehow the consensus was above -- it did show acceleration. But you pretty much did what you said you were going to do this quarter. But now you lowered '26 revenue, which you talked about due to the supply chain issues. Can you give more detail on the near-term bottlenecks here?

And related, we appreciate the long-term outlook. But how confident are you in that 2027 CIS revenue? And can you help a little on how that ramps through the year, assuming we all listen to what you're telling us?

Yes, good question. These were minor kinds of things with the data centers. You can think of it like a punch list on a construction project, not uncommon, and we normally plan for some delays, they were worse in this case than we planned for, not a big deal, a couple of months of delay. No loss of revenue, it just pushes to next year.

We are very confident in our estimates next year. And you can see from the ranges that we've accounted for pretty substantial delays if they were to take place, and there's upside if that doesn't happen. But very confident next year. And ramping, you want to think about it ramping heavily through the year. And Ed, do you want to talk a little bit more about that?

John, thanks for the questions. Yes, the way to think about it, and I'll break it in halves for you because, again, quarters, you may have a little shift from quarter-to-quarter. But think about 25% to 30% in the first half, 70% to 75% in the back half if you're thinking about sort of the ramping -- take the total amount and you sort of spread it out that way.

But I think it's important, you asked the question about how confident we are? We've made some commentary here. A, we're starting to recognize revenue already. That's a big step. Number two, we have all of the colo that's needed for the large transaction we announced a couple of weeks ago. That's pretty significant because, as you know, colo is a tight market, but we've got it all signed up and ready to go. We're just waiting for it to be delivered to us over the ramp period.

We have all the approvals needed. So we don't get caught up in some of the issues people are worried about with the politics. That's not an issue for us.

And Tom, that was going to be my second question. Like there's obviously, right now, a lot of fear in the market, especially when it comes to the build-out of AI infrastructure. But -- so thanks for that ROIC slide, that's really helpful. But math aside and to be clear, you're pretty much doing Inference only at colo facilities.

I say only, you're not building these huge data centers, out in the world. And so this can be done a lot more quickly than the large data centers used for training. So I guess, I think, anyway, tell me what's right or wrong, it seems that the risk would be lower. There's still risk in execution, but risk you can't control, is lower.

Like you're talking, Tom, the permitting. Can you talk a little bit more about any risk outside your control, including power, real estate, and permitting and how that comes into your business versus others doing mainly training and building out large data centers?

It's a great point and they're very, very different businesses with very, very different risk profiles. We have secured the power and space. It is already permitted. Not only that, we're not talking about 1 or 2 locations so that if something unexpected did happen, you got a big hit with Akamai's approach and architecture, we're in a lot of spaces that are much smaller. And so yes, some location has some small delay, not going to impact us in the big picture. Also makes it a lot easier for us to get the space because it's small chunks. It's not caught up in some permit for a nuclear power plant next door, and you're right, we are not in the business of training the big frontier models.

We don't buy that kind of equipment. It's a totally different approach. As I said, we're doing mostly CPUs, we do training for enterprise-sized models and post training and inference. And as much -- most of the time, you're not even using GPUs for that. And this is where I think a lot of the future market is going to be, which is why we're focusing on it.

So it's much more profitable for us, much lower risk profile, and I really don't see a lot of execution risk in what we're projecting for next year. This is stuff that we know how to do. And yes, we had a couple of months, okay, of a push, not really a big deal, and I don't think that's any impact on next year.

The next question will come from Samik Chatterjee with JPMorgan.

I know you said you won't announce sort of the incremental maybe deals or wins going forward for the CIS business. But maybe if you can comment about -- comment on the pipeline of demand that you're seeing right now?

And when you sort of -- tying that together with your comments about capital raise, should we expect capital raise only if that's associated with more incremental wins? Or maybe if you can clarify sort of how to think about pipeline conversion vis-a-vis capital raise, that will be helpful, and I have a follow-up.

Yes, we have a very strong pipeline with hundreds of customers including some pretty substantial accounts and we're very optimistic about being able to close that business. And we have contracted for the power and data center space to accommodate that. And I think we are in a very good position, as Ed said, with our capital, our ability to raise a modest amount of capital if we find -- if we need that, and to do that in a way, as we've always done, that is very cognizant of shareholder interest.

Got it. And I think in the prepared remarks, you did mention you're seeing the recent deals at a much stronger premium than the $22 million per megawatt that you had signed some of the earlier recent deals on. Can you just help think about sort of the magnitude of the premium you're seeing?

And how does that sort of change the returns that you're putting sort of showing us in the slides today what is the sort of magnitude or difference that you're seeing in the market? How does that change the difference between the IRR relative to the cost of capital that investors should expect?

Yes, I'll start and then Ed will get into the details on this. But yes, apples-to-apples, we are seeing improved pricing. And I think part of that is because we're getting a lot better known in the marketplace which is very helpful. And Ed, do you want to talk about the details on the IRR and the economics?

Yes, sure. I mean we gave you those numbers just because I know there's been a lot of debate in terms of how you would calculate it. So we provided the more information so you can calculate it. And those are good returns for us, obviously. Now I want you guys to keep in mind, this is -- these are for the largest volume and the longest term, and the standard in the industry today is, if you were to go buy something uncommitted, you're going to pay the highest rate.

So no commitment, low volume, you have the highest rates. So these are the lowest prices you're going to see on a unit economic basis. So look at that as sort of your floor as far as what pricing is. So somebody is going to commit to half the amount or 1/3 of the amount and 1/10 of the volume, they're not going to get anywhere near those prices. So that's pretty good to see, like for the largest possible deal to lock in that kind of return for 7 years is a very, very good business for us and for our shareholders. And the deals that are coming in, are coming in higher.

Now keep in mind, this business, a lot of what drives the unit pricing on the other side of like the on sort of buy side is based on the component costs. So as component costs are going up, the market accepts higher prices. So the margins are insulated in the fact that if we see co-location costs go up or if we see power cost -- sorry, compute -- component costs go up, we can raise our pricing for the next deals.

We lock in, as we talked about, these long-term deals, we've got mechanisms in place to deal with any pricing. But usually, what happens is we've got the colo locked up for a long period of time, and we've got the equipment purchase. So we're doing a really good job of securing margins. And that's what you see there on those big deals is the lowest you're going to typically see.

The next question will come from Jackson Ader with KeyBanc.

Ed, if I can follow up just on the margin. I appreciate the first half, second half look for revenue build next year. But what is -- is the margin going to lag like it is here in the fourth quarter before that revenue really takes a step up? Any direction would be helpful.

Yes, sure. So just to give you a little bit more picture generally, when you do these big build-outs, you're going to have -- your colocation cost is going to come in ahead of revenue. And the other issue is when you're doing a long-term colo commitments, they do come with price escalators. So you do have to put the straight-line effect of that. So what ends up happening is you end up taking a noncash charge because you're actually booking a cost higher than what you're paying for with colocation.

We estimate that will put about 1 point or 2 of pressure on margins next year. That said, I think we'll bottom out probably in the first half in terms of margins, maybe a point or so lower from where we are now, and we should be operating in the mid-20s. Keeping in mind that you've got this 1 to 2-point noncash pressure on the margin, that sort of works itself out over time.

The other thing is just with some of these build-outs there are other costs, too, what we call network build and support, we have some onetime costs for things like freight, small equipment, consulting, contractors that go and build and stuff like that. So that can also put some near-term pressure. But as you can see with the slide we provided you, what you're going to see happen here is margins sort of bottom out in the first half of next year start to improve.

And where you're going to really see a ton of leverage on the EBITDA margin. That should expand pretty significantly through '28.

Got it. Okay. Great. That is really helpful. And then Dr. Tom, the CapEx per megawatt being higher than the neoclouds, that's a good thing. I just need a quick explanation of that.

That is a great question. And CapEx per megawatt is higher at Akamai because we use less megawatts for our CapEx. And so if your denominator gets smaller because you're using less power, the ratio of CapEx to megawatt gets bigger. So the more -- the less power we use and the more efficient we are, the higher our CapEx per megawatt is, in fact, say we could power our servers with no power.

We would have infinite CapEx per megawatt. And so the trouble people get into when they model based on per megawatt is that your CapEx per megawatt, you think, oh, that should be small. In fact, you want it to be large because that means you're using less power for your infrastructure. So it is a big advantage for us. Now revenue per megawatt matters, just as revenue per CapEx matters, and we have very good revenue per megawatt. Did that help...

You tend to see higher revenue per megawatt and CPU deals than you do in GPU deals. So you could have 2 identical deals: 1 that's, say, $15 million a megawatt with say 1 megawatt is going to cost you about $4 million in colocation. So you have $4 million of depreciation, $1 million of other costs, that gets you to a 40% margin. Well, you could be getting $25 million per megawatt on a CPU deal with, again, $4 million of colo costs, $10 million depreciation $1 million of other costs, and that's 40% margin. So you got to keep in mind that you do get higher revenue per megawatt out of your CPUs than GPUs.

CPUs more power efficient. So you're in a better position.

The next question will come from Jeff Van Rhee with Craig-Hallum.

This is Vijay Homan on for Jeff. First one for me, it just looks like the decline in delivery and other cloud apps kind of moderated a little bit this quarter. Just any update on delivery pricing or bit growth and then expectations for those 2 factors, how they will evolve over the next 6 to 12 months?

Yes. Good question. Really nothing on pricing. I'd say we're probably holding pricing better now than we have for a long period of time. And actually, if you look at the last 4 quarters, you see kind of the Delivery and cloud apps is sort of stable, right? It's within plus or minus $1 million or $2 million every quarter. And that's usually seasonality. There's some seasonality in there, maybe there's a big renewal in there.

But generally speaking, it's kind of flattish. And I'd say, going forward, I think we're sort of in that stable period, down a point or 2 or 3 or whatever. But it's stabilizing, pricing is getting a little bit better and we are being very selective as we move forward here with our -- especially in our big media business.

Got it. Got it. That's helpful. And then just 1 follow-up on the LayerX acquisition, just now it's been closed for a couple of months. If you could just talk a little bit about how that integration is going? And then any commentary on the revenue potential maybe over the next, I don't know, couple of years would be great.

Integration is going great and already a lot of sales, including with some big name enterprises, including a leading satellite and streaming TV company in the U.S. is using it. And I think there's a very strong potential for growth there. And what LayerX does is, it make sure that your employee workforce is not uploading company secrets and data into prompts when they're using AI, among other things.

So it makes sure your workforce stays safe as they're using AI. And the feedback in the analyst community very strong and the customer feedback there, recent report, we came out very well with a score of 4.8 out of 5 in terms of customer satisfaction. So early days, of course, but good integration and strong results so far.

The next question will come from Sanjit Singh with Morgan Stanley.

This is Abhishek Murli on for Sanjit Singh. Maybe one for Ed. Previously, you've spoken to a willingness to drop below investment grade. And I understand some of that is kind of a mechanical thing as these deals ramp and the EBITDA disclosure is helpful for that. Curious how you're thinking about that dynamic going forward just given the recent uptick in scrutiny around financing? And anything you can share around cost of debt or cost of equity?

Yes, sure, you broke up there a little bit. I didn't quite hear exactly what you said. You said we made a comment about investment grade. Can you just repeat that part?

Yes, sure. Previously, you've mentioned a willingness to potentially not -- or to drop below investment grade. So I'm just curious how you're thinking about that around the broader scrutiny around financing and whether you guys are still open to that possibility.

Yes. I mean just to be clear, I mean, our goal would be to maintain investment grade as long as we certainly can. If there is a time that you had to raise additional capital to go below investment grade, which we don't think we're in that position.

But maintaining investment grade is helpful. It's actually for our colo investments that we make with our colo providers having that is helpful for us to be able to get space and power. We're more attractive as a customer to those customers. So obviously, we've been investment-grade for a very long time. We're going to try to keep it that way.

If there was a reason and you were to go and raise capital, and you happen to get a notch down or something like that, we would work as quickly as possible to get back to investment grade. And you bring a good point. I mean the expansion of EBITDA is interesting, right, because that's the primary metric. And if you look at what our leverage ratio is now just the EBITDA that we're going to be adding in the next 2 years is going to substantially increase our ability to buy and remain in investment-grade territory. So we got an enormous amount of flexibility. So we're -- I think we're in great shape.

The next question will come from James Fish with Piper Sandler.

I did want to go back to Mr. DiFucci's earlier question around confidence here. Just because -- look, I get this is a new business for you guys, but -- and you guys did talk about a deceleration on the last print for compute. But let's face it. We've kind of missed compute or only been in line on virtually every quarter that you've had this business.

And now you're putting out a target that's well above what we were all expecting. So what's giving the confidence or why do you feel the need at this stage to put this out when, frankly, we haven't seen you guys hit your corporate -- or your compute numbers really yet. Maybe, Ed, walk us through why -- what has to happen to hit the $1.1 billion versus the $1.3 billion next year? And why -- and just help us help you.

Yes, sure. I mean I'm not going to go litigate the past. So let's just focus on going forward. If I look at the range that we've given you, obviously, you have to look at a few things. One, do we have the contracts signed to be able to hit those numbers? The answer there is, yes.

Number two, do you have all the power that you need? The answer is, yes. Do we have all the equipment that we need? We have the majority of it and the remainder of it is really more just a timing issue that we don't want to be sitting on that for too long and carrying all that cost for everything. So we can get all that and we have a pricing mechanism to adjust for it. All that remains at this point.

We'll have 3 of these deals ramped up and generating revenue and there's a schedule of when equipment comes in and we put it into the data center and once it's made available after it's been tested and it's up and running. The customer can start using it and we start billing. So there's very little risk in terms of the big picture of what's left to do. Is there a timing potential of having something being off our schedule, and that's essentially what happened where we assumed we would be in the data center at a particular date and that we have all of our equipment at a particular date and things slipped a couple of weeks. That happens, right?

So what we've done now is we've given you a wide range to show that, look, if we were to experience some delays from what we have as a fairly conservative view of the world, yes, we'd be down closer to the lower end. If we do better in terms of getting things on time or even slightly better, we could do that higher end of the range.

And also we're not assuming a lot of incremental productivity in there in terms of giving out that range because that adds a lot more risk, right? If I were to put big deals in there that I haven't signed, well, that's something that I can't control the day I'm going to sign that, that's more risky. So this is based on signed deals where we have the space already locked up and lined up, and we've got the supply chain primed and ready to go.

And so then the question we always get is, obviously, we have these large contracts in place, now over $14 billion, so congrats on that number. in terms of what you guys got to go execute on, what's the willingness given the environment to do another large deal before executing on these at this stage?

Yes, I'm glad you mentioned it, Jim, because in my prepared remarks, I did say for that $7 billion number that we're talking about when we said we believe we'll surpass that in 2028. I made the comment that we don't anticipate us taking on any of those additional tranches from the Anthropic relationship.

Not saying that we don't want to or anything like that. We just think that, a, we've got to execute on what we have in front of us, and b, there's an incentive for them as our stock goes higher with the warrants that are attached, so we just don't assume they'll be really pushing for those orders yet. But I think once we execute through that, our willingness to go and grow faster is there. We do plan to have capacity, both capital as well as data center space to do larger deals in terms of something in the several hundred millions and that sort of thing

But in terms of the multibillion dollar deals, we don't have that in our forecast or our plans. Let's get through the execution here on next year and then we'll talk more about it then.

The next question will come from Param Singh with Oppenheimer.

Firstly, I wanted to get a sense of the growth in security. It seemed a little bit light this quarter. How do you think that should translate over the next year or 2? As you talk about -- more about this cross-sell with the API Security and other modules that you mentioned earlier in the call? And then I had a follow-up question.

Yes. So I'm not sure, were you just talking about the quarter we just delivered, I think we came in slightly ahead of where you guys were or are you talking more about the guide?

So I guess I was a little bit ahead of the Street then for the security portfolio for this quarter, but I really want to think about the next 2 years and how that embeds into your $7 billion guide. Because you know the $2.2 billion that's coming from CIS. So some color there would be helpful.

Yes, sure. So we're not going to give guidance on every different component, certainly for '27, we'll do that in the February call. And then certainly for '28, we'll give you the guidance on that. But I think you can assume sort of normal business as usual. We've been pretty consistent if you look at the -- if I look at the sequential growth, and I adjust for some of those anomalous license from last year.

We've been adding 12 to 13-ish, maybe a little bit more every quarter of sequential additional security growth. So it's continuing to grow at a very healthy clip. We have a long way to go with our API Security and obviously, LayerX is brand new. Guardicore is still growing at a very healthy clip and has a relatively low percentage -- penetration in the base compared to some of our other products. We're adding new customers at a faster clip. We've made an investment in adding customers. So we feel pretty good that security is going to continue to grow, and we'll give you more guidance on that in February in terms of exactly how fast next year.

Understood. And as my follow-up, I hate to beat a dead horse but I understand your argument of why CapEx per megawatt is higher. And at the same time, you're talking about higher revenue but let's just take the middle part out, right? If we do revenue per CapEx, right, because you're adjusting for the megawattage efficiency. In that scenario, I mean it's still pretty light, right?

On our revenue per megawatt, it seems lower than a lot of the other Tier 2 hyperscalers and definitely in line with some of the older contracts of the neocloud. So I'm having a very difficult time understanding the dynamics behind this deal. And if you just kind of delineate from a revenue per CapEx perspective, that would help me understand this better.

Yes. So you got to make sure you're comparing like-for-like. If you're comparing like say, the rental business for GPUs and what people are getting for short-term contracts and the dynamics of CapEx relative to a megawatt of power for GPU in that environment versus a 7-year commit for CPU, you can't compare the 2. They're very, very different. I believe what I heard you say is that our revenue per megawatt is similar to what you see with hyperscalers, I think that's what you said. They're a mix there, mostly primarily of CPU. I don't know off the top of my head what their CapEx is from that perspective.

But again, we're showing you what the economics look like for the absolute best pricing you can get for significant volume for significant term. If I go to the big 3 hyperscalers today, and I want to sign a 1-year deal, no commit versus a 3- or 5-year deal with a big commit, I'm going to get dramatically better pricing for the long term and higher commit. So it's hard for me to answer comparing to these other guys because I'm not sure what you guys are looking at in terms of what the dynamics are?

Are you talking about what short-term contract, long-term contract, et cetera. I just don't know their numbers off the top of my head. But we're very pleased with the returns we're getting on this business, and you can see we're in for a period of significant revenue growth and significant profit growth.

The next question will come from Rishi Jaluria with RBC.

Wonderful. Look, maybe I want to ask kind of a more technological infrastructure question. So as you're continuing to build out this fleet, right, not just for the CIS deals that you're signing, but I think for future contemplated demand, Tom, in your prepared remarks, you did hit on this a little bit. But there's clearly a lot of other potential use cases that you can be going after. Physical AI presumably Sovereign AI, especially with the kind of growth and capabilities around open weight models.

Can you maybe walk us through what are you doing to ensure that the fleet you're building is fully fungible, that you can pivot in real time to kind of how adoption trends in AI tend to shape, especially just given leadership and models tend to change? And maybe your thoughts on why you feel Akamai is uniquely positioned to benefit from both of those trends.

Yes. In fact, we've signed some very large physical AI deals. Many Physical AI contracts so far. Sovereign AI also makes a lot of sense for us with our distributed platform. We do have diversity in the hardware in our platform, in the servers. But as we talked about, we emphasize the CPUs, they're very flexible. And increasingly, for the kinds of enterprise applications that our customer base is doing CPUs work very well for that.

Again, we are not going after training the next frontier model. We are interested in post-training of, say, open source models on our platform. Now with our inference, Akamai Inference capability and having agents on our platform with Agent Sandbox. And we will assign that to the hardware that is most efficient and capable of doing the particular workload. And we take that responsibility off our customers' hands.

So I think we're really ideally positioned with our distributed platform that we can get these workloads on the right kind of server that makes it be efficient, we can get it close to the robot, if that's what it's being used for the sensor or the alert sensor, if it's what it's being used for, or the user, if you're trying to maintain some kind of dialogue or chat with the user or some video with the user and their proximity matters for latency and scale.

And we're really unique at that. And that's why even the hyperscalers use our compute platform today for the latency-sensitive applications. So I think -- yes, and you compare it to the neocloud, it's an entirely different capability.

And then, of course, with Akamai packaged also with the security you need, especially important for AI and the delivery for a lot of these applications, if you're doing some kind of video with a user or a video with a robot or a car, you need that kind of scalability that you get from our Edge Platform.

The next question will come from Peter Levine with Evercore ISI.

Great. Maybe just 2 quick follow-ups. The 13% IRR, I'm assuming that represents a blended, obviously, from your slide deck. But given the scale of like Anthropic, I mean, is it fair to assume Anthropic received favorable pricing terms versus your other customers, so therefore, generate some lower returns? Just kind of help us walk us through like how we should think about the dispersion between some -- your returns versus some of these larger customers.

Yes. No, I think you just summed it up well that they're going to get a better return from a pricing perspective because they're, a, willing to commit for 7 years, and b, they're committing a lot more volume. So that would be the lowest type of return you would get. So other customers that are not committing to that term or are not committing to that volume are going to have higher IRRs.

And maybe just a follow-up to John DiFucci's question, I guess, more on the political side. Do you view this as a tailwind I guess to your CIS business, assuming that you can accommodate incremental AI workloads within your existing kind of colocation footprint? Like just kind of walk us through what you saw post midterms? Is that a tailwind? Is that a headwind? Or is it just kind of -- it doesn't really change your outlook on this business?

In terms of the business, we're not negatively impacted by it. And so in a way, we're relatively advantaged compared to the competition, who do have serious challenges, I think. Now of course, the market got pretty well spooked as you know, and we got dragged down with the market. I think unfairly so because the issues that spooked the market really aren't relevant to Akamai.

As I've talked about, we're in a unique and great position to be able to go get the kind of colo and power that we need, which is in many locations with much smaller footprints. And as Ed talked about, we already are permitted for everything we need through 2028 for all the deals we've talked about and more. So it's not an issue.

Now to the extent that it causes prices to go up, costs to go up, well, again, we're contracted, so that won't hit us. It will potentially enable us to raise prices and that may be a relative advantage because the other guys are going to have potentially challenges with these issues. And we are really in a great position with our distributed platform to not have those kind of problems.

The next question will come from Roger Boyd with UBS.

Dr. Tom, you made an important point about the CIS business being 90% compute-driven, and you're not selling bare metal, but a lot of that compute is fungible across the rest of the platform. When you think about capacity investment needed to scale the security and delivery businesses over the next couple of years. Do you anticipate a point where you might be less willing to take on new delivery and security business, just given the opportunity and the margin uplift you see on the compute side?

Yes. Well, obviously, security, very good business for us. About $2.5 billion, just grew 10% in constant currency. We are continuing to invest and grow that business. Delivery, as we've talked about, we're not investing in the same way there. Obviously, we want to keep the platform. So it does all the new protocols is very reliable and has great performance.

But we're not going to the very low-end deals in terms of pricing or the deals where there's huge spikes where it just doesn't make economic sense. We're much better off investing that CapEx into our CIS business, much better returns. But I wouldn't equate security with the delivery business, security, very strong business and doesn't take a huge amount of CapEx.

Great color. And then just to tie a bow on the $7 billion target. If you assume kind of security and delivery follow a consistent trend over the next few years. It looks like that target is achievable just by executing on that $2.2 billion runway backlog you have in compute today. Is that the right way to think about it?

And is the implication around the other segments directionally correct? And any color you can give on the assumptions for delivery and security in that guide would be helpful.

Yes. No, I think you're right to think about it that way, that is very achievable based on what we have signed today and there's upside to that. Now keep in mind, '28 it's going to be a ramp year. So it's not -- not all of the $2.2 billion is for the full year, obviously. But yes, you're thinking about it the right way that it's a very achievable number based on what we've signed. We're not expecting massive big deals. That's why I made that comment about not assuming there's any additional Anthropic revenue in there, which obviously would change the trajectory quite a bit. So that's -- you're interpreting it the correct way.

The final question today will come from Rudy Kessinger with D.A. Davidson.

Great. Appreciate all the additional details here. I guess, just going forward, as you look at the large deals in your pipeline, understanding Anthropic got more favorable pricing and is probably weighing down that 13% after-tax IRR a bit. But if you look at your pipeline and the large deals you're in negotiations with, is that IRR on those deals on average at a similar level? Is it a couple of points higher? Is it meaningfully higher? I'm just trying to get a sense of what future deals might look like from that standpoint?

Yes, it depends. I mean the nice thing is we're able to pick and choose, too. So we are walking away from deals that we don't believe meet our profit threshold or return on capital. But I would say, in a lot of cases, they are significantly higher than what we're seeing here today. But again, no one's at the scale of committing $11 billion. So that's very anomalous. Someone in a couple of hundred million would have a much higher IRR than that.

This concludes our question-and-answer session as well as today's conference call. Thank you for attending today's presentation. You may now disconnect.

Good day, and welcome to the Akamai Customer Announcement Overview Conference Call. [Operator Instructions] Please note this event is being recorded. I would now like to turn the conference over to Mark Stoutenberg, Head of Investor Relations. Please go ahead.

Good afternoon, everyone, and thank you for joining us today on short notice. We have some very exciting news to share from a customer win perspective, and we will get to that in just a moment. Speaking today will be Tom Leighton, Akamai's Chief Executive Officer; and Ed McGowan, Akamai's Chief Financial Officer.

Please note that today's comments include forward-looking statements that include revenue guidance. These forward-looking statements are based on current expectations and assumptions that are subject to certain risks and uncertainties and involve a number of factors that could cause actual results to differ materially from those expressed or implied. The factors include, but are not limited to, any impact from macroeconomic trends, the integration from any acquisition, geopolitical developments, and other risk factors identified with our filings with the SEC.

The statements included on today's call represent the company's views on September 24, 2026, and we assume no obligation to update any forward-looking statements. As a reminder, we will be referring to certain non-GAAP financial metrics during today's call. A detailed GAAP to non-GAAP reconciliation is available in the Investor Relations section of akamai.com under Financials.

Before I hand the call off to Tom, I have 2 quick housekeeping items to cover. First, we have published a presentation of slides outlining the financials and timing of this new contract. You can access this presentation in the IR section of our website in both the Presentations and Events or Quarterly Earnings sections. Second, during today's abbreviated Q&A, please limit yourself to one question and one follow-up focused strictly on today's announcements. We will not be addressing questions related to our third quarter results. With that, I'll now hand the call off to our CEO, Dr. Tom Leighton.

Thanks, Mark, and thank you all for joining us. As we announced in our press release earlier today, I'm very excited to tell you that Akamai has signed the largest contract in our company's history, a commitment of $11.6 billion over 7 years with Anthropic to support their accelerating CPU workload demands with Akamai Cloud's distributed AI infrastructure and software. Anthropic is advancing the AI revolution, and we are thrilled they chose Akamai's capabilities for building and operating AI infrastructure at scale. When combined with the $2.8 billion in multi-year commitments for our cloud infrastructure services that we signed earlier this year, this transaction will significantly accelerate Akamai's CIS business and overall revenue growth.

For nearly 30 years, Akamai has amassed industry-leading expertise in building and operating the world's most distributed platform for content delivery and cybersecurity at global scale and with a strong reputation for reliability, quality, and trust. We were also pioneers in edge computing and took a major step forward with our cloud strategy when we acquired Linode's developer-friendly compute platform in 2022 and then integrated it with our world-leading capabilities for delivery and security, and our unique distributed platform with more than 4,000 points of presence across 700 cities in 130 countries.

Over the last 4 years, we've made major investments to upgrade Linode's core compute and storage infrastructure to create Akamai Cloud, a leading cloud platform that provides the performance, security, and reliability required by the world's largest enterprises and their most critical applications. We've scaled and deployed the platform into dozens of cities around the world, all interconnected with our global network fabric, which is one of the largest in the industry. As a result of this hard work, we've made Akamai the cloud company that powers and protects an AI-driven world. Our cloud platform extends high-performance cloud computing from the core to the edge, enabling organizations to build and scale next-generation AI applications while also providing comprehensive multi-layered security to safeguard enterprises against evolving cyber threats.

Customers are using our cloud infrastructure services today for a wide variety of applications with many powered by AI. Examples include conversational voice agents for customer service and personal assistance, translation for local language and cultural etiquette, speech recognition for subtitle generation, photo-realistic image generation for personalized commerce, live transcoding and ad break detection, robotics and physical AI, site reliability and root cause investigation, real-time virtual world generation and rendering for simulations, real-time AI video intelligence to transform raw CCTV feeds into actionable insights, strategic findings from complex gameplay data in seconds instead of hours, and B2B agentic DevOps assistance designed for large-scale enterprise workflows.

Enterprises choose Akamai for their cloud infrastructure needs because of our low latency and global scalability, our proven ability to manage and scale distributed systems, our decades-long track record of reliability, our global network fabric, our market-leading security services, threat intelligence and expertise, our ability to get the hardware and space around the world to support our contracts, our talented and experienced technical team that puts the customer first, and our reputation as a stable, trusted and dependable partner.

Akamai Cloud spans the full spectrum from core to edge and training to inference with dozens of core regions in major cities around the world, hundreds of cities with edge container support and thousands of locations for Function-as-a-Service. We plan to continue expanding our cloud platform with greater capacity and scale with diversified hardware, and with new capabilities for agent sandboxing and security, AI gateway and firewall, Model-as-a-Service and intelligent orchestration to ensure that each workload runs on the most cost-effective hardware with minimal start-up time and low latency to users and data.

With the AI business we've signed this year, backed up by a very strong pipeline, we're building upon our global footprint and years of experience in serving the world's largest enterprises to enable and secure responsible AI as we position Akamai to be the infrastructure provider for the next generation of AI-powered applications. Now I'll turn the call over to Ed to say more from the financial perspective. Ed?

Thanks, Tom. As Tom just outlined, today, we announced a significant expansion of our strategic relationship with Anthropic, signing an $11.6 billion 7-year commitment for CIS services to support their accelerating CPU workload demands. Additionally, our agreement provides for the potential to expand to an additional $9 billion of revenue commitments for a total of up to $20 billion over 7 years. As part of this expanded relationship, Akamai has issued a warrant to Anthropic to purchase up to 7.7 million common share equivalents of Series B non-voting convertible preferred stock of Akamai, which is equal to approximately 5% of total shares outstanding.

The common stock equivalent exercise price of the warrants is $111.33. The warrants have a term of 7 years and that's based on the size of committed revenue from Anthropic. Approximately 3.1 million common share equivalents or 2% of shares outstanding are expected to vest as part of today's announcement and related to the $11.6 billion commitment. The remaining 3% would vest as Anthropic commits additional revenue in commercial agreements with approximately 1% of total shares outstanding vesting for each additional $3 billion of committed revenue, up to a total of $9 billion of additional revenue commitments.

While I can't talk to all the specific details regarding the $11.6 billion commitment, what I plan to discuss today is the following: the expected timing and amount of revenue we expect to generate from this contract over the next several years as the business ramps, the expected capital expenditures and expected timing of the CapEx, some details on colocation required to power deals of this size and finally, accounting treatment for the warrants and other items for you to consider as you update your models. These points I'm about to cover are also detailed in the supplemental slides we published today to the IR section of our website.

Starting with revenue. Please note that the fair value of the warrant will be deducted from the revenue ratably over the life of the contract and is included in the $11.6 billion total. So first, we do not expect to generate any revenue in 2026 related to this contract. Second, we expect revenue to begin in the second half of 2027, and we expect to generate approximately $150 million to $300 million for the full year 2027. Finally, we expect revenue to continue to ramp throughout 2028 and to be fully ramped by year-end 2028, with a go-forward annualized revenue run rate of approximately $1.7 billion per year thereafter.

Because of this take-or-pay structure in which Akamai's guaranteed payment upon delivery, revenue will hold steady for the remainder of the contract once revenue is fully ramped. From a CapEx perspective, we expect to spend approximately $5.5 billion over the next 2 years to support the $11.6 billion commitment announced today. First, we plan to spend approximately $1.7 billion of the $5.5 billion to secure and pre-purchase critical supply chain components, including memory, in the fourth quarter of 2026. Second, for the full year 2027, we expect CapEx to be approximately $3.1 billion to support this contract. Third, for the full year 2028, we expect to finalize capital deployments with an additional approximately $700 million of CapEx in 2028.

Also, as a reminder, depreciation will begin when the equipment is placed in service. Overall, we expect depreciation to be substantially in line with revenue. Finally, there is no contractual requirement to refresh or upgrade the equipment over the length of the contract.

Moving to power requirements. While I also can't share specific power details for this particular deal, what I can say is that over the past year, we've announced several major CIS contracts totaling approximately $2.8 billion of multi-year committed revenue with various leading enterprise customers. With today's $11.6 billion announcement, our year-to-date signed total contract value reaches approximately $14.4 billion. That combined total once fully ramped, is expected to generate approximately $2.2 billion in annual recurring revenue. We believe that the required power for all of these deals combined will be approximately 95 to 105 megawatts, which averages out to roughly $22 million in annual revenue per megawatt of power for the entire portfolio.

Please note, as we bring new sites online, we typically anticipate a 60- to 90-day ramp period before they reach full revenue potential. Therefore, margins will likely experience temporary pressure over the next several quarters while we ramp revenue. And before we take your questions, I want to cover a few more things for you to consider as you model the impact of this transaction. For this contract, we are required to net fair value of any warrants issued against revenue. In addition, under ASC 606, any price escalators over the term are required to be straight-lined over the life of the agreement. Please note that the revenue projections I have provided today take into account both the cost of the warrants and any price escalators in this agreement.

As a side note, under ASC 842, we're required to treat colocation costs the same as revenue and any cost escalators would be factored into any guidance we provide in the future. Additionally, because we are using substantial cash to fund CapEx, please adjust your financial models for the lower interest net income. We recommend referencing my earlier comments along with the CapEx schedule on Slide 4 for the amount and timing, and please use a 4% to 4.5% yield on cash balance as you estimate the lower interest income from the cash used to fund the CapEx. Your models also should reflect an adjustment to our share count to account for the issuance of the warrant.

Once we expect a tranche of warrants to vest, we will include those shares in our calculation using the treasury stock method. And finally, the treasury stock method assumes that the warrant that is expected to vest is exercised and that the proceeds from the exercise are used to buy back shares at the current stock price to offset dilution. More dilution will occur if the stock price increases and if we sign additional revenue commitments with Anthropic. So with that, I'll wrap things up, and Tom and I are happy to take your questions. Operator?

[Operator Instructions] The first question will come from John DiFucci with Guggenheim Securities.

This is Lawrence Vensko on for John DiFucci. First, I just wanted to say our congratulations to the entire Akamai team on this huge CIS win. Ed, maybe a question for you. Would you be able to provide us with a little more color on the profit metrics of this deal? You disclosed that CapEx is a little less than half the value of the deal, but we assume the useful life of the CapEx might go beyond the duration of this deal as it's defined today. You mentioned it a bit earlier, but if you could, what might be the approximate gross margins over the entire life of the deal?

Yes. Lawrence, thanks for the question. So I can't get into the specifics of this deal or any one deal in particular, but I did provide a lot of information, including the total power. So the way to think about cash gross margin, the largest cost is your colocation. So you have the total power number across that entire portfolio, so you can sort of back into that. The majority of all of this across all the portfolios is in the U.S. So that's where you can assume the cost of colo, $3 million to $4 million a megawatt, something in that range, and it's a pretty decent place to put the model.

Other costs, there'll be some warranty costs, a few other things, but colo is generally the major cost there. And then from a depreciation perspective, yes, since we don't have a requirement to refresh the hardware, you can assume 7 years for depreciation for this particular deal. We have servers in our network where we get greater than 7 years. So obviously, at the end of this contract, if we can continue to use the machines, we certainly will. But I think that's probably the right way to think about your model. That's pretty much all I can say in terms of general direction for overall margins, not only for this, but for all deals in particular.

Congrats again.

The next question will come from Samik Chatterjee with JPMorgan.

Congrats from my side as well. For the first one, can you just elaborate on how to think about the $9 billion expansion opportunity? Are there certain milestones you need to meet in terms of execution on the first part of the contract to get that expansion opportunity? And is that also all CPU-based? And I have a quick follow-up.

Yes. I would think of that as an option to do additional business where there's an incentive in place, where there's additional warrants that would vest if they were to do additional business. So yes, there's no dependency for us to do anything or milestones or anything like that. That's just purely looking at additional opportunities over the life of the relationship. And to the extent that we find opportunities that Anthropic is interested in us bidding on and we find an opportunity, we'll certainly tell you about it, but there's no connection in terms of the deliverables of the first $11.6 billion to that. It's just really more of an option for us over the future to do more business with them over time.

Got it. Got it. And my follow-up, I mean, you've been talking about ARR to CapEx, the best way to sort of think about it is more like 0.5 to 1. For this deal itself, it seems more closer to sort of 1/3 on that front. Are you able to elaborate sort of what are the drivers that drive that sort of ARR to CapEx metric? It seems to be a bit of deviation from what you've been sort of suggesting as a more longer-term trend for most of the deals that you signed.

Yes. So I would say that $0.5 to $1 is pretty much holds true for the majority of the number of deals, the volume of customers, et cetera. For these larger deals, though, there's a lot of things you have to take into consideration. Number one, this particular deal, CPU, so you get much better flow-through in terms of the efficiency on power. Also, it's a 7-year deal. So when you look at the overall economics of a deal like this, you take all that into consideration. So if someone were to go buy, say, a smaller contract from us, the yield will be much higher. The pricing would be obviously different because obviously, the size and scale of this is pretty substantial. So we factor that in. And occasionally, if we can do a deal like this at this level, we'll do it, provided it fits within our margin guidance, it certainly does.

The next question will come from Sanjit Singh with Morgan Stanley.

Congrats on this nice, strategic deal. In terms of the financing of the $5.5 billion CapEx, is that sort of implied with the warrant structure here? Or is there additional financing you need that the company is going to have to execute to, to deliver the $5.5 billion over the next couple of years?

Yes, good question. So keep in mind, if you look at the schedule in my prepared remarks, the CapEx is spread out over 2 years. So we don't have a bullet, if you will, in terms of having to spend it all upfront at once. So it will be spread out over time. And we're -- the business is very profitable, obviously. We have $4.6 billion in cash as of the last report that we reported last quarter, and we also have a revolver for $1 billion. But to the extent that if there's additional capital needed, we've always been good stewards of our shareholders. And we'll -- if we need additional capital, we'll certainly have a discussion with our Board and with our senior management team and do what we think is best for shareholders overall.

The next question will come from Param Singh with Oppenheimer.

So my first one, I saw in your press release, you mentioned that the new deal was for CPU workloads. So I want to make sure the $14.4 billion of commitments you're talking about, are they all CPU architectures? Or is there some sort of GPU architecture also underlying it? And then I had a follow-up.

Yes. So this particular deal, the one we announced today, is all CPU. The 14.4, which is a mix of customers and what we've talked about is our considered, what we consider our large multiyear revenue contracts, that's a mix. That's got some CPU, that's got some GPU in there. So it's a mix of everything.

Understood. For my follow-up, 100 megawatts is a lot, and I'm glad you're expanding at a very good pace. Maybe you could give us some color on any commitments you've gotten from your colo providers? Are they new colo providers? Are you looking at having your own lease? And is this like behind-the-meter or grid power? Any kind of clarity on that 100 megawatts would be really helpful.

Yes, sure. So first of all, we're very comfortable with the portfolio and pipeline we have for colo. We're very unique, I would say, in terms of how we go about building out 100 megawatts of power, for example. We have many, many providers we work with, and we're able to get agreements for multi-years for, say, 10 to 30 megawatts in certain locations. In some cases, it might be a little bit less. But we're very comfortable with what we can put together. And there's no dependency for us to go build. We're going -- working with our colocation providers. So this is spread out among numerous providers. And like I said, we're very confident in our ability to secure that. Some of that's already been secured. So this is kind of a -- think of it as what we do on a regular basis, but just at a slightly greater scale in terms of the size that we would be ordering from some of our colo providers.

The next question will come from Jackson Ader with KeyBanc Capital Markets.

The first one I had was, Ed or Dr. John, did you guys talk about how many different sites? That once you actually get fully ramped by 2028, how many different colocation sites this would actually -- like the footprint would actually be?

It would be numerous, but we haven't disclosed the number.

Okay. And I think you mentioned that it was going to be mostly based in the states. But -- and then, Ed, when you talked about like this fitting into your profitability profile, like the profitability profile you gave was mostly for the GPU and the inference cloud. But if we're more efficient maybe with CPUs on a power consumption basis, do you see yourself thinking maybe more CPUs might be kind of the way forward for incremental margin improvement versus the inference cloud?

Yes, it's a good question. And obviously, what we gave you the numbers for, and you can sort of put it all together, you'll see it's a pretty attractive portfolio. It is a mix. But obviously, CPUs have a different dynamic in terms of you can get a lot more CPU, therefore, generally more revenue per dollar of megawatt of power. So yes, I'd say in terms of demand going forward, we see demand for both. Obviously, there's a lot of talk in the industry about how much CPU goes alongside these big AI deployments. There's an awful lot of CPU that's needed to run all this. It's not all just GPU.

So look, I think depending on the mix of business we have, you can get margin expansion also through GPU depending on the pricing and the efficiency there. Not all GPUs behave the same either. So again, this is really good business for us. I'm very happy with the entire portfolio we signed up this year. And it's -- I think a great mix of business that we have with customers and with both GPU and CPU.

The next question will come from Rudy Kessinger with D.A. Davidson.

Congrats on getting this deal done. I have a 2-parter because the answer to the first part maybe impact the second part. But the additional $9 billion opportunity on this deal, would that require additional CapEx? Or would the $5.5 billion of CapEx be suffice to serve that additional $9 billion of revenue? And then just based on that answer, I guess it might change the second part. But the second part is getting a lot of questions on just the ROIC here of the total CapEx outlays. Obviously, the operating margins from all the math I'm doing, look could be about 30% or so based on the assumed depreciation in colo. But just talk about the ROIC on the total CapEx so far for this $14.6 billion in commitments.

Yes. So the answer would be the next $9 billion, we don't know what that's going to be yet. That would be additional CapEx. And so once we sign those, those would obviously be large enough that we would give you some color in terms of what we were doing along with the CapEx and expected revenue. As far as ROIC, we don't break that out, but I think I've given you guys plenty of information to do math not only on this deal, but on all of our deals, and you can see it is a very attractive return on capital.

The next question will come from Fatima Boolani with Citi.

Congratulations on the marquee transaction. Dr. Tom and maybe Ed, I wanted to just get a sense of the motivation for the financial and investment relationship pursuant to the warrant that is tied to this arrangement with Anthropic. I wanted to just unpack some of the motivations behind that as being part of this announcement.

Yes, sure. The warrant actually provides incentives for future business and growth, as you can see. I think it helps align interest. And it's compared to the volume of the deal, it's pretty tiny. The value of a warrant at the strike price that we issued at about where we closed today, $111 or so, is about a little over $150 million, and that's part of a transaction that's worth $11.6 billion. And if you look at the $20 billion deal, if we were to do the extra $9 billion, that 5% warrant has a value today of less than $400 million. So you can see the financial aspect of the warrant is very, very small compared to the overall revenue to Akamai in what is a very strategic relationship for us.

And just as a...

Just one thing to add on that. Just a -- I was going to say -- just to add on something here. Just as Tom mentioned, the value of the warrant, when you issue a warrant like this, you have a grant date, which was the day we signed the contract. So therefore, the value is set. So to the extent that there is future issuances or future vesting of the warrants, that number is already known. So when -- so you don't have to worry about, hey, the value of that's going to accrete over time and that's going to have an impact on the revenue. It's going to have a very small netting impact on revenue going forward.

I appreciate that clarity. And then just as a related matter, just given the strategic nature of this transaction, does this preclude you from similar engagements with some of the competitors of this particular customer? Just wondering if there are some limitations in your ability to replicate the success with other comparable frontier labs.

Yes, no limitations at all. And you should think of Akamai as working with all the major players out there in AI, including the hyperscalers who are all customers of Akamai and all use our cloud infrastructure services. So yes, we work with all the major players, and this deal doesn't change that at all. In fact, maybe probably helps our ability to do that.

We have time for one more question, and that question will come from James Fish with Piper Sandler.

Maybe just working off of the theme of last question there. It's a bit of a unique deal structure. First, how should we actually think of other major potential customers come here given Akamai's attractive footprint versus just needing to focus in on executing on this massive expansion? And then second, should we expect more deal structures like this with some of these larger frontier models out there?

Akamai is very good at scale. As you can tell from just our footprint today, the volume of traffic we carry, the volume of security work that we do, the number of major enterprises that we work with. So I think we have no problems, really, at all, working with other major players on deals like this. And you should think of us as already working with the major players in AI. And of course, it's our intention to keep growing the CIS business and Akamai overall. And Ed, do you want to add to that?

Yes. The one thing I would add here, Jim, is since we've started to do these larger deals, we -- the opportunity set has increased. So it's actually, as Tom said, it's sort of the opposite that by doing something like this, there's no contractual like reason why we couldn't work with anybody for sure. But you do find that folks will reach out to us that may have not worked with us in the past or some that have that say, "Boy, I didn't know you could work at that scale. I would love to sit down and have a conversation about doing something larger than we had ever contemplated doing with you."

So it does sort of feed on itself. And obviously, we've been pretty busy here over the last 9 months announcing a lot of really big and exciting opportunities. So as Tom said, we're not limited in what we can do. Obviously, there's some scarcity in the marketplace. So it does take some time from when you announce a deal to when you start recognizing revenue. But we're open for business and I hope to do a lot more of this in the future.

This will conclude our question-and-answer session as well as conference call. Thank you for attending today's presentation. You may now disconnect.

Thank you.

All right. Good afternoon. Thanks so much to everyone for joining us at the Akamai session at the Goldman Communacopia Conference. My colleague, [ Max Campbell ], and I delighted to have on stage Ed McGowan, CFO of Akamai. Hey, thanks for being here, Ed.

Thanks for having me. Great conference so far, so thank you for doing this as well.

The pleasure is ours. Ed, you're in the middle of a very interesting inflection point for Akamai. Give us a little bit of a download of the last 6 to 12 months when you all as an executive team realized that this network that you've built over multiple decades can now, essentially, be upgraded along with the compute assets that you have into something that is a next-generation AI compute network?

Sure. Yes, it's been probably the most exciting time since we first began. I've been with the company since 2000. And back then, demand was insane, right? Pretty much the Internet was just exploding with incredible demand. We're seeing that now with AI. And we made a decision 5 years ago to get into the compute business, and we acquired Linode, and we always had this vision of, A, using it ourselves to start because we're spending so much with hyperscalers, but to make that into an enterprise-grade network that would be comparable to using a hyperscaler. It may not have all the capacity and all the features, but it was an alternative. And we thought that, look, we can leverage a lot of what we've done over the years with our CDN network, leveraging a lot of the same teams that build some of the functionality, running a distributed platform, the operational teams, et cetera.

And I think the big turning point for us is when we sat down with NVIDIA about a year ago or so and realize that there's this big opportunity to lean into not only GPUs, but compute in a much bigger way and start building out bigger locations and offering an offering for compute from large-scale CPU and GPU deployments, doing things like post-training or support systems that go along with a lot of these big AI companies, and leveraging our platform all the way down to our MCS, our managed container service, where we're taking the Linode technology and slicing off part of the CPU that we're using for CDN and for compute to run company's code, to do things like manipulation of video and synchronization of video, doing live, real-time ad decisioning and things like that.

And it goes way beyond what we were doing before with Functions-as-a-Service where you could do basic programming and web assembly or JavaScript. Now the customer has a lot more ability to do a lot more compute right out at the edge. So from core data centers all the way to the edge.

Tell us a little bit about how you've been able to merge the best of the Linode assets, of the best of the Akamai assets to the point where you're signing multi-hundred-million, multi-billion-dollar deals.

Yes. So it was really pretty simple. The first thing we did is we, over the years, built a massive backbone. So we carry hundreds of terabits a second of traffic across our backbone. We stitched together our CDN locations across the world. So we have this asset that's connected up to all the locations that we acquired from Linode, and we built out another 20 or 30 or so sites that were initially 5 to 10 megawatts in size. So they could do a decent amount of revenue, a couple of hundred million of revenue per site.

And all we're really doing is just now expanding those sites to be a lot larger. So leveraging the same teams that build the CDN network, that negotiate for colo, working with a lot of the same colo providers we've worked with in the past, and building out the platform to be able to take on all sorts of different chipsets, both CPU and GPU, and it's been a relatively light lift in terms of investment in people and technology. It's been really just doing a lot more of what we're doing at a bigger scale.

Tell us a little bit more about how you go through network upgrade cycle. And what I'm getting at here is we sometimes hear the bear case that Akamai's network is built over multiple decades. It has a legacy field to it. Therefore, it's not going to meet the performance and cost requirements of a frontier model today. Tell us how that statement is wrong.

Yes. So first of all, most of the 4,000 PoPs, we wouldn't use for a frontier model, right? So maybe if they were using, say, there was a CDN application that was going alongside that, sure, we could use the...

Different use case.

Exactly. Very different use case. In terms of what we're buying today for building out, let's say, a customer comes to us and says they want 5,000 GPU of various shapes and sizes, or they want 10 million virtual CPUs, or whatever. We're using the latest hardware, latest chipsets, et cetera, and we have the visibility of getting a big order for multiple years, and it's just the most , and it's just the most modern equipment that's necessary, that performs very well.

And to the extent that it's, at all connected to the network, if they're, say, running some code in the servers that we have for managed container service, that's all modern and usable today. We don't need to go and upgrade the network. There's no notion of saying, "Hey, we have to throw out what we've done and rebuild something else." We're just adding locations with more servers.

Makes sense. Let me ask you one more here, and then I'll turn it over to Max. When you sign these big initial commitments with customers, you've given us a couple of nuggets on the unit economics. But I find what's interesting is this could be the beginning of a much longer-term relationship where you cross-sell into these initial customers that have signed up on the compute side. What does that customer journey look like for -- maybe take the 2 larger deals that you've announced today?

Yes. So it's interesting. When you sign one of these big deals, it creates this sort of network effect where we get a lot of inbounds today, like the pipeline is bigger than I've ever seen. I've never seen anything like this where people realize, "Okay, you are an option at a scale that I didn't think you operated at." And then once you get in and start doing work with any of these bigger companies, like the biggest companies that we're working with, foundation models as well as others, you uncover a lot of other opportunities, right?

I think what people don't quite understand about us is that our business isn't just GPUs. That's not all that's growing. As a matter of fact, the pipeline is mostly CPU. A lot of our revenue today is CPU-generated. It's a little bit different in terms of the unit economics, power usage, and things like that. But what goes along with these -- even the big investments in GPUs, there's always some drag along for CPU and storage. And my CTO said to me today, think of the GPU is the brain, but the CPUs are all the arms, legs, the tooling, et cetera, that these systems have a lot of subsystems.

So I might be working with someone who is spending billions with one of the hyperscalers, but I may carve off a piece of support system to several hundred million of CPU. And as you start working with these customers, you find more and more opportunities to grow with them. Over time, think there's opportunity to cross-sell security. A lot of the customers we're working with today are already existing customers of security and delivery. And the demand is coming from new logos as well as existing customers and expansion orders with some of the customers that we've already done some compute business with already.

You've now signed over $2.8 billion in multi-year cloud infrastructure commitments. What are the most important milestones, operational milestones, between signing those deals and then seeing them show up in revenue?

Sure. Yes, great question. And it's funny, the way we're managing the business is pretty interesting. So we start with the sort of 3 components, right? There's colocation or power, there's capital, and there's equipment, okay? And each one of them has a little interesting twist and dynamic to it. So we're very fortunate that capital is not an issue for us. We have plenty of capital, and we have the ability to raise capital if we need it. So that's not a big problem for us. Colocation is probably the most challenging issue. Power and space is probably the area where -- that is the biggest locker to near-term growth. And nobody has a lot of power and space just laying around.

We have a very different model where we work with hundreds of colo providers around the globe, and we're actually a very interesting buyer for them because one, we're investment-grade credit. And two, we buy a fair amount of colo. So someone's -- like some of the major colo providers are building 5 or 6 sites across the country. We may take 20 or 30 megawatts from them in 5 or 6 different locations or whatever. So we can make long-term commitments of good financial backing. And we're able to cobble together in multiple locations, stitching that all together to our -- with our backbone in areas that might be cheaper than some of the major cities and that sort of thing. And we're not building these gigawatt facilities or anything like that.

So getting power is sort of step number 1. So I've got a lot of demand right now. And what I do is I sit down with my operations team and say, "Let's map out how much power we can light up between now and say, the end of '27 and then give me a look at what '28 looks like." And we'll start to map that out by month and by quarter. And then I'll go back to the customers who will come to me and say, "Ed, I want to do 5,000 GPU. I need it to perform like this. I want to use this chipset. I want to do this or that. And I'd like to have that up as fast as possible." I say, "Okay, I can slot you into these locations over this time period," usually, you get an agreement, it might take you 90 days or so to kind of hammer out an agreement. At the same time, you lock in your colo, you're now locking in your supply chain.

So I'll be ordering the chips, be putting together the equipment. We use contract manufacturers to put all that stuff together. Sometimes you're buying, say, like a total NVIDIA stack. Other times, you're just using CPU and putting the pieces together. And we'll make sure that if there's any delay or any sort of time difference where we're, say, risking memory prices going up, we've sort of factored that into our agreement. But it gives us an enormous visibility to where we can start to slot in some of these bigger opportunities. And I'm making my colo buying decision based on what my pipeline looks like.

I want to secure -- let's say, I'm going to sign 100 megawatts, stay in round figures. I'd like to get, say, 70% to 80% or greater of that secured with deals behind it before I place that order. I'll leave some amount, 10 or 20 megawatts or whatever for stuff that's not done or maybe there's options that I have to take additional colo, but I try to keep that as tight as possible because that's the one area where that can do the most damage to your P&L near term. So if you think about signing the deal, ordering the equipment, getting the colo, it probably takes 6 to 9 months between signing the deal and getting to revenue. And you can start to take the expense for colo the minute you get the keys to the facility. Depreciation can be pretty tightly linked to revenue. So there's a bit of a delay. So if I take on 100 megawatts of power, I'm going to have a pretty big expense. Even if it's for 90 days, it can put a point or 2 of pressure on my margin.

So I try to keep it as tight as possible. So it's really about as soon as I light up power, and I know when power is coming in, I can start to slot deals into that. So it's been -- operationally, it's a fairly easy thing to run and power is definitely the one area that is sort of the gating factor, if you will.

Is there any color you can provide us on the CIS pipeline beyond the $2.8 billion committed amount right now? Is this the limiting factor mainly colo and power? Or is there anything else that would hold you back from onboarding more customers?

Yes. So I would say we have line of sight to add a significant amount of colo between now and the end of '27 and even more in '28. So I would say that's not necessarily a big factor as far as prosecuting the pipeline. I've got a very, very robust pipeline. You need to go and qualify some stuff that's in there, doesn't meet our margin requirements, might not be the right fit. You don't want to take too much of a bet on startup companies, for example, that might have a different financial profile, especially if I've got enterprises in there that are better credit quality and that sort of thing.

So the near-term problem I have here is that I don't have a ton of colo just sitting around. So there's -- the growth -- we've already talked about having growth accelerating from the 6%, 7% now to the teens for next year. And that's with what we have signed today. So I still have the opportunity to sign more, but now I'm starting to build stuff that will come on late in '27 and into '28. And I see really no issue as far as having this massive pipeline. It's just a question of how quickly you can get the colo lit up and then there's backlog for computer equipment. So it takes some time to get certain equipment.

Let's say, for example, I want to sign -- sell Vera Rubin. I can start taking order now, but I won't get the chips for at least 6 months to -- months to a year, right? They're just not physically available. So there is -- we're in that period of time now where we are signing up big deals. Some of them are already inked and we've already announced them. They're starting to produce revenue in Q4 and will ramp up throughout '27. A lot of that's in the first half of the year. And we're starting to sign up additional deals to put more revenue into '27 and a lot of revenue to '28.

That makes sense. And you talked about a meaningful acceleration in CIS revenue starting in Q4 and then into 2027. Are there any directional clues you can give us to help us model this ramp from the outside in?

Yes. So we've given you a little bit of a heads up on -- in terms of some of the bigger deals we've announced and how much revenue they'll produce. We're expecting $15 million from one of them and somewhere around $20 million for the other. So that will significantly increase the growth rate for Q4. There's some risk that if stuff comes in a week or 2 late, maybe it pushes a couple of weeks. But generally speaking, that's -- you're going to start to see a big ramp here, and then it continues to ramp into '28 as we -- sorry, '27 as we get more equipment and we get more power lit up. But it's all structured that we should start to see a significant acceleration in Q1 and then into Q2 and a little bit into Q3.

Understood. The upside from AI compute is clearly large, but investors are also focused on protecting the downside. So how are these contracts structured to protect if any of these deals continue to slip? Or you mentioned that memory costs, you have hedged against that. But what about other delays? How are you protecting [ those ]?

Yes. So generally speaking, what you do both on the colo side and on the customer contract side, if you're signing, say, a 4-year deal or a 7-year deal, you will build in like some type of an escalator to cover costs, labor costs go up, colocation costs go up. And with our colo cost, we do colo deals, we'll do the same thing. So I might have a 2% or 3% escalator each year. The way the accounting rules work, you have to straight-line that. You also have to straight-line your revenue. So it just comes out as flat across the period of time. You don't have that bump in terms of revenue and costs and that sort of thing.

But that's generally locked in. So there's really no risk of that blowing up on me. If something comes in, say, a few weeks late, let's say, there's a delay in construction on a facility for colo and we expect to start on October 1, and it starts on November 1. All that does is just start the revenue clock later. There's no penalties. The deal doesn't get shortened or anything like that. It's just literally you have a -- you're working very collaboratively with your customer saying, this is the best estimates we have from the construction site to the manufacturer in terms of when we get the equipment. And as soon as the equipment is lit up and it's available for the customer to use, these deals generally are like take-or-pay, right?

So as soon as the capacity is available, you start billing for the amount of capacity that's available. And sometimes it might be a ramp of, say, 3 or 6 months to get to full ramp. Other times, you can do it in a much shorter period of time. Now if I'm building out a, say, a big deployment and I say, once I sign the contract, I'm signing a contract with the manufacturers and with the colo providers. Colo is easy. Usually, you can get that locked in and I can actually say in the contract, "Here are the 3, or 4, or 5, or 6 locations we're going to put you in. And if there's any kind of an issue, maybe there's a backup location where I might put you in."

And with the equipment, let's say, I can get 80% of it contracted with all the memory and everything like that locked in the day I signed the contract. And now there's this 20% or so that could potentially flux, memory being the biggest driver of that. You work something in the agreement where you would say, "Look, if the prices go up by a certain percentage, we just pass it straight on. If it goes up by, say, 50% or more, we'll have a conversation and maybe we put a pause and wait a few months or maybe we just continue to build and we just pass the price on." But the concept is that we've sort of agreed on what the "margin" looks like at a high level. In other words, like how much pretty transparent with what our costs are, that if the component parts go up in price, we are going to capture that.

Now once I have that locked in, now I've got the hardware, I'm just depreciating that sort of the cost is already set. The colo, we've taken that into consideration. We're building in slight revenue ramps as well. So we have definitely modeled in the "risk" or downside as we're signing these big deals. And to the extent that I'm building ahead of demand, I have the ability to raise my price. So let's say, if component parts go up or colo costs go up, and I'm not able -- I have to secure at a higher price, I'm just going to raise my price. And we've done that. We've actually raised prices in some cases.

Got it. Historically, you talked about CIS resulting from $1 in CapEx and turning into $1 of revenue over time. Is that still the right way to look at this? And what's the time line to get there?

Yes. Yes, good question. So when we first bought Linode, that certainly was the case. So we would see that over time, you spend $1 in CapEx and you get about $1 of recurring revenue. I think initially, when we started, there was some confusion because we were our first biggest customer. So if you looked at our CapEx and you looked at our revenue, you say, "Well, wait a minute, Ed. I'm not seeing that manifest itself in revenue." Keep in mind, we're spending well over $100 million on third-party cloud, and we brought a lot of it in-house. So I guess if I translated that into revenue, if I were a paying customer, it would probably work out to be pretty close to that.

With some of these bigger deals, we're not necessarily getting $1 for $1. It's generally between, say, $0.50 and $1. There may be an occasion where I might take something a little bit less. But I'm doing that with the -- keeping in mind that there's sort of 2 main drivers of cost. There's power and there's equipment. So if I'm willing to take a slightly lower yield, that means I'm doing much better on the power throughput, right? So I'm getting significantly better dollars per megawatt of power in a situation like that where I have slightly higher depreciation. I sort of gave you on the last call some margin guidelines.

So as we look at these big deals, I'm getting somewhere in the, call it, mid-20s on the low end on the operating margin to mid-30s on the high end. And that's for the bigger stuff. For the regular way business, it's even better than that, right? Because you don't -- customers don't have the purchasing power. They're not locking in for 4 to 7 years. You also have to assign some value to the relationship as well. There's strategic deals where you might say, "I'm willing to maybe go a little bit lower on that margin scale in the 20s because I get a relationship with customer X that I've always wanted. I think I can either make up that margin with other products down the road or it's just an opportunity to get significantly more business."

Ed, I'm curious on this point. We have so many data points that are a little bit apples changes on ARR per megawatt, revenue per megawatt, GPU payback period across the hyperscalers, across the neo cloud. I'm curious when you do your internal benchmarking, how you think about your price point relative to all of these other data points floating out there on the ether.

Yes. So obviously, with the bigger deals, it's a lot easier to do the math, right? You can say, generally speaking, you're getting $15 million to $20 million per megawatt. There's cases where you, with some CPU deals, you might get 25% or better. You could even see sometimes in the 30s with some of the smaller customers and things like that. We won't take something that's significantly lower than, say, 15%. Like maybe there might be some strategic reason you might take something slightly lower than that. But generally speaking, you're somewhere in that range.

Now initially, there can be some build-out where you would say, "I need to build out for spares and some excess capacity." So if you look at some of the metrics, you might say, "Hey, you're putting an extra 5% into this deal to build out some headroom. Now as you get bigger and you have more deployments and location, that comes down over time. So I would expect your sort of revenue yield to be a little bit better. And then also, not every deal behaves the same, right? So I can have 2 different customers that are using 2 different chipsets, and let's say they're using compute. I might get a better yield on my revenue per megawatt from a particular customer just because of how efficient the chipsets are and the servers are and that sort of thing.

But generally speaking, we try to land somewhere in that range. And if you think about cost, the cost of a megawatt of power, say, $3 million to $4 million a year, sort of on the higher end, I would say, in the U.S. I'm sort of building in a little cushion there for inflation and that sort of thing. Your depreciation can run $3 million to $5 million, something in that range. So you kind of -- as you work through it, you can see where you get to those margin targets that I talked about. Now there's some other ancillary costs in there. There's a little bit of people costs and maybe a little bit of, say, like maintenance cost for your switches, and your equipment, and that sort of stuff. But generally speaking, the bulk of the cost is colocation, and by far, the biggest cost is depreciation.

One of the debates we've been having actively in the last few weeks is we're in a period of time where supply and demand is just so tight. What happens to that $15 to $20 when the industry potentially goes into excess supply? And how does Akamai think about their role in perhaps derisking some of these contracts in an eventual excess supply environment?

Yes. So it's funny. If you look at the like history, you just do some research and just say -- ask one of the big AI models to do it for you, and say like, "What is the hyperscalers typically got? What do the other folks get out there?" You see that, that range is pretty consistent, right? It's sort of held up historically. Could you get into a situation -- maybe we decide there's business we're not going to take. And also, we have a value proposition as you think about kind of the next generation of where sort of this is going where there's a lot of infrastructure going out for building models. You've got these open source models. You've got a lot of post-training going on. You've got now just starting to see some of these agentic applications that are being built that will use a lot of compute, but not in massive amounts, right?

So you might have someone who used to have a website now has this agentic agent that's being like a travel agent, or a personal shopper, or something like that. Well, that's going to require a lot of compute and their latency really does matter. So I think you're going to get a higher yield for providing that type of a service, right? So maybe the bottle switches and we're not going after some of these if the yields don't make sense. But here's the interesting thing. Right now, I can see demand into '28, right? So I'm able to park business out into '28 today, which is very unusual, right?

And that's sort of the way the industry is. Nobody is sitting around saying, "Hey, I've got X amount of power available, just I can deliver it to you on Friday. I can get you all the equipment. It's not sitting around now." Will that work itself out over time? Probably. But I think it's several years out. I don't see this being like a '27 or '28 phenomenon.

One more on CIS before we move on to some other topics. One of the advantages for Akamai is clearly the distributed network. But how do you think about the balance between centralized GPU deployments and more distributed inference at the edge? What customer use cases are actually ready for edge inference today versus still more of a longer-term vision?

Yes. So there's -- right now, I would say anything that has a latency requirement like robotics or autonomous driving, anything that has to do with interacting with the consumer where you don't want to introduce latency into the experience would be anything that could be effectively a use case for that more distributed type platform. There's also an element of performance, too, that comes into play where customers will have an expectation of investing a dollar and getting a return out of that equipment, whether it's a number of actions that they can produce per dollar of spend, et cetera.

So that's another element of performance. And in some cases, we can outperform competitors just based on the economics that they can yield for however they're looking at, how their platform runs on a particular set of machines and equipment and that sort of thing. So it's not -- that -- and also in some cases, for inference, you don't necessarily need GPUs either. You can run it with CPUs, right? And so we're perfectly positioned for that. And I'd say we're just starting to see more enterprises heading in that direction. I think so far, the dollars have been primarily in building the big training models and all of the subsystems that go along with it. So that need for more of a centralized, but -- centralized but distributed, meaning in a handful of locations, not in tens of or hundreds of.

Do you have a latest update on how many locations you have deployed GPUs thus far or -- and you also have RTX, like lower-end GPUs deployed. Are these also included in these...

Yes. I get the number we -- last quarter, it might have been 20-something, might be the number. And that will just go up based on customer demand, right? So if somebody wants to go more -- there's nothing preventing us technologically. As I talked about with colo, we can get colo in many, many locations. So it would be fairly easy for us to increase that. It would just be based on customer demand.

Got it. Moving on to capital allocation. So you've raised convertible debt earlier this year. You still have a very strong balance sheet. As the pipeline grows, how do you think about the right funding mix between cash future debt capacity and then internal cash generation across the business?

Yes. So obviously, I want to fund as much of it as I can out of my cash flow, right? So the -- it's interesting if you model this business out and you say -- just doing a fun exercise and double the size of the company with just CIS revenue and use the metrics we gave you and then say -- at the end of that, just go to say the company goes to 10% growth or something like that and watch what happens to the margins. It's fascinating. EBITDA margin expands dramatically, free cash flow expands dramatically. And if you want to grow at 10%, say, you're twice the size of the company, you might need 20-something percent of CapEx to grow at that level.

Say, you've had a 50% yield, $0.50 on the dollar for CIS. So the multiples will expand and the free cash flow will expand, EBITDA margins will expand. So higher EBITDA margins give us a lot more ability to raise debt if we needed to do that. We're also going to produce a lot more cash. These models are great. Once you spend the money, you get the payback relatively quickly on your CapEx. Sometimes it's 2 years, sometimes it's 3 years. And they're very high EBITDA margin, very high free cash flow margin after the -- after you spend the initial capital.

So we will be able to fund a lot of this growth in the future from our cash. So far, the convertible market has been very good to us. We do want to maintain an investment-grade credit rating to the extent that we can. And the only reason for that is it is an advantage when you're dealing with colo providers. If we lose it, it's not the end of the world. I think if we lose it, it would be very temporary because of the EBITDA margins will expand very quickly thereafter. So if you -- sometimes you see the rating agencies if you do an acquisition and you have a thesis where you drive synergy, they'll give you credit, they might put you on a negative watch, but you get back to investment grade or stay at investment grade.

So that will be our goal. If not, I have to put maybe some deposits down when I do some of these colo deals or get letters of credit. So it's a little bit more costly, but not the end of the world. But our goal would be to try to keep that. And so far, debt has been the right instrument for us in terms of raising capital.

Moving on to security. AI seems to be helping some of Akamai's more mature security products, including web application firewalls, DDoS protection. Do you view this as a durable reacceleration in that more mature security product category? Or is this more so a one-time upgrade cycle that you're seeing?

Yes. It's interesting. I talked about it like demand at the top of the funnel has increased dramatically since Mythos has been released, that CISOs are bringing all their vendors in and say, "Okay, what am I doing today? Am I utilizing everything like I should?" For us, an example would be, "Hey, I'm using web app firewall for all my public-facing applications, but not my, say, supply chain or my, say, private network client portals and things like that." Well, they're just as susceptible to these types of attacks. I need to put WAF on every application. We're seeing a big jump in bot management, demand for bot management. what's happening with these AI models is it's creating a lot of machine traffic.

So understanding what is that machine and then what action do you want to take? Do you want to block it? Do you want to send it different information? Do you want to keep it away from the crowned jewels, maybe keep it away from the paywall, but let it get other free information. So there's a lot of opportunity for that business to grow. And I think that will continue. And then for newer products like LayerX, there was an announcement today from the Meta around that whole -- what's it called? Muse, I think, where they're creating basically the ability for individuals to have their personal agents do things, let's say that. Somebody told me about that in one of our meetings.

Anyway, what that does now is it creates a big threat vector for all the endpoints right? So LayerX actually can help defend the endpoints from, say, data exfiltration or data loss and that sort of thing. So I think there's an opportunity for that as agentic applications grow. And then for Noname, with our API security, with these AI models, there's a lot of back and forth. It might not be API, but maybe MCP as the protocol. That also creates a big threat vector. And I think that's -- we're perfectly positioned to try to leverage that to solve that problem.

On that point, across the security portfolio, where do you think the biggest tailwind is from AI? Is it the product tied to CDN, Zero Trust, or API security?

Yes, it's a little of everything really. So for example, Guardicore, you might be deploying Guardicore and you say, "Okay, the benefit of Guardicore is once something gets in, you limit the damage, you can set up rules in your network to segment things that if a virus were to get in, it will block certain things, bad things from happening." And you might say, "Hey, I'm going to roll that out to a piece of my network today and then slowly roll it out over time." This puts more pressure on the CISOs to have that, if that's the way they believe in the last sort of layer of protection to roll that out faster. So I think Guardicore gets a leg up.

As we talked about, LayerX has got a big opportunity initially with just protecting the browser and the user from taking the company information and sharing it out with either an open source model or 1 of the foundation models. You can use LayerX to prevent that from happening. And then we're working on with LayerX, and with Noname, and some other technologies, creating the ability to protect the cloud deployments for agentic applications running in the cloud, not just at the endpoint.

And you made a really good comment earlier in this conversation about Akamai has been through a cycle like this before in 2000, 2001. And you're one of the few management teams that actually has that tenure under your belt. What do you think is different this time?

Well, I mean, obviously, it's a lot easier than the equation we talked about in terms of having capital, being in a great position to be able to start a business like that, that is more capital intensive. Back then, we were -- we had a metric called QTL, quarters to live, right, because you were burning cash, right? Here, we're in a position where we are a profitable company, making an investment. I think we're being very smart about the customers we're working with, the amount of capacity we're adding, trying to keep that supply chain as tight as possible, right, knowing that, look, these things, they don't live forever.

There will be some -- at some point, things will slow down. We don't want to overextend ourselves. We do want to invest in growth. So I think that definitely helps you. But also, we're leveraging -- like we're monetizing a part of our business, which was our ability to scale and build a network. That now is a very valuable skill. The relationship we have with our colo providers has provided us this sort of niche where we're really the only ones out there working in this model of cobbling together across many colo providers, an interesting fabric where you can do billions of dollars of CIS revenue, right, without having to go and build your own data centers.

So I think it's sort of leveraging all that experience and then the 20-plus years of enterprise relationships with customers, right? That's a huge advantage. So you hear about some of the neo clouds and DigitalOcean saying they want to get to the enterprise customer. That's hard to do. Like we've built up 20 years of trusted relationships. And so we're already having conversations with customers that say, "Hey, I want to build my own open source model and run it or I have some agentic application that I want to build. How do I think about performance? What can you do for me?" So it gives us a significant advantage across many different vectors.

Ed, well said. Congratulations on the new milestone. Please join me in thanking it for his time.

Thank you.

Everyone, I'm Fatima Boolani. I jointly head up the software research team here at Citi. And I'm very excited to host my next fireside chat session with Dr. Tom Leighton, CEO and Founder of Akamai. Thank you so much for being here.

Well, thank you. Nice to be here.

I appreciate it. Well, we've got a lot to talk about. And where I want to start the conversation is Akamai's strategic evolution. The business has undergone a substantial change and transformation, certainly in the last 3 years, but I'd even narrow the scope and say in the last year, having essentially evolved from a predominantly content delivery services business to a majority of compute and security business. So in your view, as we think about the evolution and some of the milestones you've hit in the last 12, 18 months, what comes out most sharply in your mind as things that have been most salient in that evolution?

I think there's been a big advance with our cloud capabilities. We bought Linode several years ago now, huge investment to make that be of a caliber that big enterprises can use for big mission-critical applications. And now we're to the point where we can sign up these very large cloud customers and support their AI needs, end-to-end, training all the way through inference and get their compute applications close to users so they can take advantage of low latency and also high throughput for video kinds of situations. Maybe you're in robotics and the robot is seeing a lot of things around it or it's a car or something, and you need to process those videos at scale really quickly. And it's really been the last 12 to 18 months where we've gotten to that level, we can go out there and compete head-to-head with the hyperscalers, the neo-clouds, whoever it might be and succeed.

Dr. Tom, I wanted to take a step back and actually ask you maybe a little bit more of a bigger picture question. As a founder of the company, you've been at the helm and seen this business see massive evolutionary leaps, right? You were at literally the front seat guiding and architecting the role you would play in democratizing the Internet, right? If we go back all the way back into the late '90s. So seeing that transition and the explosion of the value that you created in the content delivery. Then you pivoted to providing security around all this traffic that was being created. And now we're in the AI era where compute is really shining.

So can you maybe juxtapose or compare and contrast for us with the benefit of having watched and witnessed these paradigm shifts in the past, how would you compare where we are from the AI-native and AI adoption era relative to these past other cycles or computing evolutions that we've seen? What's different? What's more dramatic? And maybe what's the same and maybe history is rhyming again. So I would love to get your perspective.

Yes, it's a great question. I think there are a lot of parallels. What we're seeing today and what we saw in the late '90s is the Internet exploded onto the scene, a fabulous new technology is going to change everything, a lot of great use cases, also some bad use cases. There's...

Solution for that.

We were working on it, yes. I think a lot of bubble kind of excitement in a way, but also a lot of reality. This is going to be here and change our lives. So that's all very similar. In terms of Akamai and what we can do, there are similarities there. Back in the start when the Internet came on the scene with the web, with the website, it was hosted in one spot, 1 data center, maybe East and West Coast. Well, you see that today with some of the first AI use cases, all 1 big data center. That doesn't scale well. It didn't work in the web. You had the World Wide Wait. And of course, that's where Akamai helped by being distributed and getting the content closer to users and staged there, so you get faster delivery and scalability. Same thing happening with AI.

To scale these applications, you can handle and process lots of videos or create a lot of individual videos for users that are personalized or to have conversations in real time, you need to be distributed. And we're just at the start with that. And we're using the same kinds of capabilities we developed back then now for AI. Initially, costs were very high back then, and then a lot of work was done to decrease cost. I think you're going to see that in AI. Even the notion of caching back then, which made it a lot more effective, we're at the early stages now with things like semantic caching.

So 2 people ask, really, they're asking the same question, but in a different way. You don't want to go back to the expensive model to answer it the second time. You want to realize, hey, I've answered that question, and here's what it is. And to be able to do that, we've got some partners now starting to do that with video, where instead of generating the video from scratch, which is expensive, you've already got a frame for it and you tweak it a little bit. You change how the lips move as you change what they're saying, the avatars are saying.

So there's just, I think, a lot of parallels there. And what we do is it's amazing how similar it is. Same with security, again, back then, you'd have your WAF in your data center, which at the time was fine, but then you got these bigger attacks, the WAF is overwhelmed and you couldn't do it anymore. And so of course, we built the cloud WAF. So it's fully distributed. And now you can absorb all that attack traffic because you are distributed. And just in the last year, the size of the attacks has gone up a factor of 10 because the bad guys have AI, they can go take over a lot more devices to launch these attacks. So it's striking the similarities we're seeing today.

I know we're going to talk a lot about the compute business, a.k.a., the artist also known as the CIS business. But before we focus on that, while delivery is only 1/3 of your business, it is also still 1/3 of your business. I know you've worked very, very hard to diversify from the content delivery, building adjacent product capabilities and security and on the connected cloud vision. But as you talked about the explosion of the Internet, the Internet is exploding again by way of AI traffic and traffic growth. So the surface area of the Internet is just expanding and metastasizing because of AI.

You have bots, you have AI bots, you have machine-generated traffic. So can we unpack a little bit of that in terms of because you see so much of the Internet traffic essentially flowing through your pipes, how should investors get a handle on what the complexion of Internet traffic looks like today than what it did even, let's say, 3 years ago, ChatGPT came out, back half of '23, (sic) [ '22 ], right? So that composition and complexion of Internet traffic today and how it differs or contrasts traffic patterns from 3 years ago even.

Yes. Good question. And I think there's probably a lot of misconceptions out there. Already for probably most of our customers, most of the entities that were coming to their sites and apps weren't human. They were bots of various kinds. And of course, now the variety of those bots or agents that are coming to their site is proliferating. And this does present challenges for our customers in terms of what they want to do in response. And the good news is we have the market-leading solution for bot and agent management. And so when our customer decides how they want to treat any particular kind of agent, we can make that happen for them.

And of course, with the proliferation of agents, they aren't all secure and you got a big security challenge there. I was talking to a banking customer yesterday, and they're going to block agents, user agents just because they don't have the level of security and the bank's on the hook if user data gets out through them. So there's a lot more entities out there, but it was already mostly automated interactions. Now in terms of traffic and the revenue for our delivery business, that's based largely on the number of bytes we deliver. And that's not yet changing because of AI.

When the agent is having -- transmitting text back and forth, that is tiny. The bytes come from video watching. So if everybody watches more video online or watches in a higher quality, that can make a difference. And for software downloads -- gaming releases or all the stuff you put on your devices. And AI is not really changing that yet. Now, someday, it might. If our agent manifests itself as a human, and we're talking to it all the time, it's like our personal shopper, we become buds with our AI personal shopper, well, now video is going.

Or if really all the devices out there are recording what's happening and video is being processed, okay, that makes more bytes. I was in China recently and saw some cool AI glasses that are recording everything that's going on that the person sees and sending it into the cloud real-time and then information comes back and is displayed on the glasses in real-time. In this case, they were using the glasses to translate my talk.

So I've got English up on my PowerPoint and they're seeing Chinese on their glasses. If you start having that at scale, that can generate more traffic. But today, those things really haven't started happening yet at scale. But you're right, there's a zillion agents out there and they're only increase, and that's a big security challenge, helps with our security revenue. And of course, helps with our compute revenue because that kind of stuff runs on our cloud. And that's where we're seeing the big growth.

So is it fair to characterize that the way you win AI dollars and AI budgets, so to anoint you an AI winner, let's say, would it be as a manifestation of your security business outperforming expectations? That's actually where you would be capturing a lot of this value from AI? Because it doesn't seem like geographically, it would land on the delivery side.

Yes. Not -- I don't think it's going to change our delivery business anytime soon. It's helping our security business, which obviously is important. But really, it's our CIS, our cloud business. We're talking next year that revenue overall for the company going to get into the low-teens, which is a big acceleration. That's being driven by the cloud business, the compute business.

And then just to kind of round out on the delivery side, another sort of monumental change in the delivery market is it's a much more rational competitive or competitor and peer landscape, right? You yourselves have been in a position to consolidate some of these assets from these companies that are no longer around, right? So can you give us a state of the union on -- in the classic delivery realm, how much of that has influenced the stabilization you've seen in the business, right, your ability to capture more traffic growth in the classic sense with the live events and the software downloads.

And then relatedly, the influence that's had on pricing because I think there is generally a view that this business has had to fight or circumnavigate commoditization effects and price deflation, right? So I'm wondering if you can just characterize the health of the core delivery business from the price side of the equation -- and we've discussed the volume, but the volume side of the equation, especially in light of the fact that the competitor and peer landscape has changed so dramatically.

Yes, delivery is a good business, really important for our customers. We're the leader at that by far. Pricing is still very competitive, and we walk away from deals where we don't like the price and also spiky stuff that we don't like. We're not getting paid enough. We won't do it. We do have competitors that will go out there at any price to win business still, even though at least several of them went broke in the last few years doing that. There's still others that do that.

And okay, we don't operate the delivery business to go get all the traffic or to drive revenue growth. We operate it strategically and to drive cash flow. And so in some cases, we'll say, you got to pay us more or we're not going to do it. And in some cases, we won't take it. For most of the major players out there, we'll do the majority of their delivery, but we really do care about the price there.

I want to shift gears into the security business. I would say, Guardicore on the microsegmentation side and API security have just been all-stars in that portfolio. Wondering if you can help bifurcate kind of the health of the segment by way of some of these newer kids on the block, so to speak, and some of your more mature growth businesses where you have a tremendous amount of market share on the DDoS and WAF side. If you can kind of walk us through some of the dynamics because we talked earlier about, hey, security is going to be the landing zone for you to win AI wallet share, right, for the most part. So is there -- are you starting to see maybe a little bit of a renaissance on some of the older products in the security portfolio? What has been coming up in your conversations?

Yes. The majority of the security revenue is the more mature products where we've got huge customer penetration, WAF and DDoS. And we are seeing some tailwinds there. As I mentioned, in the post-Mythos world world, the bot networks are 10x as big, and that's helped our DDoS business. And with WAF, a lot like the world's major banks, emergency programs to make sure all their apps are behind Akamai's WAF. So they'll be protected from the zero days. And Bot and Agent Manager getting tailwinds from the fact there's all these agents out there. So no question, the security business as a whole is benefiting from these trends.

That said, I think the big excitement from AI in terms of driving revenue growth at Akamai is on the cloud side. Like we talked, next year, we're going to have dramatically accelerated revenue growth going forward, and it will be faster than the security business even is growing. So security is great, where it is today. But cloud is a much bigger market, and we have a compelling value proposition, and we're seeing incredible growth there, enough to take the whole company into a different ZIP code with growth.

Good segue into talking about CIS and compute. Initially, there were aspirations to compete with the hyperscalers, but now the opportunity has become more multifaceted, right? So can you help walk us through sort of the vision for the compute hyperscaler, which was kind of the first step or the first kind of foray in scaling the cloud business by way of Linode and kind of your rationale to acquire Linode, your aspirations to be a specialized edge cloud and then also your aspirations to be a quasi-neo-cloud as it relates to being a destination for model training and distillation. I'm wondering if you can sort of go a layer deeper on where you think and feel and believe your most killer advantage is going to be between those opportunity sets inside CIS.

Yes, great. I'd say we're the first 2, very relevant, competing with the neo-clouds, not so much. Most of the neo-cloud revenue is training the big foundation models, gigawatt data centers. That's not us, not where we're going. We are doing the end-to-end AI for our big enterprise customers, training all the way through inference. And now today, a big data center for us is tens of megawatts. But that's fine for training enterprise-grade models. And of course, we already have the edge platform where we now can deploy containers, function as a service, so you get really close to end users, all on one platform connected by one of the world's largest backbones.

In fact, all the hyperscalers are using our compute service today. Obviously, we compete with AWS, Azure and Google, but their parent companies are using for applications where latency really matters, using our cloud to do that. And 2 of the 3 are very large Akamai customers. So yes, we compete with the hyperscalers. We have for a long time. Now we compete also on cloud where we didn't before. We don't compete so much with the neo-clouds. I think one of the big deals we did, they were in there competing. But generally, we're not head-to-head with them like we are with the hyperscalers.

I think what's captured everyone's excitement for the business and the opportunity is just the edge inferencing use cases. Those are absolutely lighting up. I wanted to ask you at a very technical level and feel free to go in the technical nuts and bolts, professor -- the edge latency and the use cases, you've built over a $2.8 billion book of business in signing some very marquee transactions that absolutely validate what you're doing. But can you help in plain terms, express why you've been winning these very watershed deals for you and why investors should be paying attention to the use cases here?

Yes. The edge platform enables us to get the compute logic closer to the user or the data or the bot or the robot or the sensor. And that's important for applications where the latency matters, driving a car or a robot that's doing things in a factory. You're talking to the agent. You're getting a personalized video from a commerce site showing you're wearing some clothes, you're talking to your personal shopper. And before, AI wasn't to the point where the AI could generate something like that in real time. Now it is.

And so now the last barrier is, okay, if the AI is running across the company or across the world, it's still going to be too slow just to be able to have that conversation because of the latency. Or if you've got a sensor kind of a robot kind of application and you got to process a lot of videos, you don't have the bandwidth in the central location to do that. You've got to -- it's just like you can't broadcast the World Cup game from a data center. You don't have the bandwidth out, and it's the same problem in reverse.

Or if you're doing video applications for commerce sites that are personalized at scale. You're not watching the World Cup. You're watching yourself wearing the sweater. You can't do that at scale if it's all coming from the central data center. You just can't get it out. Same problem we solved 20-plus years ago with delivery. You need to be distributed. And so that's what we do. And I think now is really the time where that's just starting to see and a big runway ahead. And of course, we do it end-to-end. You don't have to do your training somewhere else and then use us. It all works on the same platform connected by one of the world's top 10 backbones.

Dr. Tom, what about the data center architecture and your point of presence footprint is extremely advantageous in this environment. And a lot of peers talk about maybe having a different network topology, some kind of differences in opinion on, again, network architecture. But you've got the 4,000 POPs -- you have been very assertive on the capital expenditure and been very front-footed about that to support these growth opportunities. But can you help investors understand what about your distributed footprint is very advantageous and where you are just actually rolling out racks of GPUs to be able to immediately support this demand in your compute and CIS pipeline?

Yes. There's a bunch of reasons, but we probably deal with all the world's data center companies, probably more than anybody. We're in 700 cities, and you said, 4,000 POPs. And so we have good relationships there. We have good relationships with all the hardware vendors. We've been buying from them for a long time. And so there's -- it's a tight market for both, but we're in a great position to get what we need in both cases in a reasonable time frame.

And then you've got the platform where we have large -- for us, large, small compared to the hyper clouds, but plenty big for us in hundreds of cities around the world and then smaller locations in hundreds more in a platform connected by one of the world's largest backbones. So we can communicate very efficiently and quickly. It allows us to have a lot lower egress costs than the hyperscalers charge. So for a lot of applications, we're head-to-head with the hyperscalers competing for a deal will be less expensive because of that.

And also on top of this, we've got the world's biggest and best delivery network and a lot of the applications ultimately need delivery and security. And that's going to be super important for these agents in AI that you actually have really good security. The hyperscalers, most of them use us for security. Even though we compete with them, they do have security offerings in their clouds, but they'll use us for ours because they're stronger.

I wanted to spend a little bit of time on -- you are spending a lot of effort, energy and executive focus on scaling up what is structurally a very attractive franchise in your compute business, right? It is going to be the power source for the acceleration that you're anticipating in total revenue growth next year. But so far, you've been in the mode of winning the beachheads with the very large frontier lab customer that you sort of referenced earlier in the year as a big win, the $1.8 billion deal.

Subsequent to that, there's been a couple of $100 million deals, excuse me, that you've bagged. So these are very consequential wins. But at the same time, they have come with increased CapEx intensity, right? So how can you help investors get comfortable with the medium- and long-term unit economics of this business because you are in investment mode, right?

So we are seeing gross profit and gross margin compression as you tackle these opportunities. We are seeing operating margin compression as you chase these and execute on these opportunities. But how should investors get comfortable around deals from here against the CapEx footprint that you have and that you will continue to build to support this demand are going to come at better unit economics for Akamai at large?

Yes. The economics on these deals are pretty good. We've talked about the very biggest ones, which would have the, I guess, the least good economics, but still very good. Gross margins, mid-60s to mid-70s -- operating margins, low to mid-20s to low-30s, very profitable deals for us. In terms of the CapEx investment, we've talked about $1 of CapEx generating $0.50 to $1 a year in revenue. The biggest deals down at the lower end, other deals up at the high end or beyond. These particular deals are 4 to 7 years. The biggest deal was a 7-year commitment. And it's not that we're upgrading the hardware during those 7 years. It's those servers for 7 years. These are great deals for us financially.

Now where you see it and you could say, that's going to hurt your cash flow upfront. Yes, you put the cash in upfront. And then 6 to 9 months later, well, now you're generating a ton of cash going forward. And as you do a lot of these deals to jump-start the business, well, you notice it with your -- hey, your cash flow is much worse. Yes, but the generation of cash going forward is huge, really good. And yes, when you do these deals upfront, it hits your margins because you're locking down and paying for the data center space, take another 6 months to get the revenue for it.

And yes, we've said margins, we're taking down about 5 points to operate in the mid-20s, but we're going to become a growth company now, okay? And we were not that before. Growth is not going to be in the mid-single digits anymore. We're talking about much, much bigger growth for the company going forward. And so I think a very attractive proposition for investors. And yes, we start with some big deals, and we're going to try to sign more big deals like that and then take that through our large enterprise customer base. A typical customer today will spend 10x doing cloud than what they would spend in security or delivery with us. And that's a great value proposition.

And you have been very disciplined about managing the cost structure in the business, very rigorous. So is there an opportunity to continue to drive efficiencies in other parts of the business at large to essentially mitigate some of the dilution impacts as you pursue growth and accelerate growth by way of compute?

Yes, we're always looking to be more efficient. AI is actually a little helpful there. So there's positions we aren't backfilling or not hiring into because we are going to be gaining more and more efficiencies going forward with AI. Our software base, we're always working on to make it be more efficient so that we get more bytes delivered or more computations done per kilowatt hour or square foot of colo space. That's always an ongoing effort.

One of the other sort of elements that's unfurled since the last earnings call is, hey, you're really getting good at deploying the CapEx and building more capacity for, let's just call them, bluebird inbound demand, right? So correct me if that characterization is not quite precise. But when you think about the pipeline for the compute business and the pipeline for the CIS business, we've clearly seen you bring down the whale size deals, a.k.a., the $1.8 billion, but we've also seen you do the $250 million deal that you did in 4Q of last year, and then you just did the $600 million.

So it's quite a wide range and a wide spectrum. What does that normal distribution look like for that 66% of deals kind of in the middle? What's that sweet spot and zone of deal size? And do you believe that the CapEx that's been deployed thus far and that's on target to be deployed can support without incremental push, better revenue leverage, let's say, on that deployed CapEx?

Yes. Good question. Before the big deals, it was all small deals coming out of the Linode business that became our cloud business. When the $1 billion whales jump into the pool, they take up a lot of it.

And all the water.

Well, you noticed them. They become a big part of it. And so now we're very quickly transitioning to where a large fraction of the revenue is going to be large customer deals. And I think you'll see that for the next couple of years. We're signing up the full spectrum as we go. It's just when you have very large ones, if you measure by revenue, they're going to be a big part of it. And I think over the next couple of years, it will be these large tranche deals that grab the headlines. And then there's a part that's going to be, I think, growing and become important in the long term, which isn't reserving GPUs of a certain type in a certain city, it's a model as a service, which we're beginning to do.

And the customer doesn't even specify what the hardware is. We take care of that. And we're developing the capability to do these things in a serverless way. So they're not even saying, I want 10 models in this city. It just -- it spins up in on demand. And I think that's where the future goes. And it's a lot more customers, not all, obviously, it's a $1 billion a year level or even $100 million a year, probably more in the tens of millions a year that are signed and brought on to the platform. But for the next year or 2, the spectacular growth that I think we're looking forward to will be large customer deals that you notice.

Dr. Tom, just in terms of CapEx utilization, how do we get comfortable that you're not going to be on this CapEx treadmill, so to speak, for the next 3 years, right? And is there a distinction you can draw between kind of latent CPU capacity to power these edge AI and inferencing use cases? And is that something you believe maybe is not well understood in the investor community because maybe the sense is, hey, you need the Ferrari GPUs for every AI and edge use case, right? And is that what you're seeing in the trenches with...

Yes, we don't even buy the Ferrari, right? The neo-clouds buy the Ferrari and they need them to train the next amazing frontier model. That's not our business. We are buying at least a stage down GPUs that are really good at what our customer base wants.

The Toyota or the Lexus.

Yes, maybe some Toyotas and Lexus and a bunch of CPUs, too. In fact, before these big deals, today, the majority of inference and AI on our platform is done on CPU. And there's the next-generation CPUs coming out and AI is going to get a lot more efficient. And so I think a good portion of it is going to be done on CPU. Now we'll have CapEx to drive the CIS business. It's that kind of a business. It's not like security, where it's you're the world's best developers developing the latest product and not much CapEx. This is not so much that as the CapEx for CIS.

And we will have that, but you will see, I think, very strong growth on the top and bottom lines. It's not like margins keep deteriorating. I think margins, we've said is best -- we haven't given guidance for next year, but ballpark where we are today, but big increase in revenue. And as you get a big increase in revenue growth and your margins are pretty consistent, you see that now on the bottom line, good increases in your bottom line growth, which is pretty exciting.

And that brings me to capital allocation. You have been a voracious buyer of your own stock, but we have put a pin in the share buyback program. How is the philosophy around capital allocation going to change and give us sort of the current state of where you are, what you're focused on?

Yes. So we said we paused for this year for now. It doesn't mean we wouldn't do something, but we paused for now because we already bought back a lot more than we ever have in 1 year this year. So we bought back all the dilution we incurred with equity programs for our employees and then some. And so it seemed like, hey, we don't need to do more than that this year, especially when we've got these really compelling opportunities in terms of the CapEx investments.

Last question I have for you. What's one thing that with a magic wand, you could wish away in terms of or educating or disabusing the investor community on whether they're underappreciating something or something that you believe is very severely misunderstood about the Akamai story?

I think we're going through an interesting transition where we're going to have some pretty exciting growth. We've got a great security business doing very well. Delivery, okay, not exciting, obviously, in terms of the growth, but a very, very good business. And now we've got just CIS is exploding, and you're going to see some real excitement there.

Great. Well, thank you so much, Dr. Tom. This was a fantastic conversation. I appreciate it.

Very good. Thank you.

Good day, everyone, and welcome to the Q2 2026 Akamai Technologies, Inc. Earnings Conference Call. [Operator Instructions] Please also note today's event is being recorded.

At this time, I'd like to turn the floor over to Mark Stoutenberg, Head of IR. Sir, please go ahead.

Good afternoon, everyone, and thank you for joining Akamai's Second Quarter 2026 Earnings Call. Speaking today will be Tom Leighton, Akamai's Chief Executive Officer; and Ed McGowan, Akamai's Chief Financial Officer. Please note that today's comments include forward-looking statements that include revenue and earnings guidance. These forward-looking statements are based on current expectations and assumptions that are subject to certain risks and uncertainties and involve a number of factors that could cause actual results to differ materially from those expressed or implied.

The factors include, but are not limited to, any impact from macroeconomic trends. The integration of any acquisition, geopolitical developments and other risk factors identified with our filings with the SEC. The statements included on today's call represent the company's views on August 6, 2026, and we assume no obligation to update any forward-looking statements. As a reminder, we will be referring to certain non-GAAP financial metrics during today's call. A detailed GAAP to non-GAAP reconciliation is available in the Investor Relations section of akamai.com under financials.

With that, I'll now hand the call off to our CEO, Dr. Tom Leighton.

Thanks, Mark. I'm very pleased to report that Akamai continues to build momentum and gain wide industry recognition as a key infrastructure provider for the AI-driven economy. Akamai has long been known for operating the world's most distributed platform for content delivery and cybersecurity at global scale and with a reputation for reliability, quality and trust. And now we're leveraging our global footprint and years of experience serving the world's largest enterprises to do for the AI-driven economy, what we've done for cybersecurity and content delivery.

The response from industry leaders and major enterprises has been very strong. In fact, we were very pleased to announce today that a U.S.-based technology company has committed more than $600 million over 4 years for our cloud infrastructure services to power their robotics development. The addition of this new customer brings the total volume of the multiyear commitment signed so far this year for our cloud infrastructure services to more than $2.8 billion.

As a result of these commitments and the exceptionally strong pipeline we're seeing, we now anticipate that Akamai's overall revenue growth will accelerate into the low teens in 2027. For investors who want to understand how the AI market is evolving and why Akamai is such a unique and vital player in this new ecosystem, we encourage you to read the July 23 article in Fast Company by Victor Dey.

In the Oracle, Day explains the role played by Akamai's distributed platform in transforming the content delivery and cybersecurity marketplaces and how a similar approach can benefit the AI ecosystem with many agentic workloads being processed at the edge, close to users instead of in massive centralized data centers. In a related blog, IDC analyst, Dave McCarthy, considers a world transformed by AI agents and the practical challenges of automated execution of queries against a trained model whereas training frontier models requires massive data centers and energy consumption. The next challenge for AI is what it will take to run those models everywhere at low latency and with affordable cost.

It's also helpful to read a recent blog by Akamai's CTO, Bobby Blumofe on how AI inference is reshaping the cloud. Bobby's Post explains why the infrastructure needed for AI agents will ultimately be a flexible continuum, stretching from the core to the edge. As the market for AI moves beyond centralized AI factories, Akamai's strategy is to provide a unified distributed grid for AI inference. By pushing AI inference to the edge in combining it with our massive deployment of CPUs for delivery, security and Functions as a Service, our platform will enable customers to run agents and models within milliseconds of their end users with the responsiveness of local compute and the scale of the global web optimizing performance while reducing latency and cost.

Customers are already leveraging our cloud infrastructure services for a wide variety of repeatable use cases. For example, an AI patent intelligence platform in Singapore shifted its inference workloads from a hyperscaler to Akamai, boosting performance by 30% and cutting infrastructure costs by 20%. An AI-powered decision intelligence platform in Poland chose us to run low latency game theory simulations globally and with predictable costs.

An enterprise data storage provider in the U.S. chose us to run their high-throughput storage observability and analytics platform. A global AI software company in India chose us to run their GenAI image creation and a $12 million win over a hyperscaler. An AI-powered communications provider in India chose us to eliminate unpredictable hyperscaler billing and accelerate their customer engagement. And a SaaS media workflow platform in the U.S. chose us for high throughput, live media and coding.

Looking across our business as a whole, AI is not only driving adoption and revenue growth for Akamai Cloud, AI has also been a tailwind for our security solutions. In Q2, Akamai Security revenue grew 10% year-over-year as reported and 9% in constant currency. Security growth was led once again by strong demand for our market-leading web app firewall, API security and Garda core segmentation solutions. Our WAF continued to see strong demand in Q2 from customers eager to protect against vulnerabilities that could be exposed by frontier models like Mythos or GPT 5.5.

Customers that added or significantly expanded their WAF usage in Q2 included one of the world's leading commerce sites, one of the world's leading automakers and one of the world's largest banks in a $14 million upgrade to have Akamai secure all of their applications. We also signed renewal upgrades for security and CIS products with one of the world's largest telcos, in a contract worth more than $20 million over 2 years.

And today, we're also very pleased to announce that CrowdStrike, a leading AI native cybersecurity platform has switched to Akamai for its web security and content delivery needs. CrowdStrike told us they were dissatisfied with the inconsistent service they received from one of our SMB-focused competitors. We see this validation from another security leader as a strong endorsement of Akamai's enterprise security capabilities and our hard one reputation for trust, reliability and dedicated customer support.

On the go-to-market front, we recently announced that Akamai has been selected as a strategic partner for WWT's AI readiness model for operational resilience or ARMOR for short. This is the industry's first holistic vendor-agnostic AI security framework and our inclusion positions Akamai as a foundational security architecture for the AI factories being built by WWT accelerated by NVIDIA. Without this collaborative security framework for AI, organizations are often forced to piece together fragmented security strategies.

By aligning the Akamai security portfolio with ARMOR we provide a methodology to protect large-scale AI clusters proactively by preventing the lateral movement of threats. As WWTs Chris Conrad said, no single vendor can secure the AI frontier alone. Through our close partnership with Akamai, we're turning the hype of secure enterprise AI into a tangible scalable reality for customers. Akamai is also proud to be one of the industries must have security providers partnering with Anthropic and OpenAI to help ensure the safe and rapid deployment of AI-enhanced defenses.

With our access to programs like Glasswing and Daybreak and our participation in programs like the open secure AI Alliance, we're applying our expertise to help keep major enterprises and critical infrastructure secure. As employees across all industries use AI tools more frequently, enterprises need more help to secure their workforce and prevent sensitive data from being leaked. That's why we acquired LayerX, a leading provider of secure enterprise browser and AI usage control. With LayerX, now rebranded as Akamai Workforce Protector we're able to give security teams the deep visibility they need to see exactly how users are interacting with web content, SaaS applications, file uploads and even raw AI prompts.

By adding this browser security platform to our portfolio, Akamai is now uniquely positioned to protect enterprises from unauthorized AI agents, whether they're operating on a user's laptop or living inside internal applications that talk to external LLMs. What makes the combination particularly exciting is how nicely Workforce Protector aligns with Akamai's existing Zero Trust portfolio. including our Guardicore micro-segmentation, Zero Trust Network Access and DNS security solutions that thousands of enterprises rely on today.

Ultimately, we believe these combined capabilities will deliver a powerful unified workforce security solution that directly addresses one of the industry's most urgent challenges securing and governing how employees, partners and supply chain ecosystems interact with AI. LayerX is the latest in a series of acquisitions we've made to build out our security portfolio and make it easier for enterprises to obtain more comprehensive solutions from Akamai as their strategic security partner.

Overall, we expect our security portfolio to generate more than $2.4 billion in revenue this year, making Akamai one of the largest security providers in the market. In summary, we're excited to see AI driving demand for our cloud and security portfolios, and we're grateful for our talented team here at Akamai, who continued to perform extraordinary work to deliver flawless digital experiences for our customers. Our employees' dedication to our customers' success is one of the reasons why so many of the world's top brands and most demanding businesses rely on Akamai as their trusted, dependable and reliable partner.

Now I'll turn the call over to Ed for more on our results and our outlook for the remainder of the year. Ed?

Thanks, Tom. Before I begin, the standard review of our quarterly financials and updated 2026 outlook, I want to build on Tom's remarks. From a strategic and financial perspective, we couldn't be more excited about the trajectory of our business. Securing a 4-year $600 million GPU services deal with a leading U.S.-based technology company focused on robotics development brings our recently announced major wins to over $2.8 billion this year, a huge validation of our platform.

Beyond providing great multiyear revenue visibility, it also underscores the scale of our cloud infrastructure services business. For this new customer, we don't expect any material revenue impact for 2026, but we do expect revenue to fully ramp throughout 2027. And finally, as Tom mentioned earlier, and backed by these multiyear commitments, along with an expanding pipeline, we have a clear line of sight to accelerating our top line revenue growth from single digits this year to the low teens in 2027.

With that, let's dive into the Q2 results. Revenue in the second quarter was $1.1 billion, of 5% year-over-year as reported and in constant currency. Cloud Infrastructure Services or CIS revenue was $99 million, up 39% year-over-year as reported and in constant currency. This was in line with our expectations, and we continue to expect CIS revenue to meaningfully accelerate in Q4 and then further accelerate in 2027. Security revenue maintained strong momentum with revenue of $604 million, up 10% year-over-year as reported and 9% in constant currency.

The strength in the second quarter continued to be driven by our fast-growing API security and Guardicore Segmentation solutions. Moving to delivery and other cloud applications. Revenue was $396 million, down 6% year-over-year as reported and down 5% in constant currency. International revenue was $549 million, up 6% year-over-year or up 7% in constant currency representing approximately 50% of total revenue in Q2. U.S. foreign exchange fluctuations had a negative impact on revenue of $2 million on a sequential basis and a negative $1 million impact on a year-over-year basis.

Moving to profitability. In Q2, we generated non-GAAP net income of $236 million or $1.59 of earnings per diluted share, down 8% year-over-year as reported and down 6% in constant currency. These results include our expanded colocation investments, higher depreciation and increased headcount costs all to help fuel faster growth in our cloud infrastructure services. Our non-GAAP operating margin for Q2 was 25%, in line with our expectations. We expect operating margin to remain in the mid-20s for the remainder of this year as we continue to invest to capture the exciting growth opportunities in CIS.

Our Q2 CapEx was $347 million or 32% of revenue. Second quarter CapEx was below our guidance, primarily driven by the timing of receipt of some GPUs. These shipments arrived a few weeks later than expected following the quarter end, pushing the bulk of the planned spend into Q3.

Moving to cash in our capital allocation strategy. In May, we raised $3.5 billion via 2 equal tranches of 0 coupon convertible debt maturing in 2030 and 2032, to fund our growing CIS pipeline and for general corporate purposes. In Q2, we spent approximately $410 million to buy back approximately 3 million shares. Year-to-date, we have repurchased roughly 5 million shares for a total of approximately $616 million. We ended the second quarter with approximately $565 million remaining on our current repurchase authorization. However, given the strong market demand for CIS, we are temporarily pausing share repurchases to reallocate capital to support our high-growth CIS pipeline.

And finally, as of June 30, we had approximately $4.6 billion of cash, cash equivalents and marketable securities. Now before I provide Q3 and full year 2026 guidance, I want to touch on a few housekeeping items. First, on July 2, we completed the acquisition of Security Company LayerX for approximately [ $205 ] million. From a financial perspective, we expect the acquisition to have no material impact on full year 2026 revenue. On the bottom line, we expect LayerX will be dilutive to our non-GAAP EPS by approximately $0.12 for 2026, split evenly across Q3 and Q4.

Second, I'd like to turn to foreign exchange and its expected impact on our second half 2026 performance. Based on currency movements since our last earnings call, we now expect second half revenue headwinds of approximately $9 million. Finally, I'd like to provide additional color on our capital expenditures for the remainder of the year. For Q3, CapEx is projected to step up significantly, partly due to the shift that slipped into July that I referenced earlier and as we ramp up build-outs to support the major contracts announced earlier this year.

As mentioned on our last call, GPU demand remains exceptionally strong. As a result, all of our GPU capacity is completely sold up. Therefore, driven by the $600 million new customer win we announced today and a very robust pipeline, we expect to invest up to $500 million in CapEx to replenish and expand our GPU capacity. We expect to spend approximately $60 million of that this year with the remainder hitting in early 2027.

Moving now to guidance. For the third quarter, we are projecting revenue in the range of [ $1.105 billion ] to $1.13 billion, up 5% to 7% as reported and up 5% to 8% in constant currency over Q3 2025. At current spot rates, foreign exchange fluctuations are expected to have a negative $2 million impact on Q3 revenue compared to Q2 levels and a negative $8 million impact year-over-year. At these revenue levels, we expect cash gross margins of approximately 70%.

As a reminder, gross margin is impacted by the significant increase in colocation as we accelerate the growth in our CIS business. Q3 non-GAAP operating expenses are projected to be $347 million to $359 million. We anticipate Q3 EBITDA margin of approximately 38% to 40%. We expect non-GAAP depreciation expense to be $153 million to $155 million. We expect non-GAAP operating margin of approximately 24% to 26%. And with the overall revenue and spend configuration I just outlined, we expect Q3 non-GAAP EPS in the range of $1.60 to $1.80. This EPS guidance assumes taxes of $55 million to $62 million based on an estimated quarterly non-GAAP tax rate of approximately 19%, and it also reflects a fully diluted share count of approximately 150 million shares.

Moving to CapEx for the reasons I highlighted earlier. We expect to spend approximately $475 million to $525 million in the third quarter. This represents approximately 43% to 46% of total revenue. Looking ahead to the full year 2026, we expect revenue of $4.445 billion to $4.53 billion, which is up 6% to 8% as reported and up 5% to 7% in constant currency. For cloud infrastructure services, we continue to expect year-over-year revenue growth of at least 50% in constant currency.

We continue to expect security revenue growth in the single digits on a constant currency basis in 2026 and for delivery in other cloud apps continue to expect a decline in the mid-single digits year-over-year on a constant currency basis. At current spot rates, our guidance assumes foreign exchange will have a positive $9 million impact on revenue in 2026 on a year-over-year basis. Given the recent strength in the U.S. dollar, this impact is significantly less than the positive $20 million we discussed on our last quarter's earnings call.

Moving to operating margins. For 2026, we are estimating a non-GAAP operating margin of approximately 25% to 26% as measured in today's FX rates. Turning to CapEx. At this time, we anticipate our full year capital expenditures will be approximately 40% of total revenue. Moving to EPS. For the full year 2026, we expect non-GAAP earnings per diluted share in the range of $6.40 to $7.05. This non-GAAP earnings guidance is based on non-GAAP effective tax rate of approximately 19% and a fully diluted share count of approximately 150 million shares.

Before wrapping things up, I want to provide additional color on how we safeguard profitability and manage risk across our cloud infrastructure services business, specifically as it relates to our recently announced large contracts. While initial capital deployment happens upfront, these contracts are structured to deliver strong cash flows over their life backed by take or pay commitments.

To give you visibility into how we manage these opportunities, I'd like to outline the economics that are typical in these multi-hundred million dollar and multiyear contracts we have seen to date and expect in the future. While I won't speak to any one contract in particular, the following is meant to illustrate large deals that have been signed and what we see in our current pipeline. For these multi-megawatt large-scale deployments, we take several factors into consideration, including customer credit quality, contract duration, the specific compute architecture deployed, i.e., GPU or CPU and data center space, power availability, CapEx and direct operating expenses such as networking, power along with any software and hardware maintenance.

Taking all these factors into account, our signed deals and active pipeline consistently reflect a highly attractive profile, delivering non-GAAP cash gross margins spanning from the mid-60s up to the mid-70s. And after factoring in hardware depreciation and other operating expenses, these large-scale contracts typically generate non-GAAP operating margins ranging from the low to mid-20s up to the low 30s.

So with that, I'll wrap things up. And Tom and I are happy to take your questions. Operator?

[Operator Instructions] Our first question today comes from Sumeet Chatterjee from JPMorgan.

Congrats on the new wind customer win here. Maybe if you can just help us think about the -- what you're seeing in terms of more appetite from customers for the CIS business in terms of pipeline, you have announced these sort of significant deals, but you're also sold out as you're indicating. So does that sort of preclude you from any significant more sort of award where our customers in the near future as well. So any visibility on both those fronts would be helpful.

Yes, very strong pipeline really across the spectrum of industries and scale of the deals. And no, that doesn't keep us from signing up more customers because we're ordering obviously more hardware. And typically, we're looking at about a 6- to 9-month window on the larger side. So no, we're continuing to sign customers up. Ed, do you have more color on that?

Yes. I was just going to add. So the way to think about the commentary we sort of gave you a little preview last quarter that there was a good chance we might be placing an order for more GPUs based on what we saw from the pipeline, and we were able to sell that all out. Some of that large deal will take up some of that remaining inventory. And we're placing another order with NVIDIA for significant chunk of additional GPUs, some to satisfy the new order, but a lot of that is covered by what we had before.

And then the rest of it is based on the strong pipeline we see for additional GPU demand. Now in terms of the way the market is going at this point, customers are essentially preordering in a lot of cases where a lot of folks don't have a ton of inventory on hand. So usually, the conversations are about getting some GPU capacity generally, customers want to lock that in for a long period of time with a -- think of it as like a reserve instance is a good analogy to what folks do in the cloud market, and we'll generally reserve that months in advance. So it actually enables us to maintain a good inventory, if you will, and not extend ourselves too far out and meet the market demands.

Our next question comes from Jackson Ader from KeyBanc Capital Markets.

The first one I had was actually about the timing between deal signage for these multiyear deals and -- the multi-million multiyear deals and when you actually expect them to fold into revenue. Tom, you said something about like a 6- to 9-month window. And I was curious, are you talking about from deal signage to actually seeing the revenue? Or was that window referring to something else?

Yes, I'll take that one, Tom, for you. So it's usually between signing and when we recognize revenue. For the really large deals, we generally won't keep the type of inventory, like, for example, the last deal we mentioned last quarter, we're not buying that much sort of speculatively, so that will be informed by a large customer coming in ordering. And generally speaking there, you may need to get some additional data center space.

We have a nice pipeline of data center space that's coming online between now through the end of next year, and it's sort of a core competency of ours anyway. So it's sort of a normal motion for us. But when we talk about the delta between signage and revenue recognition, that's usually 6 to 9 months typically with these large deals. And to the extent that they are large enough, we will call them out on the call, like we did over the last 3 calls, actually and give you some indication of when we think these contracts will start generating revenue.

Okay. All right. That makes sense. And then a quick follow-up. Ed, it was really helpful to kind of walk through like an illustrative example of the margin profile of these deals. But as that -- does that remain pretty consistent as the component pricing and your own pricing on GPUs has fluctuated. And is there any risk that some of the margin profile were to structurally shift lower if there is more upward pressure on components in the future?

Great question. Let me use a term in terms of the pricing environment with -- especially with the big customers, I'd say it's a bit more collaborative. And what I mean by that is there's a general understanding that inventory is tight, availability of data centers is tight. And customers understand that, a, if they don't order when they say they're going to order, things may not be there and prices may go up.

So prices are moving relatively quickly in the marketplace, and we are able to pass on any sort of increase in pricing. And generally, we'll have that structured into a large contract. So for example, let's say, I'm receiving goods over a period of 6 to 9 months or something like that, if there are movements in pricing, whether it's memory or hardware costs or things like that, we do have mechanisms to take that into consideration. And we're constantly adjusting pricing.

So for example, somebody may do a follow-on order where they bought 6 months ago, and now they're buying today, they're buying the next set of inventory, and that's at a higher price. We just marked that price right up. But it is very collaborative with customers, and we are able to get price increases passed along in CIS.

Our next question comes from John DiFucci from Guggenheim Securities.

Nice job, guys. I mean it really -- things look really impressive here. And Ed, I really appreciate all the detail you gave on some of the characteristics, the profitability characteristics of these deals. I mean, the size of them and the revenue that's going to come is pretty impressive, but so is the profit. But can I ask a little bit about -- you gave some like over the life of these contracts. But just in from other companies that do similar deals.

It seems like at first, you're going to see probably less profit coming from these things. You're going to have to start spending and incurring expenses even before revenue, I would assume, and then the ramp up over time. So over the life of a longer-term deal, let's say, a year deal or whatever it is. When is it that you achieve that profit? Because I just don't want people to be surprised at first, you're seeing more pressure on the profit line until you get to where you need to be?

Yes, great question, John. And the answer is it depends a bit, and it's also a dynamic you have in the business itself, right? So -- in many cases, you will take on data center space ahead of time. So I'll talk in terms of the big deals, where you might open up, say, a half dozen data centers or whatever it may be with a particular customer. And the equipment is being received over a several month period or whatnot.

So you'll have definitely co-location costs ahead of revenue depending on the way the contract is structured, but generally speaking, when the equipment is ordered and we start taking depreciation, it's a relatively short period of time between when you rack and stack and get the equipment up and running to get revenue.

The other thing to keep in mind, too, is the accounting rules for both revenue and for colocation or leases really, if you have escalators and things like that, you generally will spread that out over the entire contract. So your costs will be amortized flat, your revenue will be amortized flat. So you do pretty quickly get up to that full profitability. But there may be a quarter where you have a little bit of noise where you have some costs ahead of time.

But generally speaking, there is a pretty quick ramp-up to get to that profitability given what we're seeing so far, both in the pipeline and also with what we signed as far as customers wanting to stand up this capacity and more, like I said, a reserve instance type capacity where you give them a certain amount of up to capacity, and they have the right to use that over the life of the contract. It's just a question of how quickly you can stand it up. So as you're modeling, I would expect to see margins a bit lower as we're ramping into these things and then they should start to ramp up into those levels that I talked about.

Okay. Great. And so when you say a relatively short time, like a quarter maybe, I mean, I know every deal is...

A quarter. I mean on the long end, maybe it's 5 or 6 months, but generally speaking about a quarter.

Okay. Perfect. And if I could, a follow-up to Jackson's question because I think he's going right down where I'm trying to think through this anyway. Are these -- if some of these longer-term deliveries, like you mentioned a deal where you're going to deliver capacity and maybe 4 data centers along the way. That's all coming online over the next year -- I mean it's coming out in the future.

And you mentioned a lot about everybody focused on inflation and component inflation. Are some of these contracts, the ones that are longer term when you have longer-term delivery, are they cost plus contracts were you're going to get a certain margin? And is that margin protected even if inflation takes off like it has over the last year or 2?

Yes. I mean, every deal is a little different. But generally, let's talk about the procurement side first, and we'll get into the the revenue side. So on the procurement side, when we will contract for a data center, we generally try to get as fixed of a term as we can for as long as we can. There is some variability in power. And generally, what you do is you agree to some sort of upfront escalator a couple of percentage points or something like that.

Like I said, on the accounting rules, you end up straight lining that just like you would on rent when you get like free rent or an escalator and rent, that sort of thing. So that generally covers that. So we pretty much know going in what our costs are for power and then our hardware is obviously fixed. Generally, if you have any hardware maintenance, you would negotiate that upfront and have a long-term agreement. So you structure the revenue contract to do the same. So there might be an escalator for each year to cover your increase in labor cost, your cost for your power escalations and things like that.

But generally, you can get a fair amount of it fixed over time. If there are long lead times in terms of, let's say, I'm getting 20% of the capacity stood up in the first quarter, then 30% and then 50% in the last quarter. So it takes 9 months to spin it all up. If there are variations in pricing, like I said, with memory, you will have contemplated that and have a mechanism in the contract to deal with it typically. So this is, like I said, it's very collaborative with the customers. We're having these kind of conversations and making sure that we're protected as best that we can. And generally speaking, so far, it's been working out as we've expected.

Our next question comes from Ram Singh from Oppenheimer.

Firstly, on the compute side, good to see all this visibility and pipeline -- just wanted to understand when you think about available capacity, what do you have in terms of megawatts? What's your pipeline? How much you can expand? And I also wanted to understand, would you need to expand to more sites versus what you have today to support this growth? And then I do have a follow-up.

Yes. We're continuing to expand the platform in more locations, more cities having GPUs and I think you'll continue to see that going forward. And we're in a very good position to get the capacity and data center space that we need. We probably deal with more data center companies than anybody. We have our service today in 700 cities around the world of 130 countries. We've got a great reputation for being a reliable partner with strong financials and established relationships. So we're in a good position to get what we need to take on these very large customers.

And as my follow-up, good to hear the commentary around agentic. How do you think about acceleration in that part of the business especially asset modules such as API security and microsegmentation attached pickup with agentic AI traffic coming into the platform.?

Yes, there's a great synergy there. in a lot of different dimensions. We use AI to make our products a lot more capable. We help our customers identify their shadow AI with API security. Now with LayerX, we help them protect the workforce. So when the workforce is using AI, they're not inadvertently exfiltrating corporate sensitive corporate data and of course, in the compute business, we're enabling the AI on our platform. So we're in a great position to support our customers' use of AI and to secure it at the same time.

Our next question comes from Rishi Jaluria from RBC.

Great to see continued momentum in the CIS business as well as some major customer wins. Two for me. First, look, I think in addition to these large CIS deals that you're talking about, a lot of us have been debating the opportunity for true AI as inferencing to leverage the edge network that you have a long history of and the largest network in the world. Can you talk about some of the opportunities you're seeing there? And -- is there an opportunity for some of these large more central cloud-driven AI deals to start to expand more to Edge Cloud, especially because you're the only one out there that has Edge Cloud and Central Cloud on one platform. Maybe help us understand that, and I've got a quick follow-up.

Yes. I think you said it very well. It's really a continuum and a big advantage that we have is that we can do the core for enterprise customers that want to train their model and we can do the edge for when they're doing the inferencing in situations where it's latency sensitive or bandwidth sensitive. A great example is robotics. For example, the robot has sensors, which could include the equivalent of video, they're seeing the environment around them. And that's high bandwidth to get the video input and ingested.

And then our GPUs that we deployed are actually very well suited to processing that video and then using AI to figure out what's going on and then with the AI to give an instruction back to the robot, what to do about it. And in many of these situations, it's also latency sensitive, not only the bandwidth of the video, but telling the robot what to do quickly, think about the robot being a card, a driverless car or the robots in an environment with humans or it's in a factory where if it does something in an untimely way, you could have a problem.

And so that's a great example where, yes, maybe the training is done to be most efficient in a more centralized fashion. But the inferencing and the usage of it as it's operating Well, there's examples you want to do that really close to where the robot is. And we see that really across the spectrum, and it's a big advantage for us to have it all on one platform.

Now to be clear, we're not in the business of training the foundation models, the giant models that we're not doing. But usage of the models or training medium or smaller models, yes, that works very well on our platform. And the key is to have the right resource in the right place for whatever your agent or model is trying to do. So you optimize cost and you optimize performance.

Wonderful. Very helpful. And then maybe just on the security side right now, especially with LayerX acquisition coming in when it closes. But as we see kind of the proliferation of agents, right, it's -- every brand has to figure out how do they embrace them but also protect themselves, especially from not just nefarious, but even competitive agents. Maybe can you walk us through how we should be thinking about that as potentially becoming an accelerant to the security business and where LayerX can fit in that overall strategy?

Yes. LayerX, think of that as protecting your employees who, for all the right reasons, are using a variety of AI tools and agents and models and so forth. And the danger, of course, is that they're actually leaking sensitive data when they use those tools. And that's what LayerX prevents, identifies and prevents and it enforces the business rules that the security team has set up for an enterprise.

Now the dealing with the agents really goes far beyond that. In fact, we're a leader today with our bot management and agent management solutions because our customers with their sites and applications, more and more agents are coming maybe instead of people. some of the agents are just fine. They're authorized by a user. And of course, that we want to be sure that's the case and then gives them very good service. But there's a lot of agents doing things that aren't authorized. And we help our customers by identifying what's the agent? Is it authorized? What's it doing? And then our customer will tell us what they want us to do in response.

A good example is the LLM search engines. Of course, it's really important today that your brands show up in the next generation of search engines. And to help with that, we identify that, hey, this is a scraper coming from a search engine and we give it different content than the regular app or the site. And we give it content that has all the right keywords and also is structured so that the agent can -- the scraper can process it much more efficiently.

And as a result, the customer gets much better search rankings. And that's just one example of an agent, that happens to be one example of a scraper, but there's just a myriad of different cases, and that's what we help our customers do is identify what it is, is it legitimate and then take the appropriate action for their business.

Our next question comes from Sanjit Singh from Morgan Stanley.

In a year that or less in a year that CIS has been stood up as a visit, you guys have come up with $3 billion in bookings, which is super impressive. It sounds like these trend lines are going to continue, just given your comments on the pipeline. And so I wanted to get a sense from the team about to what extent or what sort of the financing strategies going forward? And to what extent is the company willing to go into either a net debt position or access more equity capital to fund what seems to be a very strong pipeline of future business?

Yes, I'll start and then hand it over to Ed. First, we've been working on this business for years, and it's built on top of a platform that we've been developing for decades. So it's not an overnight phenomenon. But you're right that it's in the -- really, the last year, we've made more of an investment in go-to-market because the CIS and the platform, our cloud platform is now in a position where we can take on major enterprise customers at scale. And so you are seeing very rapid growth. And we're in a good position that we have over $4 billion of cash, and we are going to invest to continue the growth of this platform. And Ed, do you want to give some more details around that?

Yes, sure. As we talked about, we have $4.6 billion of cash on the balance sheet today. We've structured our guests so far to have every 2 years or so, there's a payment of -- or later, if you will, of retirements of debt coming up. But the way these big deals work, even the free cash flow is excellent after you deploy the initial capital. So you dip down a little bit here and then as you start to recover some of this on the other side, it's extremely high free cash flow margins on these deals.

But that said, we have excellent banking partners. We've been so lucky to work with a lot of amazing firms and they give us excellent advice. So far, converts have been very attractive for us and have offered the best economics we work with our board all the time on different strategies and scenarios and look at what makes the most sense for our shareholders, there's still a lot of debt capacity if we need it, if that's the right thing to do. We will do that. So far, we haven't done -- we've done equity-linked debt, but we haven't done any equity offerings.

But we will look at whatever the best opportunity is and what makes the most sense, the lowest cost of capital for us. I think we can grow the business quite a bit just in the model that we're in right now. We're still holding an investment-grade credit rating, which is important and helps us with our colocation providers. But we -- you heard us talk about suspending the buyback temporarily here to use all the available cash we have for growth because we haven't seen a growth opportunity like this in a very, very long time, so to make sure we capture that. But I'm very comfortable with our ability to finance the growth going forward.

Understood. And as my follow-up, in terms of the trajectory of growth for CIS, we've had 2 quarters of sub-4% growth was a little bit less than what we exited last year with. And then you are targeting at least 50% growth for the full year. In terms of how you guys are sort of converting those big contracts and hopefully getting into revenue generation capability. What is the -- how should we think about the ramp Q3 to Q4? Is it more of a sustained growth in CIS and then sort of a big hockey stick? And what gives you the confidence that, that hockey stick will emerge in Q4?

Yes. Good question. So we don't provide quarterly guidance. One thing I will say is we did talk about how we did receive some of our GPUs a little bit later than we expected. They just pushed from Q2 into Q3. And -- so there will be a few less weeks of revenue for deals that were signed that were scheduled to ramp in Q3. So I'm not expecting any kind of a growth acceleration certainly for CIS in Q3.

But in Q4, I do expect to see a big hockey stick of acceleration for a number of reasons. One, we're now able to start getting revenue for the GPUs that we've sold that we've taken into inventory and starting to rack and stack and get revenue for those contracts. The big contracts I've given you guys some guidance on when I thought those would start to produce revenue and they do start in Q4. Right now, as we look at everything is on track, it's not a question of having to execute of getting anything signed. It's just the time that we have in terms of everything being delivered, set up and generating revenue.

So right now, we feel like we're on track with everything and we expect a big hockey stick, and that should continue to ramp pretty significantly into Q1 of next year as well because you get a partial quarter of revenue from some pretty big deals and then you start to get a lot more going into Q1. And I expect that acceleration to continue throughout the year of next year.

Our next question comes from Frank Louthan from Raymond James.

How challenging is it to get the power at the facilities where you're doing CIS? And can you give us an update on how many locations you have today? And how many will you need to have built out for this new business that you just signed?

Yes. Great question. We're really in a very good position to get the data center capacity and power that we need. We've been in this business for an awfully long time. We probably deal with more data center companies than anybody. We have our infrastructure today deployed in over 700 cities in 130 countries.

You've got a great reputation built up over many years for being a reliable business partner for financial strength and having a very strong business. And so we're able to get the capacity that we need. We are continuing to grow our data center footprint in more locations and we are taking on larger locations, more power as we grow. And I would expect to see that continue. But we're in a really good position there.

Yes. So one thing I would add -- one of the advantages we have, Frank, is we're incredibly fetal given that we run such a large backbone that for us to get power, say, for the West Coast, we have many different options of where we can go. And it's very attractive for data center providers where not all power is created equally, even as you go state by state.

So you might be able to go a couple of hundred miles from really expensive state into a cheaper state, deal anchor tenant for somebody who's building out data centers, make a long-term commitment, get guaranteed power with great performance as connected to our backbone, et cetera, and you're able to satisfy whatever the demands are for that particular customer, and get much better economics because the power is cheaper.

So that's one of the other advantages we have is we're very, very flexible with where we can build out. And we're interesting enough to these big data center builders we're not asking for gigawatt facilities. We're talking oftentimes 5 to 10, maybe 15, 20 megawatts, which is a big enough commitment that can really help them scale and get good IRR on their investment as they go and get credit to build these facilities.

Yes. Ed raises a great point. People don't think about it much. But we operate one of the world's largest backbones. And of course, we use it to connect all of our locations. We don't sell connectivity per se, and that makes a big difference when it comes to going into new locations. We can go places very successfully and connected to our platform that others may have a challenge.

Our next question comes from Rudy Kessinger from D.A. Davidson.

Could you guys -- I know you don't really talk a ton about these massive deals and the use cases that you're doing there. But to the extent you're able to maybe just talk about whether it's this robotics deal or some of these other large deals or just the deals in the pipeline. What are the specific metrics, whether it's latency or what have you, that these customers are looking at and saying, Akamai delivers a definitive edge versus the neoclouds and the hyperscalers, therefore, we're going to go this route with Akamai. What are the specific performance advantages that are making you guys choice for some of these large workloads?

Yes. No, that's a great question, and we can't talk about the specific deals, but in general, Akamai would be chosen because we have a great hard-earned reputation for reliability. They can trust us. We have a massively distributed platform that's really unique in the marketplace. As I mentioned, 700 cities in 130 countries. We can get our customers' compute needs close to the users, close to the data that provides better performance, lower latency, better scalability, particularly if you're doing anything to do with video.

We have full stack compute and storage combined with the world's leading cybersecurity solutions, the world's leading delivery platform. And at a really good cost, because we have the world's leading delivery platform and this massively distributed capability, our cost for egress is a lot lower than the competition. And so we can give a compelling value proposition of, hey, great performance, great reliability at a lower cost.

Our next question comes from Fatima Boolani from Citi.

Ed, I wanted to talk to you about the capacity that you're bringing online and you framed it as a reserved instant type model, which gives you a lot of visibility and it gives customers a lot of predictability. But I know you have entertained sort of up rental business within the confines of your CIS business.

So I wanted to get a sense of where you are on that journey. And to the extent you do have excess capacity against which you can have more of a spot pricing orientation to see maybe faster growth in CIS. And to the extent that is relevant to think about over the course of this year? And then I have a follow-up on the security business.

Yes. So, we offer both. And what we're finding though is there is a stronger demand for customers who want to lock in for a longer period of time. And those obviously are better for us in terms of locking in the value over a longer period of time and you got guaranteed ROI and what you're investing in. We do offer both. We even have some customers that will come to us and ask us for different models of whatever you have excess, I'll take it for a period of time and you just give me some amount of heads up if you need to take it back.

So the market is, I'd say, probably more skewed certainly with our customers anyway for those who would like to have certainty versus just going and renting by the hour, but we do offer both. And I'd say most of the growth right now, obviously, given the size of these deals is coming from that more committed model, but we do what we do offer both.

I appreciate that. And just on the security business. You specifically called out kind of the tailwind that you are seeing on the Guardicore Segmentation side, API security, just on the back of the traffic mix from the Internet and by extension, your platform changing. I'm wondering if you can speak to some of your more traditional and maybe more mature product areas like the DDoS and I would say, classic WAF. What are you seeing there?

Is there a rising tide lifting all boats and knots maybe there's a renaissance of spending from a customer standpoint, maybe a reinvigorated focus. I'd love to kind of get a better sense on what some of those traditional and more mature areas that the security product portfolio are doing in contrast to some of your higher growth SKUs and the [indiscernible] .

Great question. Yes, and we have seen real tailwinds from AI and sort of the post Mythos world for our more longer-term security products. I think roughly because of AI, the attackers have assembled much larger bot armies to launch attacks. And we've seen the scale of the attacks grow by maybe a factor of 10 over the last year. And so there's more of a need for our DDoS services. And with Web App Firewall, especially there because now there's going to be a lot more 0 days.

And we see those first, and we get our firewall rules updated to protect our customers before the 0 day becomes it public knowledge. And so that they'll have time to do whatever patching they need to do and they can do that safely. And I can't tell you, I've talked to so many CISOs and CIOs and CEOs over the last few months, and many of them are having emergency projects underway to make sure all of their applications and sites are protected by our web app firewall.

Because they know if it's behind Akamai web app firewall, they're going to be okay as the 0 days inevitably come out. So we are seeing good tailwinds from AI across the board, and there's a chance for Akamai to really help major enterprise customers.

And we have time for one final question, and that comes from Patrick Colville from Scotiabank.

This is Conor on for Patrick. Just was wondering if you could double-click on the CrowdStrike customer win, which is really a positive sign, I think, for the kind of architecture. Can you talk a little bit about when did that deal close and if there's any really partnership dynamics there as well as a customer relationship?

Yes, that closed recently. And it's a real validation point for our security solutions. Obviously, CrowdStrike is a company that cares a lot about security and cares a lot about reliability. And they were -- as stated, they were not happy with their current -- the prior fighter, and they really were attracted to Akamai because of our reliability and the higher level of security capabilities that we offer them. And we do partner with them. And so it's a good relationship that way as well. But I think it's a great validation of Akamai and our security solutions.

And ladies and gentlemen, with that, we'll be concluding today's question-and-answer session as well as today's conference call. We do thank you for participating and joining today. You may now disconnect your lines.

Good morning, and welcome to the Q1 2026 Akamai Technologies, Inc. Earnings Conference Call. [Operator Instructions]. Please note that this event is being recorded. I would now like to turn the conference over to Mark Stoutenberg, thank you, and work to you.

Good afternoon, everyone, and thank you for joining Akamai's First Quarter 2026 Earnings Call. Speaking today will be Tom Leighton, Akamai's Chief Executive Officer; and Ed McGowan, Akamai's Chief Financial Officer. Please note that today's comments include forward-looking statements that include revenue and earnings guidance. These forward-looking statements are based on current expectations and assumptions that are subject to certain risks and uncertainties and involve a number of factors that could cause actual results to differ materially from those expressed or implied. The factors include, but are not limited to, any impact from macroeconomic trends, the integration of any acquisition, geopolitical developments and other risk factors identified in our filings with the SEC.

The statements included on today's call represent the company's views on May 7, 2026 and we assume no obligation to update any forward-looking statements. As a reminder, we will be referring to certain non-GAAP financial metrics during today's call. A detailed GAAP to non-GAAP reconciliation is available in the Investor Relations section of akamai.com under financials. With that, I'll now hand the call off to our CEO, Dr. Tom Leighton.

Thanks, Mark. I'm pleased to report that Akamai is off to a strong start to the year. In just a few months, we've achieved major milestones for our cloud computing strategy, marking a definitive point in the growth and evolution of our business. Akamai has long been known for operating the world's largest distributed platform for delivery and security solutions at global scale and with a reputation for reliability, quality and trust. .

Now we're leveraging our global footprint and years of experience supporting the world's largest enterprises to become an indispensable infrastructure provider for the AI-driven economy. At GTC in March, we unveiled the industry's first global scale implementation of NVIDIA's AI grid, and we announced the rollout of thousands of NVIDIA RTX Pro 6000 GPUs. By integrating NVIDIA AI infrastructure into Akamai's massive distributed platform and by leveraging intelligent workload orchestration across our network we intend to move the market for AI beyond isolated AI factories toward a unified distributed grid for AI inference.

By pushing AI inference to the edge, and combining it with our massive deployment of CPUs for delivery, security and functions as a service, we're enabling customers to run complex models within milliseconds of their end users. With the responsiveness of local compute and the scale of the global web, optimizing performance while reducing latency and cost. Those who attended GCC heard NVIDIA reference Akamai as a vital player in the industry's ecosystem for AI infrastructure, and we've seen very positive market reaction to our rapidly expanding capabilities for a wide spectrum of enterprises.

Today, we're very excited to announce another major milestone for our cloud computing strategy and the evolution of Akamai. The signing of a landmark 7-year $1.8 billion commitment for our cloud infrastructure services by a leading frontier model company. This is the largest customer deal in Akamai history, and it comes on the heels of the $200 million CIS deal we announced in February, with a major U.S. tech company, also at the forefront of the AI revolution. These leaders in AI have chosen Akamai because their AI workloads need to scale, performance and reliability that our cloud platform provides. Many other enterprises have chosen Akamai for similar reasons. For example, since the start of the year, a leading cloud and digital infrastructure provider in Asia chose our VPU to support their low-latency live streaming media service.

An AI company in the U.S. chose our GPU platform to power their voice first solution to optimize business operations. An AI-powered video intelligence platform in India chose our GPU platform to scale video analytics and computer vision workloads for retailers. A consumer AI platform in the U.S. chose Akamai cloud to run and scale live personalized agents. An AI commerce company in India chose our distributed inference platform to power their ad personalization engine. And 2 premier global retail brands chose our distributed data capabilities to improve the performance and resilience of their online retail applications. But all this is just the beginning. We have a large and rapidly expanding pipeline of prospects who are looking to Akamai for cloud solutions, including some with very large needs.

To satisfy this strong and growing demand for our cloud infrastructure services, we expect to continue to build out both our physical infrastructure and our cloud sales and support teams. And as Ed will talk about in a few minutes, we now anticipate significant acceleration of our overall revenue growth heading into 2027 and beyond.

Turning to security. I'm pleased to report that Q1 was also strong for our security portfolio, where revenue grew 11% year-over-year as reported and 9% in constant currency. Our security growth was led by strong demand for our market-leading web app firewall, API security and Gardacore segmentation solutions. Our WAF in particular, is seeing growing interest from customers eager to deploy the latest defenses for vulnerabilities that could be exposed by the ever-strengthening frontier models and AI-powered attacks.

Frontier models are changing vulnerability management, and we're proud to be one of the industry's must-have security providers partnering with the Frontier model companies to help ensure the safe and rapid deployment of AI-enhanced defenses. With our early access to their vulnerability detection programs we're applying our expertise to help keep major enterprises and critical infrastructure safe. Of course, and this is important to understand, attackers will also be using more advanced AI technology to develop even more potent ways to cause harm. This means that major enterprises will need Akamai security solutions even more than before. For example, there are many legacy systems and billions of deployed devices that can't be patched. They'll become a lot more vulnerable with the advances in AI and they'll need our security solutions to keep them safe.

For the devices and systems that can be patched, the patching process still takes time, often days or weeks, and they'll need out protection until that's done. We've seen this happen before when zero-day attacks emerged. And with the advances in AI, we can expect zero-day attacks to occur much more frequently. There's also an increasing challenge with scale because AI is enabling attackers to take over more devices and create enormous bot armies, we're now seeing attacks with unprecedented volumes. Just in the last few weeks, we neutralized a series of app layer attacks with millions of malicious requests per second from millions of widely distributed IPs. Akamai can defend against such attacks because of our widely distributed platform. Our WAF runs in 4,300 locations across 700 cities to intercept the attack traffic right where it enters the Internet and well before it can coalesce onto the target.

Having a great WAF with the needed defenses for the latest attacks is obviously important, but that alone isn't enough in the coming age of AI. The WAF need to be deployed across a vast distributed platform. And this need provides a unique advantage for Akamai when compared to the competition. In summary, we believe that Akamai's security portfolio will be needed more than ever before as attackers take advantage of the advances in AI. That's because of our massive platform scale to absorb attacks our unparalleled access to real-time attack data, our tight integration with the early warning ecosystem to provide up-to-the-minute defenses for the latest zero-day attacks, our large and very experienced human security operations team that's equipped with the latest AI tools to enhance visibility and minimize response times.

And our innovative, rapidly evolving and AI-enabled product suite to help prevent penetrations and to limit the damage when penetrations do occur. Customers who selected Akamai in Q1 for that kind of protection for their APIs included one of the largest telecom groups in Africa, a major investment management company in South America, one of the premier investment banks in the Middle East and one of the world's leading fintech companies in the U.S.

Customers who added or expanded their use of our GardaCore segmentation solution in Q1 and included the leading telecom carrier and media company in South Korea, one of the largest banking groups in Europe and a leading health care company in the U.S. Many of the large renewals we signed in Q1 also included expansions of our security services. For example, after we protected one of America's leading retailers from unwanted bots during the holiday shopping season, they increased the use of our services in a contract worth $24 million. We signed an expansion contract worth $80 million over 2 years with one of the world's largest video game companies. We signed an expansion contract worth more than $20 million with a global consumer electronics company in Korea. And one of the largest global professional services companies in the world expanded their use of our ZTNA solution to secure large-scale remote access as they move critical applications to a 0 trust model.

Our security solutions continue to receive top recognitions from the major analyst firms for their effectiveness. For example, last quarter, Akamai achieved a 99% recommendation rating as customers' choice at Gartner's Peer Insights report on micro segmentation. And last month, Akamai was the only provider to be named customers choice at Gartner's Peer Insights report on API protection. In closing, we're thrilled by the way our growth strategy has taken hold and is generating transformative opportunities for our business. We believe that Akamai is uniquely positioned to enable and benefit from the development of the AI-driven economy.

By bringing powerful compute directly to the data and the users at the edge Akamai is enabling and securing the next generation of Agenic AI. With each quarter, the massive opportunity we see ahead becomes more evident, and we're making bold investments to capitalize on that opportunity and enable Akamai to do for cloud and AI, what we've done for security and CDN to generate significant future growth for our business. Now I'll turn the call over to Ed for more on our results and our outlook for Q2 and the year. Ed?

Thank you, Tom. Before I get started and to build on Tom's remarks, I want to personally underscore my excitement regarding the $1.8 billion new customer win announced today. This is a powerful validation of the Akamai value proposition in the age of AI and a clear indicator of the scale at which we can operate. To fully capitalize on this momentum and support the accelerated growth we anticipate, we will be investing slightly ahead of revenue, you will see this reflected in the updated capital expenditure and operating margin outlook I will discuss during the guidance portion of my remarks. We view these investments in our CIS portfolio as critical to ensure we have the foundation to meet the significant demand we see on the horizon.

Also, driven by today's announced $1.8 billion win, the $200 million 4-year CIS deal we announced last quarter in our rapidly accelerating pipeline, we now expect total company annual top line revenue growth to reach double digits in 2027. We look forward to sharing more details in the coming quarters. Clearly, this is an incredibly exciting time for Akamai. With that, let's dive into the Q1 results. We delivered strong first quarter results with total revenue of $1.074 billion, which was up 6% year-over-year as reported and 4% in constant currency. Cloud Infrastructure Services or CIS revenue got off to a robust start to the year with revenue of $95 million, up 40% year-over-year as reported and 39% in constant currency.

As Tom noted, we are seeing CIS wins across a wide spectrum of industries, geographies and use cases. Even more encouraging, the pipeline for AI-specific use cases is building rapidly. We also maintained very strong momentum in security with revenue of $590 million, up 11% year-over-year as reported and 9% in constant currency. The strength in the first quarter continued to be driven by our fast-growing API security and Gardacor segmentation solutions along with strong growth from our largest product, Web Application Firewall.

Moving to delivery and other cloud applications. Revenue was $389 million, down 7% year-over-year as reported and down 8% in constant currency. These results were in line with expectations, driven by the wraparound impact of the EGO transaction in 2025. We expect this effect in the rate of decline to moderate throughout the remainder of the year. International revenue was $530 million, up 9% year-over-year or up 5% in constant currency, representing 49% of total revenue in Q1. We Foreign exchange fluctuations had a positive impact on revenue of $2 million on a sequential basis and a positive $19 million on a year-over-year basis.

Moving to profitability. In Q1, we generated non-GAAP net income of $239 million or $1.61 of earnings per diluted share, down 5% year-over-year as reported and in constant currency. These results include our expanded colocation investments, higher depreciation and increased head count costs, all tied to our strategic investment in cloud infrastructure services during the first quarter. Our non-GAAP operating margin for Q1 was 26%, in line with our expectations. We expect operating margin to remain in this range for the remainder of this year as we ramp up our investment to capture the exciting growth opportunities ahead of us.

Our Q1 CapEx was $206 million or 19% of revenue. First quarter CapEx was slightly below our guidance, primarily driven by timing and favorable pricing. Specifically, some expenditures shifted from Q1 into Q2, and we benefited from some lower-than-expected component costs.

Moving to cash in our capital allocation strategy. During the first quarter, we spent approximately $206 million to buy back approximately 2 million shares. We ended the first quarter with approximately $975 million remaining on our current repurchase authorization. Our intention with capital allocation remains the same to continue buying back shares to offset dilution from employee equity programs over time and to be opportunistic in both M&A and share repurchases. As of March 31, we had approximately $1.7 billion of cash, cash equivalents and marketable securities.

Now before I provide Q2 and full year 2026 guidance, I want to touch on a few housekeeping items. First, for Q2, CapEx is expected to jump significantly as we start to take delivery of the NVIDIA GPUs we discussed on our last quarterly earnings call, and we catch up on some of the CapEx that pushed from Q1 into Q2. Second, we expect to see an increase in operating expenses in the second quarter due primarily to continued investments in go-to-market and the impact of our annual employee merit cycle that went into effect on April 1.

Third, we anticipate revenue from the $1.8 billion customer win to start to ramp in Q4 and we expect to generate approximately $20 million to $25 million of revenue in the fourth quarter. Finally, regarding CapEx for this win. We expect to spend a total of approximately $800 million to $825 million over the next 12 months to support this customer. We expect to deploy roughly $700 million of that total in the second half of 2026, with the remaining balance falling into the first half of 2020.

Moving now to guidance. For the second quarter, we are projecting revenue in the range of $1.075 billion to $1.1 billion, up 3% to 5% as reported and in constant currency over Q2 2025. At current spot rates, foreign exchange fluctuations are expected to have no material impact on Q2 revenue compared to Q1 levels and a positive $2 million impact year-over-year. At these revenue levels, we expect cash gross margins of approximately 70% to 71%. Gross margin is impacted by the significant increase in colocation as we accelerate the growth in our CIS business. Q2 non-GAAP operating expenses are projected to be $346 million to $357 million.

We anticipate Q2 EBITDA margin of approximately 38% to 39%. We expect non-GAAP depreciation expense of $144 million to $146 million. We expect non-GAAP operating margin of approximately 25% to 26% and with the overall revenue and spend configuration I just outlined, we expect Q2 non-GAAP EPS in the range of $1.45 to $1.65. This EPS guidance assumes taxes of $47 million to $54 million based on an estimated quarterly non-GAAP tax rate of approximately 18.5%, it also reflects a fully diluted share count of approximately 146 million shares.

Moving to CapEx. For the reasons I highlighted earlier, we expect to spend approximately $433 million to $453 million in the second quarter. This represents approximately 40% to 41% of total revenue. Looking ahead to the full year 2026, we expect revenue of $4.445 billion to $4.55 billion, which is up 6% to 8% as reported and up 5% to 8% in constant currency. For cloud infrastructure services, we are raising our outlook to at least 50% year-over-year growth in constant currency. We expect momentum in DAS to continue to build throughout the second half of 2026 driven mainly by the scaling of our AI opportunities and the impact of the 2 very large transactions we announced in Q4 and today. Also, we continue to expect security revenue growth in the high single digits on a constant currency basis in 2026.

And for delivery and other cloud apps, we continue to expect a decline in the mid-single digits year-over-year on a constant currency basis. At current spot rates, our guidance assumes foreign exchange will have a positive $20 million impact on revenue in 2026 on a year-over-year basis.

Moving to operating margin. For 2026, we are estimating a non-GAAP operating margin of approximately 26% as measured in today's FX rates. Turning to CapEx. At this time, we anticipate our full year capital expenditures will be approximately 40% to 42% of total revenue, including the $700 million impact from the $1.8 billion contract we mentioned earlier. Before I move on, I want to provide some additional color on our CapEx outlook.

As Tom noted, the demand we are seeing for CIS, including our GPU deployments is exceptional. Our current pipeline for GPUs significantly exceeds our existing and projected inventory, meaning we may place additional GPU orders in the second half of the year to meet this demand. This is not factored into our current annual CapEx guide. We will update CapEx guidance on a subsequent earnings call if we place another GPU order before year-end.

Moving to EPS. For full year 2026, we expect non-GAAP earnings per diluted share in the range of $6.40 to $7.15. This EPS guidance includes the impact from the very large win. This non-GAAP earnings guidance is based on a non-GAAP effective tax rate of approximately 18.5% and a fully diluted share count of approximately 147 million shares.

With that, I'll wrap things up. And Tom and I are happy to take your questions. Operator?

[Operator Instructions] We have the first question from the line of Roger Boyd from UBS.

Question and congrats on the landmark deal there. Maybe if you can, Tom, just broad strokes about kind of the competitive set to win that deal. Are you going toe-to-toe with our hyperscalers or neo clouds? And anything you can provide on kind of the use case? Is this inference? Is it agenetic workloads? And -- when you think about your compute-enabled POPs, just how is this customer leveraging the Akamai network as a whole? .

Yes. I can't give any more details about this specific deal. But in general, yes, we do compete with the hyperscalers and the neo clouds with our cloud infrastructure services. That's the primary competition. They select Akamai because of our proven ability to manage and scale complex distributed systems, our ability to get the necessary data center space and locations around the globe to interconnect that with the world's largest and best-performing delivery network and leading security solutions. We offer the best in terms of latency, scalability.

We probably deal with more data center companies than anybody with being in 4,300 locations across 700 cities and 130 countries. So yes, we have significant competition. Every deal is competitive. But we also have unique capabilities, which is, I think, why our pipeline is so strong and why we're winning some very large deals. .

Excellent. And then just on security, I wonder if you could unpack what you're seeing from a demand perspective there. A nice result in the first quarter. Just what are you seeing around conversion rates, sales cycles? Are you seeing more urgency from organizations that are thinking about ways to limit the blast radius and defend against kind of an AI fueled attack landscape. .

Yes. I don't think I've ever seen the CISOs more agitated and feeling more of a sense of urgency than they are now. Over the last several weeks, couple of months, I've had the chance to meet with a lot of the world's biggest company CSOs. In many cases, the CEOs and senior executives, and they are very concerned about what happens when the attackers get access to advanced AI with the latest AI frontier models, which it seems that they will. This is going to uncover a lot more vulnerabilities. We're going to see the equivalent of a lot more 0 days, and they are literally scrambling now in many cases to make sure all their applications, their agents, their APIs are protected by Akamai. .

And you can imagine, most of the world's major banks rely on us for security, and they're looking at a pretty big wave of new attacks coming their way. So this is -- I don't know of a comparable time where there's this much concern about what's going to happen with security and also this much appreciation for what Akamai provides with our security platform.

We have the next question from the line of Patrick Goel from Scotia Bank.

I mean this one is for Dr. Tom. I mean when I think about Akamai, the value prop for the last 30-plus years has been the distributed architecture 70 cities, 130 countries. When I think about this mega deal, is that a kind of highly distributed use case or should we think about it as being served from a few like sub-10 type data centers? .

Well, I'm not at liberty to talk about the recent deal. However, I think when you're thinking about Akamai's value proposition, you hit a very key point with our really unparalleled distributed architecture. And I did reference a bunch of use cases in the prepared remarks. And yes, they very much rely on our distributed platform where you want to get the agents and the applications, the business logic close to users, close to the data, so you get low latency, you get stability particularly anything to do with video processing or video generation, even takes a -- needs a lot of scale. And Akamai is unique there. And so I think absolutely, what we're able to offer is very compelling. .

Yes. And look, congestion I guess a follow-up, please. Ed, you made this kind of suffer point that there's a CapEx guide that might have to increase .

Sorry to interrupt you, Patrick, your voice is breaking. If you could probably go off the speaker phone, we would be able to hear you better. .

Thanks for that. I guess this follow-up is for Ed. I mean, you gave us the CapEx guide -- but then you kind of gave us this kind of subtle point that might have to increase CapEx further. Can you just help us understand the nuances of why there might be an increase in the CapEx midyear? And I guess what that might mean .

Yes. So thanks for the question, Patrick. So what I had mentioned was we have a very, very strong pipeline for the platform. And we're just starting to get the bulk of those chips up and running now, and we've got a very large pipeline. It exceeds what we have in inventory. So obviously, we want to prosecute that pipeline, start winning all those deals, converting that into contracts, et cetera. And then the reason I sort of hedged a little bit is, one, we don't obviously fulfill that pipeline. But two, there is some time that it takes to get the chip. So even if we were to place an order, it may slip into next year. So what I want to do is just give it another quarter and if, in fact, we're in a position to place an order and receive that by year-end. We'll certainly do that and let you guys know.

I see that as a very bullish comment. And again, I just didn't want to come up and surprise you with another whatever it is a couple of hundred million or whatever the order may be without at least giving you some color behind that.

We have the next question from the line of John DiFucci from Guggenheim Securities.

My first question is for Ed and I have a quick follow-up for Tom. Thanks for all the detail on CapEx. But when I think about the CapEx for this mega deal, and I think Patrick was kind of going here. I mean this is over a long time, right, 7 years. Are you accountable, for example, like right now, we're seeing higher memory costs than we would have thought of maybe a year ago. Are you -- like when you've locked in this deal, do you also have the supply locked in? Or are you exposed to that, if that were to happen, I don't know, 2 years from now, higher prices again on that? .

Yes. Great question. So that I was fortunate enough to work very closely with the team on both sides of this transaction. So yes, we've been able to get the supply chain ready. We anticipate receiving all the goods that we need just to deliver this service over the 7 years within the next 12 months, with obviously, you saw the way the CapEx was broken out with the majority of it this year. So we anticipate receiving a significant portion. Now there's always the potential for some slippage in delays. But we have mechanisms in our contracts to deal with if, in fact, say, 6 months from now, prices were to go up.

So we've taken that into consideration. We've got that taken area. In the way this -- from a revenue perspective, the way to think about this deal is it's a set amount of capacity that we're deploying and there's no usage to it. It's a straight usage -- sorry, committed deal over 7 years. So as soon as we ramp all the capacity up, we'll start taking the revenue for full year. I expect, as I said, a little bit this year. And then next year, we'll get a partial year as we receive the remainder of what's to be deployed. And then from there, it will go on for the remaining 6-plus years.

Okay. So that will kind of look -- even though it's a consumption, it will kind of look like a subscription. Is that accurate? .

Exactly. Yes, that's exactly the way I think about it. .

Awesome. Okay, great. And Dr. Tom, a component of your delivery business is video streaming. And in March, we saw open AI they confirmed they shut down their AI video generation system, Sara. I'm just curious, do you expect that to have any effect on your delivery or compute business forecast?

No. We partner with OpenAI on security vulnerabilities, helping to find them and protecting our customers for the associated attacks. But open AI is not and has not been a customer of Akamai. So yes, no impact on us at all. .

We have the next question of the line of Jack Saner from KeyBanc Capital Markets. .

This is Aidan Daniels on for Jack Snider. With this big deal, as you allocate capacity going forward. How can we kind of think about the impact on any amount of on-demand GPU capacity you're able to offer going forward? Like how are you kind of balancing what you have committed from this deal with maintaining flexibility for newer incremental demand going forward? .

No. We support both on demand per token or per VM our access to our platform. And also, we support large tranche deals. And so it's not really a matter at this point of trading off. And as we need more GPUs, as Ed said, and that may well be the case that we would purchase more. .

Awesome. And then just one quick follow-up. I know you can't really talk too much about the deal. But I guess like how can we kind of think about the proportion of whether it's CPU or more of the GPU inference cloud going forward? Is there kind of like a framework we can think about with this deal? .

With this deal, we can't comment on this deal. However, in general, with inference and AI, you need both really. And part of the value we provide is that we can help provide the computational resource that's most appropriate for the workload that you have, which might be CPU, might be because you want to be as efficient as possible. And also you want to have it be as close as possible to the user, so you get the best performance. So it's a mix, and every application is different in the mix of CPU versus GPU that it needs. .

We have the next question from the line of Fatema Bolani from Citi. .

Just a higher level strategic question. you have opted to take more of a dedicated capacity approach in terms of satisfying demand and some of the supply constraints out there. I wanted to sort of dig deeper into why simply because the stock rates and the market rates for what otherwise could be almost entirely a rental or GPU as a service business are significantly more attractive.

So I just kind of wanted to get the division and the thought process and the decision-making calculus around steering the network and the platform more towards larger customers, longer commits and more dedicated capacity? And then I had a follow-up as well, please. .

Well, we do both. And the larger, bigger deals with long-term commits are more attractive in many ways. You have the commit. And the big deals, yes, the pricing would be lower, but we also support the on-demand where you can buy it by the token or the hour and you get a little bit higher pricing, but there can be more expense associated with that, getting the customer on if you have a rep engaged in the account. But both are attractive and we support both. So it's not a matter of us doing one or the other.

The one thing I would add there is this time, just to jump in here for a second. The customers are really driving that. If I look at our pipeline, a lot of our customers want to have dedicated capacity, say a dedicated number of GPUs or whatnot because there is a scarcity in the marketplace. So rather than going on a consumption basis, they can get slightly better pricing and lock in that capacity for themselves. So it's really a market-driven thing more than anything.

I appreciate that. And Ed, since I have you, you telegraphed for us pretty nicely that should the pipeline continue to grow and morph in the way and positively morph than the way you are seeing you will be very open to continuing to throw down CapEx and bringing and lighting up by the lots online. But wondering sources of funds and capital to fund these investments. Is that something you feel you can intrinsically do from running the business? Or should we expect maybe other of the capital to be tapped as you build out with a bigger CapEx profile for some of these larger customers under demand? .

Yes. So so far, no issues as far as financing these build-outs from our own capital today. We are obviously a company that's very profitable, produce a lot of cash. Obviously, in the years when we're investing big cash flow will be a bit lower. But these things have phenomenal free cash flow after you do the actual deployment. So that's one attractive things. We -- from a cash and equivalents, we have $1.7 billion on the books today. We also have a line of credit of $1 billion if we need to tap it. And then obviously, we've got excellent credit and have no problem raising money in the capital markets if we need to. Right now, we haven't announced anything there. And if we continue to get large deals and need to get capital, we'll certainly go to look to do that. But so far, we've been able to use our own funds. .

We have the next question from the line of Mark Murphy from JPMorgan. .

This is already Vila from JPMorgan on for Mark Murphy. Great to see the momentum you're having with the large deals with companies on the AI technology frontier. -- had a large deal last quarter, another 1 this quarter that dwarf the 1 before it. So just at a high level, can you help us understand from your perspective, like it seems like all of a sudden, you're getting some of these large deals. Has this been brewing for a while in the pipeline? Or have these been a little bit faster? What's changed that's brought a lot of this business here at doorstep seemingly pretty quickly from our point of view. And then as a quick follow-up to that, as you're dedicating the financial and operational resources to the CIS and the large deals, does it change how you're thinking about other business segments? .

Well, this has been the strategy all along. So -- and we're very pleased to be executing against it. The goal has been to be deploying a distributed inference platform distributed compute platform that would be desired by enterprises really across the spectrum and with many large customers. And of course, Akamai's customer base does feature many of the world's largest enterprises. And as we've talked about before, they spend 10x or more on compute than they do on our traditional services, delivery and security. So this is exactly what we said we were going to do. And now we're delivering those results. The platform is to a point where we can do that. And I think you'll see more of this going forward.

We have the next question from in Loaning from Morgan Stanley. .

Congrats on the big steel and company history. On that point, this might be a trivial question, but in terms of this $1.8 billion contract, is that more of a public cloud opportunity? Because I know part of the public cloud business also has a GPU component? Or is this -- was it specifically for Akamai Inference cloud? That was the first question, and then I had a follow-up. .

Yes, we really can't talk more about this particular deal. But obviously, there are a lot of companies where we've signed contracts that we did talk about across the spectrum. And those deals for our inference cloud and our cloud capabilities for our GPUs and our CPUs. And it really is our ability to bring the right hardware for the particular application and have it located where you get the best benefit for the use of that application. .

No, that's fair enough, Tom. My follow-up question is it actually goes to the delivery business. And there's a lot of people in the market kind of debating about a potential new lever for growth in CDN and delivery in a world where you have millions, potentially billions of agents running around calling tools, executing tasks doing web searches, has the team internally sort of revisited the thesis around the secular growth prospects in delivery? Or is it still a business that you're mostly looking to harvest for profitability and gross profit dollars to fund the compelling opportunities in security and compute.

Yes, great question. When you look at what is the proliferation of agents and what's coming, the biggest driver for growth is going to be the compute platform. the cloud platform that supports that. And we're really well set up to do that. Next, you have a big security issue because AI and the agents are a whole new vulnerability surface that not only do you need your web at firewall, your API security, you need special security for AI. And so we get a real tailwind from STG for our security technology group there.

Also, the agents are -- you have to interpret what an agent is, who's behind it and what they want to do when you're delivering or protecting an application or a site or another agent. And the response you give is really tailored to what the customer wants you to do when an agent of this flavor comes and interact with you. And so we developed a lot of capabilities there. And they fall generally within our security capabilities.

Now in terms of delivery, yes, there'll be some traffic that used to be human generated, now agent generated, okay. That doesn't make a huge swing in the amount of bits you're delivering. That starts to change if you have agents dealing with video, generating video like you go to a commerce site and the user wants to see what do they look like in that sweater they're thinking of buying and if you generate a video showing them wearing the sweater, that will improve the return for the site, and that generates a lot of traffic.

And so we're just at the very early days of seeing things like that, they're being experimented with now that could generate more traffic for delivery but the biggest impact for us is in the cloud business and then next in the security business. Delivery really important, very synergistic with our whole platform approach does generate a lot of cash for us, and we're plowing a lot of that cash into the growth of the cloud business.

We have the next question from the line of Mike Cikos from Needham .

Congratulations on the strong quarter and the customer win. I just wanted to make sure I'm understanding at least the mechanics of this deal. So you signed a 7-year $1.8 billion commitment. Can we expect the full $1.8 billion to show up in RPO? Or does that include anything as far as potential renewals? Is that all take or pay? Just anything to make sure we're understanding the mechanics of the deal. .

Yes, sure. So I touched on this a little bit earlier that -- and there was a follow-up question around this notion of capacity, dedicated capacity versus pay by the hour. This is more of the dedicated capacity. So as soon as we get the capacity set up, we will take the revenue ratably over the contract. As I said, we'll get some revenue this year and a little -- not a full year next year, but a partial year next year is we're still going up and getting the capacity up and live. In terms of the way you account for RPO there's -- we will see most of that in the next quarter. And then by the time we get everything delivered, it will be all in RPO eventually. There's just some odd mechanics with the first 12 months and how we're dealing with how we're receiving the goods, and we talked about a pricing mechanism to handle it prices would go up or down, that sort of thing.

So there's a little bit of nuance in there. But once we get this fully up and running, you'll see it in our RPO. There'll be some amount next quarter and then it will build there. .

I appreciate you spelling out the mechanics there. And then for Dr. Lee, just to make sure I'm clear as well, and it's great to hear that your largest security product here with WAF is seeing some stronger growth, which I wouldn't have expected. Can you just tap into that one more time as far as what's driving that? Is it really this heightened environment that we're in? Or is there something else behind there?

Yes. There's -- as you know, real advances in AI and it's getting much better at finding vulnerabilities and helping the attacker take over devices and penetrate enterprises. And you need our defenses now more than ever before. There's billions of devices out there that you can't patch. And now the adversary can find ways into those devices and take them over. And so as a result, we're seeing attacks much bigger than we've seen before.

Literally, application layer attacks for millions of distributed IPs with millions of attacks on a target per second. And you can't defend against that with just a WAF in a data center or anything close. You need the vast platform that we have to be able to intercept all that traffic and deal with it because you got to separate the bad stuff from the good stuff, and there's a huge amount of the bad stuff now.

So our platform, the physical infrastructure is needed more than ever before for our security services. And our customers know that. And there is a heightened sense of urgency now because they know the attacks are getting more capable due to AI and larger in size because they can take over all these devices and launch the attacks from many more locations. And so that's why we're seeing things like our web app firewall suddenly a lot more demand. Now AI helps on the defense, but doesn't solve that problem. And so net-net, this is a very challenging time for CISOs, and that's why they're turning to us to make sure everything that they have is protected by Akamai.

It's great to hear about that a little bit. And congratulations again on the strong customer win. .

We have the next question from the line of Frank Louton from Raymond James. .

Yes. Just a follow-up on the question about the $1.8 billion, how that's being booked. Is all of that going to come in as revenue? Will any of that be counted as paid for upfront CapEx? Or or something like that? And then I also wanted to follow up and see how many locations do you have Inference Cloud built out to currently? And what's the plan? .

Sure, I'll take the first part, Tom, you can take the second. Yes, all revenue. There's no offset to CapEx or anything like that, so it's going to be all revenue.

Yes. To the second part of the question, we have -- Inference Cloud covers all of our 4,300 locations. We have functions as a service running in a serverless way in all 4,300 locations. We have our managed container service running in well over 100 cities and conceivably could run in all 700 cities, but active and well over 100 today. We've got full IAS capabilities in several dozen cities and a couple of dozen of those are equipped with the new 6,000 GPUs.

And the goal, of course, is to have all that's orchestrated so that when there's an application or an agent that needs to be run, it's run on the most computationally efficient resource. If you can do it on an edge server with the existing CPU, fabulous, fast, very low cost. If you can do it on a container in the same city in a CPU, great. If you need a group of GPUs in one of those couple of dozen locations, okay.

And again, so you want it to be on the most efficient resource to be close to the user and to already be ready to go. You don't want to have to spin it up in response to a request. And that's what our orchestration layer is designed to make possible. And this is how it fits in with the vision from NVIDIA with the AI grid, you think of AI like you would an electrical grid. And that's what Akamai is building.

We have the next question from the line of Will Power from Baird.

Great. I'll echo my congratulations on this massive deal. Just maybe 2 questions. First, just a clarification, perhaps, Ed, when you talk about needing additional GP, do you need more GPUs to satisfy the new deal? Or is that more related to the building pipeline? And is there -- I know the timing is uncertain as to when the GPUs might be available but is there a real work for what we're talking about in terms of overall cost. And then I have a second question.

Yes, sure. So sort of if you listen to the prepared remarks, we talked about all the CapEx that we need is in the guidance for satisfying the $1.8 billion. So that's separate from the comment I made around the additional GPU purchase. And that was really tied to how we're doing with the pipeline, how quickly we can execute on that. And again, there's always the question of, can you get them delivered in time. So we'll give you more information on that.

It really depends on what we're seeing in terms of demand. We're seeing a pipeline very, very strong, some very large opportunities, some customers that want to start with a couple of hundred GPUs, some that want to start with 1,000 or more. So it's really all over the map in terms of opportunities, and it's growing every day, which is great. So we'll size it up for you. The last one was around $250 million CapEx, I don't have anything to tell you in terms of how big I think it will be. But hopefully, I'm telling you it's a really big number because we've got significant demand for it. .

Yes. Okay. And then any way to kind of frame how you're thinking about gross margin, operating margin impacts? I know 2027 sounds like it's still a partial year. As you look into 2028, how do we kind of think about how this impacts the overall financial model relative to where -- maybe relative to where you are today?

Yes. So let me talk about sort of a high level. So if you think about some -- especially some of these larger deals that are more of that someone who comes to us and say, say, I want 1,000 GPUs are in this case with this big customer, I want a certain amount of capacity over a long period of time. The biggest cost driver there is your depreciation over the period of time. The costs that go into your cash gross margin are much less, right? It's your co-location costs, maybe if there's some bandwidth or networking costs and things like that. And sometimes, there's some people cost. But generally speaking, these scale pretty well.

So what you would expect over time is your cash gross margin could improve. Now obviously, there's some push and pull here between lighting up colo for expected demand, and you have to light that up first and all that kind of stuff. So it will take a bit for this to sort of play out. But you should see your cash gross margin expand a bit. Your EBITDA margin expand a bit. And then from an operating margin perspective, it really depends on the mix.

We're willing to do some of these deals that are a lot larger, potentially margins that might be sub the 30% operating margin. If you look at the -- certainly, the GPU by the service rented GPU, much higher than the company operating margin. You're going to get a lot of scale across the OpEx as we take on certainly larger customers. So we're going to really focus over the next year or two on really capitalizing on this growth. So we won't be in a margin expansion point right now. But at some point, that will happen naturally and your free cash flow margins will improve and things like that. So that's sort of the way we're thinking about it, but we're really excited about going after this growth opportunity. We're going to continue to invest to go get it.

We have the next question from the line of James Fish from Piper Sandler. .

Look, given what you've discussed around Power in the past, with the large site having, I think, Ed, you said 5 to 10 megawatts and smaller sites, a fraction of that, it puts you above 300 megawatts. So now, you don't have enough revenue that online to this. So how much of that power is for noncompute services? And is that why you need to bring on from what I can tell, another roughly 40 megawatts just for this deal alone, as it does seem to kind of mask and you guys should be able to kind of support this if that power is all allocated to compute. So can you just walk us through how you guys are in terms of megawatts and kind of what the plan is by the end of '27 then? .

Yes. So I didn't quite follow all that but let me just start by saying your math isn't right in terms of what would be required to deliver this particular deal. I'll just leave it at that. It's significantly lower than that. In terms of our capacity, we have -- the major -- if you think about what uses the megawatts of power that we have, the CDN and the security business is a small fraction. You think about that as being kilowatts in some cases, maybe a megawatt or 2 in some of the big CDN deployment. So there's not a ton of like massive power required to run the CDN business.

When it comes to the compute business, it's a lot greater, especially when you get customers who want, say, a few thousand GPUs in a particular location or they're in say, 20, 30 locations, and they've got a lot of CPU. So you do tend to see a lot more need for power there. And what I've talked about is our typical deployment for some of our larger locations, we talked about having 40 core compute locations. And those were in the 5 to 10 megawatts expandable to say, 20 to 30 and it really, again, depends. And we can get a little bit bigger than that, but there's plenty of opportunity for us to get additional colo. We expect to light up a lot more going forward here.

So if the concern is that we don't have access to enough power or can't get colo, that is not a concern of ours right now at all. We -- as Tom talked about earlier, we've got great relationships, and we're a very attractive client for some of these data center providers. We've got excellent credit. We're not a do-it-yourselfer like a hyperscaler. We've got much better credit than, say, some of the neo clouds and we do take significant chunks of colo, in some cases and actually help some of our colo partners build out.

So I'm not concerned about that at all. And the power dynamics for each one of the different products is different. So GPUs take up more power than CPUs. And so there's math that goes along with that, and I've shared that math. The CPU math is much lower because you don't need as much power. And then also the type of equipment that you're running can also have a pretty significant impact on power. Certain -- we're seeing a lot of interesting hardware providers coming out with stuff that's incredibly efficient from a power perspective.

So it's really -- it's hard for me to answer that question specifically, but just take it that we're not too concerned at all about getting enough power to run these things and the power requirements are very different. But we factor that in to any deal that we do and ensure that we're not going to take anything on that's not profitable. .

Got it. Makes sense. And just to add on the security side, normally give us an API and Zero Trust versus kind of the core. I guess, how did that trend? And then how did compute in the quarter trend between -- with enterprise .

Yes. So with security, we didn't break out API and Guardio, we did say it was the majority of what's driving growth. And I would just say the growth rates are similar to what we had last quarter. Just remember, last quarter had a fair bit of license revenue. So when you back that out, it's kind of apples-to-apples growing at roughly the same rates when you back out the impact of the license revenue. So still very, very healthy growth rates there. .

And then on compute, you asked about enterprise compute. We don't break it out that way, really, the way to think about enterprise compute as I which is broken out separately, and we do provide what we used to call our application services, which is included inside of the third bucket, the delivery and app services. So that number is broken out for you. So 40% growth was for year-over-year, and we expect that to accelerate.

We have the next question from the line of Jonathan Ho from William Blair & Company. .

Let me echo my congratulations as well. Just one for me. Just given the types of mega customers that you're bringing on to your platform, is there more opportunity to upsell to them once they are on your platform, are there potential additional services? Or could they come back to the well if they continue to expand their growth as well? .

Yes. As you know, the demand for AI is rapidly increasing. We're really early on there. And yes, I would expect there's plenty of room to grow the existing base and, of course, add other customers of that scale.

Thank you. We have time for 1 last question from the line of Jeff Vander from Craig Hallum. .

The question is sneaking me in there. Two quick ones. First, maybe, Tom, just a lot of blowback nationally against AI data centers and the power consumption correlated to them. As you're stepping into deals of this magnitude, how do you think about staying out of the crosshair of some of those community-wide pushback on the broader AI compute environment, if you will. .

Yes, I don't think we're of a profile in the popular press anything like the giant hyper killers are. So I don't think that's really an issue for us. So yes, we're not worried about that yet. Maybe that's even a problem good to have once we're much, much larger than we are today.

Okay. And then second, the -- on the security side, given the comments about AI becoming a tailwind there, would you think this year is likely a floor in terms of growth rate, namely we should be thinking maybe reacceleration as we get into '27 and beyond? .

We'll see. We have -- we gave you guidance for the year. We're obviously very pleased with what we saw in the first quarter. There's -- we do like what we see especially around API security, -- still early days, low penetration rate. GardaCORis growing very, very consistently. So yes, we'll see how it goes, and we'll update you as we go. .

This concludes our question-and-answer session. The conference has now concluded. Thank you for attending today's presentation. You may now disconnect.

All right. I'm Sanjit Singh. I am doing my last presentation for TMT' 26, and we're going to end it strong with Akamai, we have Ed McGowan, Chief Financial Officer at Akamai. Ed, thank you for joining us at TMT 2026.

Well, thanks for having us. Great to be here.

There's a lot to talk about. The Akamai story has gotten really exciting. And so we're going to dive into all the things that's going on in the company.

Before we get there, for important disclosures, please see the Morgan Stanley research disclosure website at www.morganstanley.com/researchdisclosures.

So I wanted to start the conversation with you, Ed, on the various acts at the company has been good going through, right? The company sort of pioneered the delivery networking market, you're sort of the market share leader there. Then you entered into an act 2, which saw you build a $2-plus billion security business over the last decade. And now in recent years, you're pursuing act 3, which has seen the company enter the public cloud market and more recently, the GPU as a service for Gen AI inferencing with Akamai Inference Cloud. So a lot going on.

For those investors who haven't been close to the story for some time, can you lay out the vision here for Akamai's Act 3 and why the management team is excited about the prospect for growth heading into the AI era?

Yes. Great question. So most folks know us as for our first act, as you call it, with delivery. And we've always had a vision of some -- we started off actually edge computing way before it was even a thing, right? This was back -- I can remember when I first started in 2000, we are doing some form of edge compute. It was pretty limited in terms of what you could do with it. It was more a functions as a service. But the notion of leveraging our platform for compute was always something that we envision doing.

Security was a very natural extension. So as we're delivering these websites, whether it's a commerce site or a banking site, there's so much malicious activity that happens, it was very natural to be able to protect those sites and do it in a way by leveraging our platform. And at the same time, I'm delivering you a shopping cart experience. I'm also using web application firewall rules to protect that particular interaction.

So that business has been phenomenal for us and grown, adding $200 million to $250 million of new security revenue pretty much for the last decade, really. And it's a lot of internal innovation. M&A has been part of our strategy. We've been very successful in integrating companies into our portfolio and getting a larger and larger share of our security wallet, and we continue to expect to do that.

And we actually think that we're the next generation of the web, if you will, of the agentic web, a lot of our security products will be applicable there. Obviously, we'll have to do some innovation and do some things to conform with the new protocols and things like that. Noname and API Security is a perfect example of that, where you've got massive amounts of data flows. Today, it's API, tomorrow, it will be talking to different models, going back and forth to different systems and all that will be exposed. And I think we have a great opportunity there to continue to grow our security business along with our compute business.

So there's a natural adjacency with delivery compute and security. And most recently, over the last several years, what was happening is some of our bigger customers, more innovative customers are coming to us saying, we want to run this application and it requires us to have root access. We need to -- we can't just do functions as a service. I can't use your kind of edge workers platform to program in Java script or program in web assembly. I actually want to run my own proprietary code.

I'll give you an example. We have customer who's using us with their own proprietary technology for live streaming. And they want to do things like synchronization to make sure that the video is synchronized across all different platforms, and they have some code that they use to get much better performance out of that particular stream. Now in order to do that, they needed to be able to run their code on our platform. And we're not designed to really do that. So we bought Linode several years ago, and we deployed our managed container service, where we're essentially slicing off a sliver of CPU from all of our servers around the world. And we have containers where they can actually run that code in that in all -- as many locations as they want to.

And we also saw, at the same time, as customers are coming to us for these more traditional compute use cases where the public clouds weren't giving them the type of performance that they needed. One good example, we have a case study on our site about Apple, what they do with Private Relay, where you couldn't have, say, 4 or 5 big clouds backhauling all that information and actually run the service because users are everywhere.

And at the same time, with our security business, we're spending a fortune on third-party climate, right? If you think of the massive amounts of data that we're processing. And we realize that there's a pretty good opportunity here for us to leverage what we do really well, running a massive distributed system and get into the compute business. Number one, it's going to save us a fortune. So we've started off as our first major customer, and we were able to offload a tremendous amount of workloads off of the public clouds, and that's continuing to grow today.

So we really proved out the fact that we could run this. We needed to acquire some technology to help us get there, and we bought Linode, which was primarily focused on small, medium business, the hobbyist, if you will, and we turned it into an enterprise-grade system. And just so happens now with AI, the demand for compute being in a distributed fashion is even growing greater and greater and did some interesting work with NVIDIA back in the summer, early fall and just most recently deployed GPUs in what we call our Inference Cloud. This is sort of the latest innovation. So it's a very natural extension of what we're doing, well informed by customer demand, and it's been growing phenomenally well for us.

But like from my perspective, I see stability in your delivery networking business, in your security business while you're seeing accelerating growth in your compute business and the CIS business. So let's dive into the compute and the cloud infrastructure services opportunity, starting with public cloud. Public cloud growth is accelerating across the industry, including for Akamai and your Linode business. What are the reasons why this part of the business has moved to a higher gear recently? And is this growth being fueled by a handful of customers? Or are you happy with the breadth of that contribution?

Yes. So I just actually came up with my last meeting with a group of investors where we were talking about, if you had asked me 3 or 4 years ago when we started this, we're now at a $400 million run rate in the CIS business, our cloud infrastructure service business. I would have thought we would have a much higher customer concentration. We would have gotten a few big customers to spend tens of millions with us, and then we would just start to get into a much deeper breadth across, say, finance and retail and travel and all that sort of stuff. And it's exactly the opposite, which is great.

We actually -- that -- if I look at the composition of that, that's hundreds and hundreds of customers. It's anything from a customer doing a couple of hundred thousand dollars with us upwards to a couple of million dollars a month, but it's not driven by a small handful of customers. As a matter of fact, very little revenue today is coming from these mega deals, right? These are starting to ramp up, and they're going to become a lot more material to revenue over the next year or so.

And we're also seeing our pipeline full of anything from customers working with our partners. So we have like a store, if you will, where you can go and say, if you're a media company, you want to do media workflow, we have media workflow partners where you can leverage their technology, use our storage and our compute. That's been going phenomenally well. Observability is one of the fastest-growing sort of use cases that we're seeing where customers are using an observability partner of ours and then ingesting information, storing it, using compute and creating all sorts of dashboards and things like that. And I wouldn't have thought that was how we would have scaled up so quickly. But it's been a great surprise and the demand is very strong across the installed base as well as new customers coming to us.

We're starting now that we've -- started with the agentic web, we're actually getting a lot of interesting inbounds. It sort of reminds me of the early days of Akamai when you walk in when people used to have desk phones and you see the red light on and people will be calling and saying, "I need this stuff." We're starting to see that kind of activity again, which is kind of neat.

As I see investors sort of reengage with the Akamai story and maybe to take a break for a couple of years, but kind of looking at the trend lines of the business. The question I often get when it comes to the public cloud business is what gives Akamai or why do customers choose Linode versus the hyperscalers versus some of the alternative clouds out there. I think one of the important points that you guys make is that all three major hyperscalers are customers of Akamai but if you can just talk to the value proposition of Linode in the public cloud market.

Yes. So I think there's a misnomer out there that it's a zero-sum game. Most companies are doing multi-cloud, and there's applications even with some of the hyperscalers where we just get better performance, right? There's an application where you're just better served, whether it could be economics, it could be latency, we're using a more distributed cloud is a better answer.

We've had companies trying to comply with standards in finance where, today we're complying with Apple Pay. And you can't get that out of a public cloud, but you can get that by using us. We have some customers that might have workloads that are in a third-party cloud or public cloud, that their data is accessed very frequently. So you store the data, but you're reaching in and grabbing that data is going all over the place. That's very expensive. You have a lot of hidden costs, egress fees, for example, it's something that we've had customers say, if I move to you, I'll use your computing storage and you're not going to charge them for the egress fees. And the reason we can do that is we have one of the largest backbones in the world and we deliver hundreds of terabits per second, so the cost for us is virtually nothing. And so we pass that on to our customers and we can differentiate from an economic perspective. So if you need economics, performance, sometimes it's just for diversity, where you want to have multiple cloud providers. It could be maybe a major AWS shop, but you have 2 or 3 workloads with us.

So it's not necessarily a zero-sum game or one cloud takes everything.

Just a couple of follow-ups there. In terms of those lower egress fees, can give us a range on how much cheaper you are versus your -- versus the hyperscalers. And the other question was around -- let me just stick with that first.

Sure. Yes. So a lot of times, even in the CDM business, we started to see this, and we actually started to build some products to help with our customers, where they might have their web infrastructure at a hyperscaler. And you have the notion of your origin, right? So we're all the -- say, if you're a video customer or even a retailer, you have lots of images and that sort of stuff. The idea of having a CDN, obviously, is you want to get great offload meaning every time a request comes in, you want it to hit an edge server and not have to go back to the origin right? Better performance, lower cost. We have customers where sometimes we'll have 99% offload or, 99.5% offload and their bill for egress could be equal to what they're spending in CDN. But just that little bit of a cash miss. Because it's so expensive. And I don't have all the numbers handy, but you can look at some of the public costs for what it would cost for egress. And it's -- compared to CDM pricing, more expensive.

With the managed Kubernetes surface that you guys have launched, is that expanding the types of workloads that customers can run? Is that becoming a criteria for why customers gravitate to [ the low end ]?

Yes. I mean that obviously like if you think going back to the beginning of the discussion, where our first origins were in this notion of Functions as a Service, programming your cash, programming some basic content awareness where you could do, tailor personalized content based on an IP address. You can program the CDM to do that. That's very basic stuff. The -- having a managed container service, now I can run all sorts of different code and workloads in that container that I couldn't do in the earlier days on programmable CDM, if you will.

And in terms of the hyperscalers as we've mentioned, that their customers of Akamai and their public cloud service, what are the hyperscalers using you guys for?

Yes. So we've talked to sort of at a high level, some are using us for video delivery, some are using us for API management, some are using us for advertising decisioning and things like that.

Awesome. So let's move the conversation to the inference opportunity. Akamai Inference Cloud was announced at the end of October, beginning of November. The edge computing opportunity has been discussed about as a potentially big market for several years kind of across the industry. I would say it's generally been slow to ramp, and maybe it's having its moment now. But what -- with the opportunity around the inference, what gives you guys the confidence that a significant portion of AI inference will happen at the edge? And what will be the drivers for this opportunity to materialize?

Yes. Good question. If you think about the next iteration of applications. Today, you might go to a retailer or to like a Home Depot or a Lowe's where you interact with it with a search, right? So you do a search, and you get a bunch of stuff and it's helpful. But imagine now you work with a virtual travel agent or a virtual contractor. The amount of data that needs to be processed and the amount of compute that happens is much more robust, right? You might be -- have a small language model that you're using in real time. And you also want to make sure that, that is trained up and informed, et cetera.

But the amount of compute for that is pretty significant. And you couldn't do that using, say, like our edge workers or program that in JavaScript and that sort of stuff. But latency is a big issue, right, where just like in the old days, when you had a shopping cart, if your shopping cart took forever to load, you see the circle going around, you're going to abandon that. You don't leave the site or refresh it. And the vendor, the partner of ours, ours customers will lose a sale.

Same thing will happen in this world. It's just now compute is going to be a much bigger component, right? Because in order to deliver that experience, maybe that virtual travel agent is booking your dinner reservations for you. It's putting markers in your calendar for travel might be getting you a car service and recommending all sorts of different things. That's a much better experience, but it requires a lot more compute, Latency is a big issue, right? Workloads like robotics, you need to be close to where the factory is. Autonomous driving is going to require extremely low latency.

So those are just a few ideas of what we're seeing from customers. And what's interesting is we started working with NVIDIA and doing the trial to see what can we get out of these RTX 6000 Pros? What is sort of that token to megawatt mathematics look like? And we were blown away with how -- what we can do with those in even relatively small deployment, say, maybe a megawatt of power gets you about 1,000 GPU, you can do some pretty powerful workloads. And we have customers coming to us, anything from, say, AI start-ups that want to do some rentals and just get access to GPUs. And we also have others that want to do clusters where they say, I've got a -- maybe it's a research project I'm doing. Maybe it's a post training. Maybe it's robotics. Maybe it's a really robust advertising decisioning model that wants to work on the fly and requires an enormous amount of data.

So there's -- we're starting to see some really interesting use cases now and a pretty nice pipeline that's building up. And latency is a factor. As a matter of fact, with this very large customer deployment, the latency was the requirement here was such that typically, within the U.S., you get pretty good latency, like you're in the middle of the country and you're on one of the coast. It's usually tolerable. In this particular case, it wasn't. We had to be very, very close to where the application was being run.

Yes. That makes a lot of sense. You mentioned that large customer, it's a $200 million 4-year deal with a major tech customer at the forefront of AI. Why did that customer choose Akamai? And what needs to get done between contract signing and the timing of revenue, which you sort of pointed to at the second half of the year, maybe Q4 in terms of what we're seeing? Just the time line between why do they choose you guys and then the time line between to get revenue to start to materialize?

Yes. So this was an existing customer, not one of our top 1% customers. It's been a customer for a while. And when we announced the Inference Cloud back in the fall, they had called us and said, "Hey, we've got this unique circumstance where we want to run a cluster of GPUs. And here is the type of performance we're looking for.

We want to trial it you, right? And we actually have some -- like NVIDIA does refer some business our way too where customers will come to them and say, "Hey, I've got this use case, I want to do X, Y and Z. I need this type of performance. How would I go about doing it?" and in some cases, working with us is the best chance you have to meet certain requirements. So it was just an idea at first, and then it was -- let's run a trial, so you do a proof of concept, and you run that on a small scale in one of our locations and see what type of performance you get. Once the customer was comfortable with that, they placed a very large order. In this case, it's a fairly good-sized cluster. We had to light up some more data center space in particular for this. So in terms of timing, signing a contract of that size depending on -- it's an existing customer, so that helps. So you usually have an MSA in place and that sort of thing.

But it does take time, approvals internally and that sort of stuff. So it might be 3 to 6 months to do a contract from proof of concept when it actually gets signed with that scale. Maybe it's 2 months on the short end. And then there's a lag between -- in this particular case, since we had to get some new space before we get revenue. But the pipeline is informed from existing customers, new customers, referrals from our channel partners, also from some of the partners we have on the technology side like NVIDIA.

That's great context. I want to have a discussion on what it takes for the build-out of the AI inference opportunity. So we have about 20 locations today supporting Akamai Inference Cloud. What is the build-out of colocation facilities look like over the next 12 months? And then maybe looking beyond that?

Yes. So in some of the -- like right now, the 20 locations we have were all existing facilities, and we just can roll out racks of GPUs. So all of the sites we were in had all the necessary cooling requirements and things like that. So that wasn't a problem.

With certain clusters, if someone who says to, "Hey, I want 1,000 GPU." we may not have space in the particular areas where we need to add on to an existing site or maybe deploy into a new site. So there'll be a mix of rolling out into existing sites where we have capacity. And a lot of our -- the way we're building our CPU or our compute centers is essentially you look for our bigger sites, the 41 sites or so are generally 5 megawatts typically the initial size. I may only be using 1 or 2 megawatts and I'll roll up to 5, but I'll usually have an option to go to 5, 10 or 15 depending on the type of relationship.

A lot of the folks we work with will be doing a multiyear build-out. So there might be campus one is available today, and then we have an option on campus to that might be available next year. So there will be continuous build-out of capacity.

Now in terms of like if you want to think about unit economics, I've given us-- a little bit of this on the call. So if you think about just GPUs in general for 1,000 GPUs, we'll stay in base 10, it makes it easy. It requires about a megawatt of power. It's about 600 kilowatts just for the GPU, but with all the other stuff, it's about, call it, a megawatt power, okay? So a megawatt of power in the U.S., you can get it for anywhere from $2 million to $4 million a year, typically. California is a little bit more expensive, maybe $6 million. But generally speaking, somewhere in that range.

For the CapEx, it's funny. I did this the other day. I did -- I'd like to find public references. So I went to ChatGPT and said, "How much would a 1,000 GPUs cost with all the servers and switches and all this other stuff." And it gives you an interesting answer. It comes up with anywhere from, say, $12 million to $16 million. So let's round that up, we'll call it, $20 million on the high side. The rental market today is around -- I think we launched on Saturday $250 an hour. Okay. $250 an hour for 1,000 GPUs for this year would be $22 million roughly. $2 is around 17.5, okay? Let's take that number divided in half year, okay? So if you think about the unit economics, I've got high end, $20 million of CapEx. I appreciate that over 6 years, it's roughly $3 million, $3.5 million of depreciation, a couple $2 million to $3 million of co-location costs. And even if I get half of that price, it's still very good unit economics for me. The operating margin is in the 60% -- 40% range or better. $2 is phenomenal.

I had a whole section unit economics, but you got in front of me and that's super interesting. In terms of -- on the theme of the build-out of the AI inference opportunity, how do we think about the existing network and growing to that? So whether it's the roughly 50 Linode reasons that you have the broader 4,000-plus points of presence, is that going to be leveraged in terms of this build-out? Or are you building out like net new or renting out net new cohorts?

Yes. So there'll be a combination, I'd say, the hierarchical compute capabilities, functions as a service available in all of our servers. That's sort of serverless request based edge Java web assembly type programming for basic things. the container service, we actually have available in all of our locations as well. Limited amount of capacity, obviously. And some of our locations are pretty robust, where you might have -- we call them our multilink e-cores is the term we use internally, which effectively is a big server farm where you have most of your big telcos all connecting at once. And so you've got maybe several hundred to maybe 1,000 servers or something like that in those locations. And there, you've got more power, more space where you could run your GPUs in those locations. So we could leverage those and roll in some racks of servers and get maybe 100 locations or something like that.

But there will be purpose-built locations for customers that have -- that when the math makes sense, where they have, say, a cluster that they need in a particular area, we'll purpose build for that. So it'll be a mix of leveraging some existing sites. But some of the sites are like in telco hotels. We don't have the power and the cooling necessary to run even a compute server, the 1U, 2U rackmount servers you use for CDM, pretty cheap. They can -- they don't require a lot of power, they don't require a lot of cooling. So you can run those in pretty small locations. But for the -- even a compute server is much more robust for CPU that you wouldn't run them there.

So -- and also, I don't know that the economics would necessarily make sense for somebody to run GPUs in 4,000 locations. The math probably wouldn't work at that point. But it's not saying that we couldn't go a lot more distributed than say, 100, but practically speaking, probably 100 locations for GPUs at some point makes sense. We'll probably grow from 20 today to somewhere 20 to 40, something like that. But a lot of this can be informed by customer demand will be the main driver of why we pick additional locations.

And just as a follow-up there, I think the team mentioned that the initial capacity for the 20 has already been contracted out on the 20 existing location. How do we think about where do we stand in terms of the utilization potential and potentially going denser in those 20.

Yes. So that's not -- we didn't spend a heck of a lot of money there. There wasn't a huge revenue opportunity.

Just sort of proving out the concept out?

Yes, it was proving the concept out and we do have a sliver of that we use for proof of concept. So like I said, the typical sales motion is someone comes to and says, "I want to run a proof of concept." And the proof of concept might be running in a couple of locations, using a few GPUs or whatever, that we do on a regular basis. So we do have some of that in terms of being sold out, that is set aside for proof of concepts. But it wasn't a huge revenue opportunity. The bigger revenue opportunity is this initial buy of $250 million of CapEx. Only a portion of that goes to this big customer. There's still thousands of GPUs to sell there. So there's an enormous amount of revenue to come with that as well.

It's also the basis of my next follow-up is the timing of these purchases on the GPU, CPU side, does that happen once a contract has been signed? Or are you purchasing ahead of that, anticipating future demand?

I mean I would love to be in the situation I was just in where I had a big customer willing to -- if I didn't have the chip inflation, that probably would have covered practically all my CapEx, right? That's an awesome position to be in. It will be informed by the pipeline, growth of existing customers and demand. We will invest ahead of time like we do, but it will be informed based on what we're seeing in the pipeline and what customers are doing.

Awesome. And I want to talk -- spend the next couple of minutes just talking about some of the debates coming out of Q4, which -- I mean the results in Q4 were really strong. You have a CAS business that's accelerating, Global Cloud business accelerating $200 million customer Inference service. One of the things, though, is that margins did come down to 26% to 28% from about 29% in 2025. As CapEx steps up -- I think to the range you provided was 23% to 26% versus 19% of revenue in 2025. So with the full understanding that growth doesn't come free, and we need to fund this investment. How should we think about the margin trajectory from here as the Inference Cloud business scales?

Yes. So if you look at -- we gave this disclosure on the call. The -- where we're seeing the pressure points on margin are in cost of goods sold primarily around colocation and then depreciation. So if you look at that, almost the entire amount is those two items. And with colocation, there's two factors. One is we are buying ahead of time so as I deploy these thousands of GPUs with this latest purchase, well, I haven't sold them yet. So I don't have any revenue against it. Even with this big customer, I'm going to have that facility up and running for several months before it ramps fully up with revenue. So there's the timing aspect.

And that goes for my entire compute platform, right? So I've got capacity. Any time I have capacity available I have some inefficiency in my math on [ colo ], right, which you'll always have that.

The other thing is, as I talked about, how we buy some of the bigger locations, we'll work with our vendors for 7-year deals, 10-year deals, in some cases, and there, you're making a fairly good-sized financial commitment saying, okay, I know I'm going to be using 5 megawatts over the 10 years, and I want expandability, but I want unit economics today at that purchasing power. So I might make a, say, a $50 million commitment over a 10-year period and maybe I'm only spending $1 million in cash. But with the lovely lease accounting standard, you need to put an asset and a liability up. And whenever you have a lease that has either free rent or have escalations or commitments, you need to straight-line that.

So there is some element of noncash colocation costs in my cost of goods sold. So there's some inefficiency there. So that's part of what's driving it. But that said, there's a trade-off between growth and margin. So if I want to grow faster, let's say a customer came to me and said, I want to spend $500 million with you, whatever, pick a number. Well, my margins are going to go down a touch before because my payback on my servers might be two years or whatever. I have to get colocation because a deal like that, I don't just have that sitting around necessarily, unless they want to distribute it everywhere. And perfect locations and fine everywhere that I have capacity that's not reality. It's probably going to be in several locations might be a U.S. opportunity or Europe opportunity or whatever. So there's going to be some inefficiency associated with that.

Where we get good operating leverage is on the go-to-market side, on the engineering side and on the operations side. And we have some overlay sales for compute. It's a relatively small group compared to the size of our sales force. The people that are going out and getting the colocation deals and managing the operations associated with that, doing the deployments, buying testing the hardware and the different routers and switches and chips and all that stuff. The same group that does that for CDN and CDN basically runs our security on the same platform. So I get an enormous amount of leverage there. We've moved about 1,000 engineers out of our CDN business into our compute business because there's a lot of similar things that you do in terms of [ your ] engineering and running the distributed platform. So we will get a lot more leverage on the operations side over time.

Maybe to look at the unit economics question from the lens of a customer so sort of lifetime value of a customer, let's say, it's maybe not this particular $200 million customer, but you signed similar sized deals with commitments for 4- to 5-year ranges, what do you expect the contribution margin to like over the lifetime of that customer commitment? And I was sort of referring to, I think, a couple quarters ago, you mentioned that you think that maybe -- the Inference Cloud business could get back to kind of where your public cloud gross margins are.

Yes. I mean just if you go through the numbers we just talked about, let's say you get per 1,000, let's say you're getting $12 million to $15 million of revenue against cost of, say, $3 million to $4 million for colocation to $3 million of depreciation. Your operating margin is north of 30%. Your gross margins are in the 70-ish percent. So you can get there certainly and that's assuming a discount off of what the published price for rentals are.

Now the other misnomer I think that's out there is people believe, well, 6 years depreciation, that's probably not the right way to think about it because what about the next generation. Well, sure, there'll be a next generation in that 6 years, probably many next generations. That doesn't necessarily mean that everyone is going to move to that next generation because you've done -- like one of the things I'm finding is customers will come to us and say, my particular application runs best on this chipset, maybe it's AMD, maybe it's NVIDIA, maybe it's Intel. And I also have a particular workload in a set of economics that I might not be able to afford the next generation of chips. So that may not make sense.

And I look to our Linode business. Before we bought Linode, they offer GPU as a service and it's been a nice little business for them, and we're selling generations that were put in place before we even bought Linode. We have grown Linode now for 4 years.

So there's -- I anticipate -- I mean, I could be wrong on this, but I don't think I am, there will be a market. Now maybe the unit price has come down a little bit. So how you get good cost recovery of say, a year to two on my service. And the same is true with CPU. You -- kind of the dollar CapEx dollar revenue gets you above a year or so. Even if it's $0.85, you still get somewhere between a year or two of -- maybe it's 18 months of recovery on your CapEx.

So it's a great business. I'm not too concerned about -- most customers would probably renew unless their application goes away, but generally speaking, it's tied to a business that has some notion of scale with it as well that you might say, "Hey, I want to start off with x amount because I'm doing an application that is tied to something else I'm doing that's revenue generating. And as long as that's growing, I'm going to continue to need more and more. And we do see that. We see very nice, call it, same-store sales growth of folks that are using us. So that's very true with the public clouds.

Maybe last question [indiscernible] time about security in the delivery business. So I'll kind of bundle the question. Your confidence in sustaining the growth that you have in [indiscernible] security, was 9% constant currency. I think you guys guided to high single digits, your confidence in stating that level of growth in security? And then the delivery side of the equation, it does seem like the pricing environment is a little better. Fewer competitors out there, we got Olympics, we got the World Cup. And how should we think about the growth trajectory and delivery over the next year?

Sure. Delivery has gotten a bit better. One of the things that's a little difficult to digest this year as we had acquired the assets of NGL. So throughout the year, we had modeled it that some of these customers, some of the bigger customers were multi-CDN and we bought the other CDN. So they have to rebalance. So let's say we have somebody who's doing 50% with us, 50% with Edgio. We combine it, we probably end up with 75% share as a general statement. So 25% goes to fight somewhere else. That happens throughout the year. It's not all like I closed on December 20 and December 21, everything rebalances. It happens gradually throughout the year. So you did see revenue declining. And part of that was like modeled and planned, we knew that was going to happen. Someone else was going to pick up some share where they didn't have share before.

Now we are seeing the pricing environment get better. And we're taking a much tougher stance with pricing now saying, look, we just talked about on our fourth quarter call, $200 million of additional CapEx just because memory prices went up, right, colocation is not getting cheaper. In some markets, it's solid, but other markets more expensive. Labor doesn't get cheaper. We can no longer eat all of this. We have to start passing along some of that price. And certainly, not accepting a 15% or 20% price decline in the delivery business. So we are taking a much tougher stance. We are trying to raise price on renewal. We'll be successful in some areas and others we may not be. But that's a much different stance from us. We've never raised price before. We're doing it for the first time in our history.

On security, I gave out some really interesting statistics on the fast-growing parts of the business. API security exit over a $100 million run rate, growing at over 100%. The penetration rate in the installed base is sub-10%. Very applicable for our WAF customers to use API security. So that's going to look -- enormous runway. And I actually think in the agentic web, as we talked about earlier, with all the data flows that go with that more robust application that's going to be a massive attack vector. So we think there could be even a second generation of growth out of API security, protecting those agentic applications in the future.

In terms of Guardicore, again, sub-10% penetration of the base, the majority of that business is coming from new customers, a very robust channel there. Very great dynamics with existing customers. The typical motion is to add more agents to protect a greater slew of your network. So you start off running maybe 1/4 of your network and then you expand our license business, which isn't a significant business, but has 90x percent renewal rates. Usually, it's an upsell. They're term licenses. So very healthy dynamics in that part of the security business that was growing at over 30% last year.

Well, we're out of time, Ed. Thank you so much for giving the update on the Akamai story. A lot of great things happening in the business and all the best for 2026.

All right. Thank you, Sanjit. Appreciate it.

All right. Good morning. Thanks, everybody, for being here. Last day of the conference. My name is Frank Louthan. I'm the senior telecom analyst here at Raymond James. We're very pleased to have Akamai Founder and CEO, Tom Leighton, here with us. We're going to start out with a few questions and then we'll save some time at the end for you.

But Tom, maybe kind of walk us through the Akamai story, kind of where do you fit into the space and what you do to kind of set the stage for everybody.

Sure. Most of our revenue is in security, growing at about 10%, market-leading products for web app firewall, stopping DDoS attacks, bot management and more recently, API security and Guardicore Segmentation growing very fast. So in Q4, they delivered $90 million, growing at 35%. And so they're driving a lot of our future growth. A lot of investment around AI-related kinds of security as enterprises adopt more use of AI. That's a big new attack surface, so developing the capabilities there.

Fastest-growing product area is our cloud infrastructure services, finished $94 million in Q4, up 45% year-over-year. Even more exciting, we think that will accelerate through this year. So we're calling for 45% to 50% growth in revenue, really, really exciting. A lot of big name companies using our cloud infrastructure services and including all the hyperscalers, which is a pretty cool validation. And of course, we operate the world's largest content delivery network, the largest and most reliable and scalable by a good margin.

All right. Great. So you mentioned the compute platform. Talk to us about how that's kind of changed the business. What are customers looking for when they come to you for compute? And how does that kind of change your go-to-market approach with customers?

Yes. What we're trying to do for compute is the same thing we did for delivery when we created the content delivery marketplace. And then with security, created -- security is a cloud service, WAF operated in over 4,000 POPs and 700 cities, so we can stop all those attacks before they get anywhere near the data center. By being close to users, it's better performance, more scalable, more reliable. And now it's the same story for compute. We can get our customers' compute logic close to users. So it's faster, very scalable and at a very good price point. So it's very competitive in the marketplace.

Yes. Closing that distance comes up all the time with investors, the concept of just needing compute nodes near population densities and so forth. So with that, you've got your inference cloud product that you're launching. Talk to us about that. How does that fit in with the compute platform? And what can investors expect from that business?

Yes. As we look to the future, a lot of the use for compute is going to be AI inference agents doing things on behalf of enterprises and users. And a lot of the applications people are developing, you want to have low latency. Now historically, that was hard to do with AI because the models were slow. And if it took you a few seconds for the model to generate a response, well, it didn't really matter if you were close.

But now with the latest generation of hardware, and we're deploying a lot of the new Blackwell 6000s. They are fast in response. You can generate videos pretty much on the fly. And so now the latency means something. Not only that as the web gets more of these agents and apps deployed, it's going to be more use of video. And that's bit intensive. And so even if you didn't care about the latency, which you do, you have a problem with bandwidth coming out of a data center.

And so it's not possible, for example, to have millions of concurrent sessions with users that are video-based with personalized video coming from a data center. Same thing just like with -- from streaming big sporting events. You don't do that from a data center or 2. You've got to do that in a distributed way. And so Akamai, we have this fabulous unique distributed platform, perfect for AI inference in the future.

So you've got a lot of the critical components here. So when we look at AI, where do you expect to have the most revenue impact? Is it going to be -- you got delivering traffic, AI bot mitigation, compute inferencing at the edge. Where do you think you're going to see the most impact from AI on all the different parts of your business?

First compute, then security, then delivery. Compute because that's where the action is with AI, you have all these individualized responses. Security in a variety of ways. First, the models are enabling the attackers to do more penetrations and be more capable, so you have more need for security services, especially Guardicore Segmentation. Second, as enterprises use AI in more places, that's a whole new attack surface and you need special defenses, which we supply with our security business.

Also with the bot management or the agent management, you want tailored individualized responses based on who the agent is and what it's doing. And that's what our bot management solution does. And so that's very helpful. And then third is delivery. Now with delivery, the revenue generally is based on traffic, which is generally gigabytes delivered. Now the big sources of traffic on the Internet are video and big software downloads, so a big gaming release.

A video is orders of magnitude more traffic than an image. And an image is orders of magnitude more traffic than a text, an interaction buying a sweater. And so the agents themselves aren't generating a lot of traffic per se. There can be a lot of hits, but not a lot of traffic. Where you will see the traffic generated, I think, as more of the applications on websites become video-based and more of your interaction with the web is video. You're talking to your shopping helper. And it's a person who you know and it's a video of a person, and you're getting real-time responses.

And then your personal shopper says, hey, look at this sweater, here's a video of you wearing the sweater at a dinner part, okay? And so it's a personalized video. That will generate more traffic. That said, the value in terms of revenue is more for the generation of that video than the transmission of it. So compute would have more revenue benefit for it.

Okay. And so if we -- let's look at this as customers move to more to block it from optimize it on the CDN side and security, how are you positioned to kind of benefit that as you're -- because with all of this traffic, there's some of this traffic that you want to get through and some that you don't. And so how are you -- on the security side and the delivery side, how are you able to get some traction there?

Yes. We've been doing that since before AI. There's all sorts of bots out there and agents doing different things. There's been scraper bots out there for, well, a decade, doing all different kinds of things for scraping. Now there's a new scraper bot doing it to train models for AI. But there's been just a whole plethora of different kinds of bots doing different things, and our bot management solution gives a differentiated response that our customers want to give.

It could be a partner bot, you want to give a good service. It could be a Google search bot, you want to give it a version of your site with all the right keywords and make it fast. It could be a price scraper bot, you want to give it false prices because it's your enemy trying to undercut you by $0.01. Could be a bot trying to fill up all the seats on your plane so that you'll buy the competitors' seats. And so we've been dealing with that for a decade. So there's nothing really new there. There are the AI bots now, and that traffic has increased, particularly the scrapers so far. And so we just continue to give whatever service -- differentiated service our customers want to give.

Yes. Okay. So talk to us about the compute platform. I get this question a lot from investors, and I think from a lot of generalists and so forth. So how is your compute platform differentiated from your competitors? And what is it about customers using your platform versus kind of the house names and the hyperscale world of large compute companies that we all know. What is it about your business that attracts customers?

We're more distributed. So we're closer to the user, gives you better performance. We're extremely reliable. I think that's really important. And we're very competitively priced. And I think a great proof point is all the hyperscalers use our compute platform to do mission-critical things. And it's not like they don't have their own big cloud platforms that they can use, but they'll get better performance for particular applications where that matters using Akamai.

What's an example of some of those applications? Because -- and just to get like a real world something that you're able to do maybe an application or a product that everybody is familiar with that your platform is much better suited for than just the generic thing on one of the large hypescalers.

Right. Well, one of the hyperscalers uses us for live video. So if you watch a lot of sports online, you're actually using our compute platform. And they chose us over their own capabilities, again, because we're close, so they can tailor the event that you're seeing to give the best possible viewing and to synchronize it. So everybody sees the action at exactly the same time, which could be useful if there's betting applications.

Another hyperscaler uses us for ad selection, which, again, you want to do really fast, and it's better if you can do that locally. Of course, we have a lot of commerce companies and speed makes a lot of difference there. And as you adopt AI, you're going to get a much richer experience, translates into a lot higher conversion rate and performance matters a lot. Reliability matters a lot there. One of our very large new customers has capabilities to manage fleets of robots, cars, automated kinds of things. Again, you want that to be as real time as possible.

Okay. Great. And so then on the distribution side, talk to us a little bit about the status of sort of self-provisioning for CDN and how do you maintain relevance with all the large media companies and so forth. We've seen some M&A in that space recently with Warner Bros. How does that impact you? And talk to us about the supply for that.

Yes, I don't think there's any impact from the acquisition. Both sides use us extensively already today. There's a couple of exceptions, big media companies that do it all in-house. But otherwise, most of the rest primarily have most of the traffic with us. And most of them do some kind of load balancing among CDNs. But in generally, situations we'll have a majority share.

Okay. And that business trended pretty well for the industry last year. Pricing was a little bit better, traffic better than folks thought. What is the longevity of that? Do we expect -- is that -- have we kind of reached a plateau for that sort of business? How do you -- what's sort of the outlook for that going forward?

Yes. We've guided this year mid-single-digit decline in the delivery business. Revenue is the combination of traffic growth and per unit pricing decline. We're continuing to be very diligent about the pricing we offer. There is business that we don't take and share we don't take. And some of the big spiky events, we won't do if the economics aren't right. So we, I think, are seeing better traffic the last year or so and looks like that trend should continue. And we're being a lot -- very diligent on the pricing. And so that helps us.

There's some revenue we let go somewhere else, but it helps our business and certainly the profitability of that delivery business.

Are you seeing any aspect of that business, especially with AI traffic and so forth, where customers are willing to -- are seeing value to using you and paying more? Are we seeing some stability of that? I've seen that in some other areas of my coverage where the hyperscale customers want -- they want execution, and they want it as quick as possible, and they're willing to pay better margins to some infrastructure companies and so forth. Are you seeing the same thing for some delivery applications and maybe that's getting a little bit extra life?

Yes. We generally are paid more than the competition, depending on the particular product and the application could be a lot more to a little more because we offer better performance, and we are a lot more reliable.

All right. So let's talk a little bit about security on that side of the business. How is pricing holding up on security? How should we think about that for the...

Pricing is holding up well there. We have the market-leading solutions by a good margin. And in some cases, this year, we'll be increasing pricing. As you know, the cost of memory has gone up quite a lot in the last few months. And so some of that will be passing on to customers.

So how does that cost of memory affect your business? Talk to us about the capital investment you have to make on the security side. And is that an impediment to your growth? Are you seeing any issues with your ability to get the memory? Or is -- how does that...

No. It's not an impediment to growth. We can get everything that we need. We're a big buyer in the marketplace, but it's more expensive. So we're doing a lot internally to buy less than we might have otherwise. A lot of our servers that we might have decommissioned this year, the math has changed on their useful life. And so we're going to leave them in the field. And they'll be working longer just because the memory cost has gone up so much.

We've estimated that this year, after all the puts and takes, it will be an extra $200 million for us in the increased pricing. Usually, with these things, the supply gets increased, the capacity and the production has increased, and that will help abate the cost going forward. We're not counting on that. So we're doing everything we can to optimize, but we -- it will be an extra $200 million this year.

So when we look at that -- what is the breakdown of that $200 million from additional data center space or servers? Or what do you -- what is it...

That's purely the memory cost.

That's purely the memory cost Okay. So it's not -- you're not having to -- if you're keeping these older servers on, are you having to take on more data center space to...

No, it's just we're not taking on as much CapEx, buying less of the memory, using the stuff that a little less efficient. It's not a 6-year-old server, but working fine. And now the math has changed so that we're going to keep it in service longer.

Okay. Great. All right. So with that, thinking about the guidance for the year. So what does it take to kind of reach the high end of the guidance for this year? And let us -- what are your -- what's kind of built into the assumptions for that?

Yes. So the amount of traffic can be variable and a strong traffic year helps the delivery business. How fast we sign on the new compute business can make a difference towards the back half of the year. Security, that's a little bit more predictable. There are sometimes license deals, particularly as you get, for example, sovereignty, other issues with critical infrastructure, maybe they want it in-house, which we can do.

And then the accounting treatment is different if they take control of it versus our taking control as a cloud service. So that can swing things a little bit. Dollar fluctuations can make a little bit of a difference. If the dollar is stronger, that can depress the revenue through the conversion a little bit. Dollar weakens, the revenue goes up.

Yes. Okay. And then walk us through the path of inference cloud for the year. You're making an investment in that. I think it's roughly $100 million, something like that. Walk us through the pace of that investment. And then when do we start to see that revenue coming in? How actively are you selling it? And when will we start to see that showing up in the income statement?

Yes. We talked about a $250 million investment into inference cloud and a large purchase of the Blackwell 6000s. The initial tranche we deployed last fall in 20 cities. That actually goes GA at the end of the quarter, but it's already sold out from the beta customers. So we're deploying a much larger tranche now. And as we talked about on the call, we have a good chunk of that already committed in a 4-year deal with a large account.

And as we sell out the rest of it, which is not deployed yet, then we would add more after that. The revenue associated with the tranche that we're in the process of doing now, we'd be looking at towards the end of the year to start to get revenue. First, we got to get the servers. We've got to deploy them, get them all turned on and then used. So that's towards the end of the year, we're looking for revenue generation there. So that's really more a big impact next year.

And is this incremental data center space that you're taking on this year to deploy that? And have you had any issue -- how about how many -- roughly how many megawatts of space does that require?

Yes. So a bunch of it we already have from -- as we initially deploy, we signed long-term colo deals for increasing amounts of usage. We are adding new data center space now on top of that. So it's a blend. The typical large-sized data center for us now is 10-ish, maybe 10 to 20 megawatts that we take on as a large-sized data center for us.

Okay. And then as you -- with breaking down some of that investment, how much of that is the servers? How much of that is the data center space you're taking on to deploy this?

Yes. So that would be the CapEx side of things. The -- so the -- say, the Blackwell 6000s, associated hardware. Typical use cases are not just the 6000s, but we use actually our whole platform. And so it's the CapEx needed for that. In addition, there's colo space that, in some cases, we're already paying for or is already on our books because you linearize the counting when you take on a longer colo deal. And in some cases, it's new.

Okay. And so you mentioned there's 20 -- I think you said 20 cities that you'll be going to. The thing that, I think, is interesting for me being a telco guys is looking at the 4,600 locations you have on the delivery network and being able to -- finally being able to use that to bring this elusive concept of edge computing that has never seemed to have materialized, and you are finally in a position to do some of that. So as you look out at that network, talk to us about -- this is a question I get a lot from investors, like why couldn't they do -- how much can they deploy and where can they go?

And one of the gating factors there is the power because a lot of your locations are on telco data center facilities or telco facilities that were built for voice. And so maybe you have a little bit less power. But if you look at all those locations, how many of those can you realistically deploy? I mean you're looking at inference cloud, how many megawatts does it take to put a pod of servers out there and GPUs? How should we think about that?

Great question. And the answer is it's hierarchical and depends on what you're trying to do. So in all 4,300 locations, 700 cities, we operate function as a service, our EdgeWorkers solution. So that's everywhere, totally serverless, and that's been out there for a little while now and used actively. Then the next tier up would be our managed container service. And this is where we deploy customer containers in software into our existing hardware. So no extra deployments and stuff needed.

Now that can go into any of the 4,300 locations. We're actually using it live today. It's been between 100 to 150 cities. So not all of them, but 100 to 150 of the larger regions. And actually, one of the hyperscalers is using that capability today. And that does your container. Then stepping up another level is you have full stack compute and storage, VMs, big storage, object store. And that runs today in 36 cities, about 40 data centers. That's where we're deploying by and large, the Blackwell 6000. Now we're in 20 cities today. And the next tranche will be maybe some new locations, but mostly beefing up the existing ones for the next tranche of the 6000s.

So the critical place is having it in the city, not necessarily in all 4, 5 or 6 POPs in the city for having this deployed.

Yes. And you're right, the 6000s, they're not going into the 4,000 locations, and they're not set up for that. But they could go into 100 to 200 cities over time.

Okay. Great. That's great. All right. Well, why don't we see if we have any questions from the audience, and I got a couple more. Eric, go ahead.

Who are the buyers of the new full stack that you're building out?

Usually, it goes to the CIO. You mean by the function -- job function or...

The kind of organization. Are those hyperscalers or more enterprise buyers?

Enterprise buyers, including the hyperscaler companies. But yes, the major enterprises for us initially was big media. Now commerce is heavily engaged. And our biggest customer is industrial.

Did you mind a follow-up?

Yes, go ahead.

And how do you normally price these? Are these 1-year deals? Are they -- if they're longer, are there escalators for things like memory...

Again, it's hierarchical depending on what you want to do. So you can buy product-led motion on the website. It's $2.5 for a VM hour, and you buy 1 hour, if you want. Then as the customers get larger, now you're probably in a sales-led motion. Often there'll be a commit to a certain amount of usage over a period of a year, 2 years, 3 years and the reps comped for longer-term deals. And what's new for us now is you can buy clusters, and that would be multiyear commits. And we talked about our first customer in that motion that with a 4-year $200 million commit, and they're basically buying a cluster of the GPUs.

And what does the revenue kind of breakdown look like that at the low end of a cluster? What are we talking about as far as monthly recurring revenue, just to get an idea?

Well, it depends on the size for the monthly recurring revenue. But if you buy at list, $2.5 oer VM hour, obviously, if you're buying a cluster for 4 years, you're going to pay a lower rate than that.

Yes. Okay. Great. All right. Any other questions from the audience? All right. Okay. So talk to us about capital allocation priorities for the next 12 months. How should we think about that?

Pretty much where we've been all along, I would say. We buy back the equity generally that we distribute to employees. We opportunistically buy back a little bit more on average, maybe 1% of the equity outstanding a year. Last year, we bought back more, more than we'd ever done before, about $800 million. Part of that was we did a convert and as part of that, bought back some as part of closing the convert. But our strategy is the same. We use the capital for M&A. We also use it, obviously, for CapEx, which is part of the operating of the business. But there's no intended shift in terms of our use of capital and buying back stock.

So -- and then talk to us about M&A. You bought quite a few different companies. Usually, it's -- you're buying a product, you're buying a capability, it's more of a buy versus build strategy. You're clearly -- you're building the inference cloud yourself. What kind of -- where do you see opportunities for M&A? Is it in security? Is it in the compute? Where you think the things to plug in some platforms?

Yes, we're looking in both areas. Security, very fast-moving landscape. So obviously, a lot of interest there. We've been very happy with our major acquisitions there with no name for API security and Guardicore for micro segmentation. There's other areas we're looking at. We're also -- we're very disciplined. So we're not going to pay a crazy amount of money for something. It has to be something that will return real value to shareholders. And on the compute side, we're looking at ways that we can enhance our capabilities there. It's not quite the same thing as security doing a lot of the compute in-house.

Does anything stand out to you on the security or the compute side to say that it's something that you would need to have or is it an area that you'd like to move into the M&A might be more of a possibility?

Generally, it would be some kind of product adjacency. So it fits within our current product set, it makes sense to enhance our security platform play, something that we think our reps can sell that we know the buyer because it's close to the buyer of some of our other security products. So having it fit with Akamai is the kind of thing that we'll be looking at.

Okay. Great. Anybody else got a question in the audience? All right. So kind of to wrap up here, talk to us a little bit about the company. I've been asking all my companies this question, sort of what's one big sort of misperception about the business? I think the obvious one for you guys is you're not a CDN necessarily anymore. That's part of the business, but that's not really what you are, even though you're often thought of that. So maybe just to level set take that one off the table, what is sort of -- when you talk to investors and you're talking to customers, what's the biggest misperception about Akamai that you'd like to kind of address and set the stage for?

Well, I think the biggest change that investors are starting to understand is that, yes, we are a cloud company and have a strong capability that's accelerating at a very fast rate. I think cloud has been sort of a show-me story for investors, and I think we're showing them. We're getting very fast growth on a meaningful number now, signing up some impressive enterprises, including, I don't know any cloud company that has all the hyperscalers as customers of their cloud business. And I think that's a good proof point that if they're buying our cloud services, there's something there, really does give better performance. And for non-cloud companies that don't have their own cloud, it is very competitively priced.

Okay. All right. Great. Well, Tom, thank you very much for being here. Appreciate everybody coming. We've got a breakout session after this. If anybody is interested, feel free to join us. Thank you very much, Tom.

Great.

Good day, and welcome to the Q4 2025 Akamai Technologies, Inc. Earnings Conference Call. [Operator Instructions] Please note that today's event is being recorded.

I would now like to turn the conference over to Mark Stoutenberg, Head of Investor Relations. Please go ahead, sir.

Good afternoon, everyone, and thank you for joining Akamai's Fourth Quarter 2025 Earnings Call. Speaking today will be Tom Leighton, Akamai's Chief Executive Officer; and Ed McGowan, Akamai's Chief Financial Officer.

Please note that today's comments include forward-looking statements, including those regarding revenue and earnings guidance. These forward-looking statements are based on current expectations and assumptions that are subject to certain risks and uncertainties and involve a number of factors that could cause actual results to differ materially from those expressed or implied. The factors include, but are not limited to, any impact from macroeconomic trends, the integration of any acquisition, geopolitical developments and other risk factors identified in our filings with the SEC. The statements included on today's call represent the company's views on February 19, 2026, and we assume no obligation to update any forward-looking statements.

As a reminder, we will be referring to certain non-GAAP financial metrics during today's call. A detailed reconciliation of GAAP to non-GAAP metrics can be found under the financial portion of the Investor Relations section of akamai.com.

With that, I'll now hand the call off to our CEO, Dr. Tom Leighton.

Thanks, Mark. I'm pleased to report that Akamai delivered strong fourth quarter results as we continue to make major progress in positioning Akamai for the future. Revenue grew to $1.095 billion, up 7% year-over-year as reported and up 6% in constant currency. Non-GAAP operating margin was 29% and non-GAAP earnings per share was $1.84, up 11% year-over-year as reported and in constant currency.

Q4 revenue for Cloud Infrastructure Services, or CIS, was $94 million, up 45% year-over-year as reported and up 44% in constant currency. That's an acceleration from the 39% growth rate we achieved in Q3. The rapid growth was broad-based within CIS, driven by our ISV solutions, by Infrastructure as a Service and storage customers and by customers leveraging edge workers and web assembly, which offer improved performance and lower cost for edge native applications.

In each of these areas, we're starting to benefit from AI-related tailwinds as customers make greater use of AI applications and agents across their businesses. Last quarter, Akamai took a major step towards the future with the launch of Akamai Inference Cloud, our platform to support the growing demand to scale AI inference on the Internet. Akamai's architecture uniquely positions us to power and protect AI the way we power and protect the web by bringing AI physically close to users enabling the faster performance and global scale needed to unlock AI's full potential.

We believe the AI market is entering a critical transition point, the first inning of a long game to come, where inference or the execution of queries against a trained model is the new frontier. This requires purpose-built infrastructure to enable distributed low-latency, globally scalable AI at the edge with response times measured in a few tens of milliseconds.

Akamai Inference Cloud does just that by incorporating NVIDIA Blackwell GPUs into Akamai's distributed cloud infrastructure with its unparalleled global reach and security at the edge. This enables intelligence to run instantly, securely and exactly where it's needed, right next to the user, agent or device. As evidence of our strong momentum, we're delighted to announce that we recently signed a 4-year $200 million commitment for our Cloud Infrastructure Services with a major U.S. tech company at the forefront of the AI revolution. I've had the privilege to work at Akamai for many years, and I have to say that it's really exciting to see such a pivotal player in the AI ecosystem choosing Akamai Inference Cloud for such a large AI use case.

We also signed many other new and expanded contracts for our Cloud Infrastructure Services in Q4. An AI chatbot platform based in India signed a 3-year contract for our IAS and enhanced compute support solutions and save 45% on compute costs they would have paid to a hyperscaler. A very well-known antivirus software company chose Akamai's cloud for their VPN service, telling us they liked our performance and support better than what they previously got from 2 of our cloud competitors.

A leading social networking platform that was using us on a pay-as-you-go basis committed to consolidate their multi-vendor stack on to Akamai's cloud platform, providing us with another takeaway from a hyperscaler. Two [indiscernible] tech companies in China chose us for our significantly lower latency and dramatically reduced egress costs. And one of the world's largest retail companies expanded their use of our edge compute platform to improve their digital shopping experience and increase conversion rates.

As a result of the strong customer demand that we're seeing and the strong AI tailwinds across the marketplace, we anticipate that the very rapid growth rate for our Cloud Infrastructure Services will accelerate further in 2026.

Our security solutions also performed well in Q4, led by continued strong demand for our market-leading API security and Guardicore Segmentation solutions. Revenue from these high-growth security products grew 36% year-over-year as reported and 34% in constant currency. Last month, Akamai was recognized as a Customer's Choice for Network Security Microsegmentation in the Gartner Peer Insights report for 2026. Akamai earned a 99% recommendation rate, scoring above market norms for both user adoption and overall experience.

Last quarter, we saw continued strong demand for our Guardicore Segmentation platform with both new and existing customers. One of North America's largest financial institutions purchased our segmentation solution to gain visibility and protection across all of their network assets as part of a 4-year $40 million contract. South Korea's largest mobile operator selected Akamai following the well-publicized BPF door security incident, which exposed gaps in East West Security and Zero Trust maturity. The customer chose our solution for workload level segmentation, deep visibility and resilient enforcement across hybrid environments.

We also signed deals for segmentation in Q4 with one of the largest carriers in the U.K., a major branch of the U.S. armed services and multinational banks in North and South America and Scandinavia. In Q4, we also saw increased demand for our API security solution, signing new customers across multiple verticals including financial services, technology, health care, real estate, retail and travel. Customers who chose Akamai API security in Q4 included a major European automaker, a telco in the Middle East as well as airlines serving Asia Pacific and Latin America.

We also signed a 5-year $47 million commitment from one of the largest hardware companies in the world, in a contract that included API Security and cloud Infrastructure Services Along with other Akamai offerings. We had many other customers in Q4 who purchased multiple security products across our portfolio including one of Asia's largest airlines, which signed a $10 million contract for multilayer protection over 5 years and a 3-year $45 million renewal with one of the world's largest financial institutions, to migrate nearly 100 critical applications away from hyperscaler security and onto the Akamai platform to ensure best-in-class DDoS and web application protection, high availability, and robust security support from Akamai Security Operations Command Center.

Earning the trust of customers is imperative for Akamai. The world's biggest brands trust us to keep their apps performing well even under peak traffic conditions. They trust us to protect them from myriad attacks and to keep their data safe. And they trust us for our reliability. We saw how much this trust matter to customers who relied on us during the recent holiday season, a time when one of our competitors took down their customers with multiple multi-hour outages. Major enterprises know who they can trust, and we're grateful for the trust that our customers place in Akamai.

Last quarter, we were honored to be named by Forbes in their list of America's Most Trusted Companies and in their list of America's Best Companies for 2026. Forbes analyzed thousands of the largest public and private companies in the U.S. across 11 dimensions, including financial performance, customer sentiment, employee ratings, reputation for innovation, executive leadership, cybersecurity and sustainability. We were also honored by The Wall Street Journal, naming Akamai to its list of America's best managed companies, The Management Top 250.

This ranking by the Drucker Institute analyzed publicly traded companies based on customer satisfaction, innovation, financial strength, social responsibility and employee engagement and development.

Before I hand off to Ed, I want to thank our employees and our management team for their achievements in 2025. Together, we're successfully executing on our ongoing transformation of Akamai into the cybersecurity and cloud company that powers and protects business online. We believe that the investments we're making today are enabling Akamai to do for cloud and AI, what we've done for security and CDN and enabling Akamai to grow even faster as a result.

Now I'll turn the call over to Ed to say more about our results and our outlook for Q1 and the year. Ed?

Thank you, Tom. I'm pleased to report that we delivered excellent fourth quarter results with total revenue of $1.095 billion, up 7% year-over-year as reported and up 6% in constant currency. We also delivered strong bottom line results with non-GAAP EPS of $1.84, up 11% year-over-year as reported and in constant currency.

Moving now to revenue. Compute revenue, which is comprised of the high-growth Cloud Infrastructure Services or CIS solutions and our Other Cloud Applications, or OCA, was $191 million, up 14% year-over-year as reported and in constant currency. For Q4, CIS revenue was $94 million, accelerating the 45% growth year-over-year as reported and 44% in constant currency, a nice jump from 39% growth last quarter. CIS now represents approximately 50% of total compute revenue.

Moving to security. Revenue was $592 million, up 11% year-over-year as reported and 9% in constant currency. Revenue from API Security and Zero Trust Enterprise Security combined was $90 million, an increase of 36% year-over-year and 34% in constant currency. Notably, API Security grew by more than 100% year-over-year, exiting the year with a revenue run rate exceeding $100 million. Security revenue was driven by strength of our high-growth product suites and a favorable tailwind from term license revenue.

For the fourth quarter, license revenue rose to $18 million, up from $12 million in the same period last year. As a reminder, our term license agreements are generally for one to 3 years and we continue to maintain exceptionally high renewal rates in our term license business.

Moving to delivery. Revenue was $311 million, down 2% year-over-year as reported and down 3% in constant currency. These results highlight the continued steadying trends we have seen in our delivery business throughout 2025. International revenue was $542 million, up 11% year-over-year or up 8% in constant currency, representing 50% of total revenue in Q4. U.S. foreign exchange fluctuations had a negative impact on revenue of $5 million on a sequential basis and a $12 million positive impact on a year-over-year basis.

Moving to profitability. In Q4, we generated non-GAAP net income of $270 million or $1.84 of earnings per diluted share, up 11% year-over-year as reported and in constant currency. This better-than-expected performance was primarily driven by higher-than-expected top line revenue in the fourth quarter. Finally, our Q4 CapEx was $154 million or 14% of revenue.

Moving to cash in our capital allocation strategy. As of December 31, our cash, cash equivalents and marketable securities totaled approximately $1.9 billion. During the fourth quarter, we did not repurchase any shares. For the full year 2025, we spent $800 million to buy back approximately 10 million shares, marking the largest annual buyback in our history. As it relates to the use of capital, our intentions remain the same, to continue buying back shares over time, to offset dilution from employee equity programs and to be opportunistic in both M&A and share repurchases.

Now before I provide Q1 and full year 2026 guidance, I want to touch on some housekeeping items. First, as Tom pointed out, we recently signed our largest compute customer contract. We're very excited that this technology company has committed to a minimum 4-year spend of approximately $200 million on our Cloud Infrastructure Services with a large majority of that spend for our AI Inference Cloud. We expect to start recognizing revenue from this contract in the fourth quarter of 2026.

Second, to capitalize on this transaction and with growing AI inference cloud pipeline, we intend to invest approximately $250 million of CapEx this year to augment our AI Inference Cloud.

Third, we have recently observed significant inflationary pressure within the computer hardware market due to unprecedented industry investment in AI, specifically, we're seeing a dramatic increase in the price of memory chips, which is driving up the cost of servers. This supply constraint has necessitated an upward adjustment to our CapEx forecast of approximately $200 million for 2026.

Next, I want to remind you some typical seasonality we experienced in operating expenses throughout the year. First, we recently completed a targeted reduction in our workforce to better align our talent with our long-term growth priorities. While this action streamlined certain areas and reduced our OpEx, we do not anticipate generating net savings for the full year. Instead, we are reinvesting those savings directly back into the business, specifically to scale our go-to-market efforts and to support our colocation and CIS infrastructure requirements to maximize our growth opportunities. In Q4, we took a $55 million restructuring charge that was primarily comprised of severance costs and impairments of certain intangible assets.

Second, looking at the first quarter, we typically see a seasonal increase in expense. This is driven by higher payroll costs resulting from the reset of social security taxes for employees who maxed out in 2025 and stock vesting from employee equity programs, which tend to be more heavily concentrated in the first quarter.

Third, as we look to the second quarter, we expect operating expenses to remain relatively flat on a sequential basis. The savings realized from our restructuring and the roll off of the higher Q1 payroll taxes will be offset by our annual merit cycle, which takes effect on April 1.

Moving to FX. Foreign currency markets are expected to remain volatile throughout 2026. As a reminder, we have approximately $1.3 billion in revenue that is denominated in foreign currency. Largest currency exposure on revenue includes the euro, the yen and the Great British pound.

Finally, as previously noted, Cloud Infrastructure Services now accounts for approximately 50% of our total compute revenue and is growing rapidly. Recognizing CIS is the primary growth engine and a significant focus of our investments. For the compute business, we will begin reporting it as a stand-alone revenue category effective in the first quarter of 2026. For simplicity, we will consolidate delivery and other cloud apps into a single reporting category starting in Q1.

To assist with your year-over-year analysis and financial modeling, we have published 8 quarters of revenue history, for these revenue categories and supplemental schedules as part of today's reporting package on our Investor Relations website. In addition, for added transparency, we will disclose quarterly revenue for OCA independently for the remainder of 2026.

Now moving on to guidance. For the first quarter of 2026, we are projecting revenue in the range of $1.06 billion to $1.085 billion, up 4% to 7% as reported or 2% to 5% in constant currency over Q1 2025. We expect Q1 revenue to be lower sequentially from Q4, driven by the following factors. First, reduced onetime license revenue in Q1 from Q4 levels; second, 2 fewer calendar days in Q1 compared to Q4, plus 2 less days of usage revenue; and finally, less seasonal traffic in Q1 compared to Q4. The current spot rates, foreign exchange fluctuations are expected to have a positive $4 million impact on Q1 revenue compared to Q4 levels and a positive $22 million impact year-over-year. At these revenue levels, we expect cash gross margins of approximately 71% to 72%.

Q1 non-GAAP operating expenses are projected to be $339 million to $348 million. We anticipate Q1 EBITDA margin of approximately 39% to 41%. We expect non-GAAP depreciation expense of $145 million to $147 million and we expect non-GAAP operating margin of approximately 26% to 27%. With the overall revenue and spend configuration I just outlined, we expect Q1 non-GAAP EPS in the range of $1.50 to $1.67. This EPS guidance assumes taxes of $57 million to $60 million based on an estimated quarterly non-GAAP tax rate of approximately 19%. It also reflects a fully diluted share count of approximately 148 million shares.

Moving on to CapEx. The reason I highlighted earlier, we expect to spend approximately $254 million to $264 million in the first quarter. This represents approximately 23% to 25% of revenue. Looking ahead to the full year for 2026, we expect revenue of $4.4 billion to $4.55 billion, which is up 5% to 8% as reported and 4% to 7% in constant currency.

Moving on to Security. We expect Security revenue to grow in the high single digits on a constant currency basis in 2026. The Cloud Infrastructure Services or CIS, we project revenue growth to accelerate to 45% to 50% year-over-year. We expect this momentum to build throughout the second half of 2026 driven mainly by the scaling of our AI Inference Cloud business.

For delivery and other cloud apps, we expect both will decline in the mid-single digits year-over-year. Specific to delivery, we expect the revenue to decline in mid-single digits for the year, with Q1 being slightly higher due to the wraparound impact of the Edgio transaction from last year.

By way of comparison and for consistency with 2025, using our former compute reporting methodology, we expect the combined growth of CIS and OCA to be at least 20% year-over-year. At current spot rates, our guidance assumes foreign exchange will have a positive $36 million impact on revenue in '26 on a year-over-year basis.

Moving on to operating margins for 2026. We are estimating non-GAAP operating margin of approximately 26% to 28% as measured in today's FX rates. The decline in operating margin for the full year 2026 is due mainly to increased colocation and depreciation expense associated with the continued buildup of our CIS business.

We anticipate that full year capital expenditures will be approximately 23% to 26% of total revenue, driven by the investments and costs that I mentioned earlier. As a percentage of total revenue, our 2026 CapEx is expected to be roughly broken down as follows: for network-related CapEx, we expect approximately 4% for our Delivery & Security business; approximately 10% to 13% for compute; and for other CapEx, we expect approximately 8% for capitalized software with the remainder being for IT and facilities related spending. Excluding the impact of the increased hardware pricing, 2026 CapEx would have trended within the 18% to 22% range. The impact of increased server costs is mainly included in the compute line item above.

Moving to EPS. For the full year 2026, we expect non-GAAP earnings per diluted share in the range of $6.20 to $7.20. This non-GAAP earnings guidance is based on a non-GAAP effective tax rate of approximately 19% and a fully diluted share count of approximately 147 million shares.

With that, I'll wrap things up. Tom and I are happy to take your questions. Operator?

[Operator Instructions] And today's first question comes from Sanjit Singh with Morgan Stanley.

Congrats on a very strong Q4 results. Ed, you played a lot of great detail on the dynamics around CapEx as well as the momentum you're seeing with the CIS business. When I look at the increase in CapEx, I mean, it's roughly coming up by, I think, $270 million. Going back to like the discussion that we've had in prior quarters, that roughly $1 of CapEx equals to $1 of revenue. Does that still hold? And as we think about this increase in CapEx, how should we think about that translating into revenue from a timing perspective, both this year and then maybe going beyond 2026?

Sanjit, thanks for the question. So obviously, I talked about having some inflation in memory chips, hopefully, that is something that doesn't last for a long time. So that obviously skews your CapEx a bit. And as I talked about, most of that is affecting your compute because there's a lot more memory in those servers. So the dollar CapEx per dollar revenue would not hold true for this particular buying CapEx, but it's not that far off. Generally speaking, we're seeing something roughly like that. Obviously, for larger deals with longer commitments, we will offer volume discounts. But even for some stuff, you might get a slightly better return like, for example, we'll be launching a rental service where you can rent GPU by the hour starting sometime later this quarter, where the list price for that's $250, so that would work out a little bit higher. But generally speaking, it's a decent number to work with -- modeled it a little bit lower for this year, just given that we've seen higher CapEx costs associated with the memory prices.

Understood. And then just one follow-up on the Akamai Inference Cloud opportunity. Really encouraging to see that 4-year deal with a major tech company. Can you speak a little bit about the pipeline? I know we have some really big customers looking at the opportunity. But just in terms of the breadth of interest pipeline. Any color you can provide there on potential more customers signing up for the service?

Yes. Pipeline is very strong. In fact, the Inference Cloud offering we announced in the fall where we deployed the GPUs into 20 cities that's already sold out, even though it's not generally available yet, just from the beta customers. And so now we're ramping up the investment there, as Ed mentioned, and very strong pipeline. In fact, with the large customer we talked about already committing to take over a substantial portion of that.

The areas of interest are broad at a high level, obviously, inference applications, also post-model training, but specifically things like transcoding, real-time translation, generative media to generate images and video on the fly and the new Blackwell GPUs, very good at doing that with much lower latencies. Vision, processing what is seen, customer support bots, all sorts of gaming applications, streaming, rendering, modifying characters as you go along in the game, in commerce, virtual fitting room kinds of applications. So it's almost like the buyers looking at themselves in a mirror wearing the clothes, also making sure the close will fit, so you have fewer returns, a lot of robotics and autonomous vehicle kinds of applications, areas that these folks might not traditionally be Akamai customers, now potentially large compute customers.

And generally, the field of local LLMs, as people -- companies do more kinds of things themselves, but want to operate their own model. That's great because that's the kind of thing you'd want to do on Inference Cloud and have it done close to where your employees are. So we're very enthused about what we're seeing so far and a lot of potential for growth for us.

And the next question comes from Mike Cikos with Needham & Company.

Congrats on the strong end to '25. The first question I have for you on that major U.S. tech customer, can you help us think about how this came together? It's great to see the duration. We're talking 4 years and the $200 million minimum commitment. But was this a new logo to Akamai? Or were they a previously existing customer within CIS or another portion of the Akamai portfolio? And then I just have a quick follow-up.

Sure. I'll take this one, Tom. So the good news, it was an existing customer. It wasn't one of our largest customers, though. This was somebody who was using us for CDN security and then had discussions with them going for several months now on a pretty exciting workload. We're not at liberty to disclose who it is. But the good news is existing customer who has dramatically increased their spend, and we hope there's a lot more business to do with them.

That's excellent. And I appreciate that, Ed. I guess the follow-up -- [indiscernible]. when thinking about the capital intensity here, and I really appreciate the disclosure [indiscernible]. But how do we think about the level of CapEx you guys are deploying here? Are you changing in any way how you're sourcing servers or going in and buying hardware versus where we've been previously, just given the heightened price components that we're seeing out there in the market and they feel that this is somewhat different as far as the cycle and persistence of these pricing dynamics? Anything that would be incremental as well.

Yes, sure. No problem. And the capital intensity isn't necessarily increasing for any other reason than we're seeing significant demand for CIS. So that's the major driver. And obviously, making that purchase of $250 million for the Inference Cloud is very well informed. And as Tom mentioned, we have -- it's great to have one customer who's taking a good chunk of that and having that committed, it's just a great opportunity for us to put that capital to use. So I hope we do more of that. So I'm very happy about that.

Now in terms of -- as the complexity of what we're doing or what we're buying changing, no, not really. We're buying mostly servers and networking equipment and things like that. We are looking at trying to reduce the impact of the server -- the memory chip increase in costs. So we're looking at sourcing things differently from different sources, et cetera. But generally speaking, there isn't really any significant change.

And as far as our co-location posture, we're still using the third-party colo providers. At some point, maybe that changes once we get a lot larger. But no real significant change. And hopefully, as I broke out the different components, if you want to think about it this way, you take out the $200 million for the price increases. And then if you look at that purchase of the AI Inference Cloud as sort of something that we did that was a little -- a little different than last year. The normalized CapEx is kind of at the lower end of what our range typically would have been. So this is a good kind of capital intensity increase when you have a chance to fuel a business that's growing as fast as CIS is.

The next question comes from Rishi Jaluria with RBC.

It's nice to see acceleration in the CIS business at scale. Maybe 2 questions, if I may. Number one, if I start to think about some of the success that you're having on CIS, it sounds like you're having that with existing Akamai customers that may have used you for delivery or security or combination there above. Maybe can you help us understand, as you think about going back to those customers, is the total ACV or whatever sort of metric you want to use, with those customers growing meaningfully as a result of this? In other words, just trying to get a sense that it's not a situation of money that maybe they would have spent for delivery in the past. And as we think about pricing and DIY, it's money that's going elsewhere that this is actually being additive to those customers' total bills, if that makes sense. And then I've got a quick follow-up.

Yes, yes, great question. It's certainly, it's additive. We don't -- we're not horsetrading any delivery for compute or anything like that. As a matter of fact, this particular large customer was done out of cycle, so it wasn't even done as part of a renewal. So it's all 100% additive. And I would say, yes, we're having good success with existing customers, but also with new customers. John talked about the pipeline. What's interesting with that pipeline is we are starting to see verticals. We don't typically are strong in from a legacy perspective as far as CDN goes. And so that's good to see. We see partners bringing us new business. And there's really a mix in that pipeline of new and existing customers. And we've actually seen total new customer count pick up over the last 1.5 years or so, and I think a lot of that has to do with having CIS as an offering that's more broad.

Got it. That's helpful. And then maybe I'd be a little remiss if I didn't ask about kind of some of the onetime factors going on in calendar year '26. As you think about your guide for the year and obviously appreciate the granularity -- just can you maybe help us understand -- I know this isn't the Akamai of 10 years ago when maybe live events were a lot more meaningful. But I still just want to understand what are kind of your assumptions in terms of the major events are happening between Winter Olympics going on right now between the Fifa World Cup in the summer, got some big AAA gaming releases that may or may not happen, obviously, release dates kept getting pushed out. Can you just help us understand kind of the puts and takes and how that ties into your numbers?

Yes, sure. Happy to take that one. So if you think about events, they come in different flavors, you get the small events like a live concert or a Super Bowl, those tend to be very small revenue. Sometimes you might get a capacity reservation fee. So maybe that might be $0.5 million to $1 million or something. So nothing too dramatic there. Something like the Olympics 3 weeks long, it's a few million dollars, depends on how many rights holders you have, how many different rights holders you sign, et cetera. So it's not a huge jump. Doing $1 billion-plus a quarter, it's fairly insignificant to the quarter. It's a good business, so we'll take it.

Something like the World Cup is a little bit longer. So you'd probably see maybe $3 million to $5 million, $5 million to $6 million, something like that. But again, nothing overly material, although it's nice to have all these events. And then the things like an NFL season much better, you're going to generate a lot more revenue there from a number of different customers. So it really depends on the length of time and the number of people that have rights.

Something like a gaming release depends if it's a really popular release that has a lot of updates to it that can be popular and can drive some extra revenue. It really depends. Something like Fortnite certainly was a big tailwind for us several years ago. If you see a new console refresh cycle, that's a much bigger impact for us because you're talking now about hundreds of millions of consoles getting firmware updates and lots of updates. So that's the way to think about the event. So it's nice to have them, but it's not overly material for the year.

Our next question comes from Roger Boyd with UBS.

Congrats on a good end of the year. I wanted to ask about the handful of larger CIS deals that you had noted last year as being delayed out of the back half of the year. Can you just update us on how those are progressing and maybe how those are embedded into the 2026 guide? And I think you mentioned the $200 million deal you signed this quarter will start to ramp in the fourth quarter. At a high level, can you just talk about the typical ramps you're seeing in compute? Is any part of this result of capacity constraints? And do you expect to see these ramps on the compute deals get shorter over time?

Yes, it really depends. Some we can get up and running pretty quickly. It really just depends on the size of the transaction and if there's any specific geo where we may need to get some additional colocation. The colocation market is tight. But we've got -- we're a big buyer of colocation, so we're doing pretty well there. We did see some of the larger workloads ramp up at the end of last year, and we've modeled in what we think those will do. And I talked about this particular really large deal will start ramping in Q4. And part of that is we have to -- we're ordering all the chips, putting them in place, getting some space. So it just takes a bit to ramp that up. It is -- obviously, GPUs are pretty tight supply chain, but we're able to get those out and launched here. So we've modeled in a variety of different outcomes on that in terms of our guidance range. But if the bigger the deal usually takes a little bit longer to ramp, and in some cases, people can get up and running very shortly.

The next question is from Fatima Boolani with Citi.

I wanted to focus on the trajectory of the delivery business. I think this has been asked in a couple of different permutations. But I wanted to ask it at more of a higher level with respect to the aggregate environment for internet traffic and traffic volumes. You've had a bunch of your peers sort of talk to accelerating or improving traffic trends. I was hoping you could compare and contrast for us what you're seeing on the Connected Cloud Network?

And then the flip side of that coin is just the pricing dynamic. So to your point, the delivery business has seen a pretty substantive degree of stabilization over calendar '25, and it seems like that is going to persist. So I just kind of wanted to unpack the P and the Q on the delivery equation?

And then I have a follow-up as well, please.

Yes. At a high level, the trends that we're seeing and projecting for this year are pretty comparable to what we saw towards the latter half of last year. Traffic environment seems very reasonable. Obviously, fewer players in the market than a couple of years ago. Pricing environment remains competitive. We still have folks out there selling, in some cases, at very low prices, which we won't do. We -- in particular, we see some costs rising as we've talked about, especially in memory and in some cases, we'll actually be raising prices to help offset those costs. But I would say at a high level, what we're expecting this year is pretty comparable to what we saw last year, especially in the back half of the year.

I appreciate that. And Tom, you had sort of talked about the rental service that you're going to launch in the upcoming quarter. I wanted to take the opportunity to have you unpack that, what the expected structure is, what the economics look like? And maybe in a more broader sense, the type of utilization that you are expecting on your network as you think about and deploy this $250 million of incremental capital to scale out the inferencing cloud ahead of the capturable opportunity?

Yes. So in terms of Inference Cloud, there's 2 models. One is the traditional model where you buy access to the GPU by the [ VM ] or the token. And that's what we'll be going GA later this quarter. The GPUs we deployed into 20 cities are already pretty much sold out. So we're adding an order of magnitude, more capacity, and that's what the $250 million investment is for. And in addition to selling by the token or VM, we will be selling clusters so that you might decide to buy hundreds or thousands of GPUs in certain locations. So that will be a new model that we're introducing this year and have some very large customers buying CIS in that way.

Yes. The one thing I would add, Fatima, is in terms of the early pipeline, we are seeing a bit more skewed to the customers who want to guarantee the capacity. So they're asking for whether it's several hundred or thousand or whatever GPU for a period of time, multiyear time kind of deals, which is obviously a better model. I like to see that. In terms of the usage, we haven't done that yet. So we don't know exactly how that's going to play out. So we've got a range of various outcomes there. But certainly, there's a lot of early excitement and demand in the pipeline that we're seeing for what we're buying.

And the next question comes from Frank Louthan with Raymond James.

This is Rob on for Frank. Congratulations on the strong 4Q. So my question is, what sort of revenue commitments are you guys able to get from customers today relative to before? How prevalent are those now versus previously what percentage of revenue on the delivery side is under those commitments? And what's your outlook for delivery growth this year, specifically with AI-based traffic, if you can give us a better sense of that?

Yes. We are seeing longer commits for -- really for all of our services, partly that's by design. And I think customers also interested in having that take place. And with the delivery growth, we're looking at about the same rate, so mid-single digits this year.

And Ed, do you want to add to that?

No. I would just say you'll see like the RPO is growing for the total company quite a bit. That's just a function of what Tom is talking about in terms of folks making longer-term commitments. And we've incentivized our sales force to get longer commitments. As far as the delivery market itself, not a huge change there in terms of commitments. There are some customers that might commit a percentage. Some might give you some type of exclusive or either a part of their business or geographic area, et cetera. So there's really no dynamic change in the delivery business. It's roughly the same in terms of committed versus uncommitted. But since the other parts of the business are growing much faster, security and compute, we're seeing a lot longer and bigger commitments.

And our next question comes from John DiFucci with Guggenheim.

A lot of interesting things happening here, Tom, and especially around the CIS business. And thanks again for breaking that out historically, too. Last year, you announced a very large contract with a social media customer. And I think this is sort of a follow-up to Roger's question. And that company had a lot going on internally, right, and externally, too. And it required the additional build-out of capacity by you. I think we're a year into that, and we believe -- I believe the buildout is complete by you, but I still think there's a lot going on with that company.

I guess could you -- because a lot of this stuff could come on lumpy. And I'm just trying to figure out how to think about this going forward? And this is like the first deal like this, and it was great to hear about that $200 million 4-year deal, too. But with this deal, have you started recognizing revenue yet from that customer? And if not, can you share a little bit about what you expect to recognize revenue?

And then I guess one other part related to this is that social networking deal you talked about that's going to consolidate on Akamai and take away from a hyperscaler that I think Tom mentioned in his prepared remarks, is that the same customer? Or is that another customer? Sorry for the long-winded question.

No worries, John. I hope by interesting you mean good interesting. So I'll take that. It's not the same customer. It's a different customer. In terms of the lumpiness you talked about, generally speaking, we don't see lumpiness per se. As I talked about with the new deal we just signed, the $200 million 4-year deal, I expect that to be fairly even -- maybe there's some upside as usage ramps. But there's not like say, a big chunk of revenue and then it goes away or whatnot. But we do expect that to start ramping in Q4 just as we start deploying, make the purchase, get the GPUs, get them up, customer has to do their testing and then they go into a full launch. So that just takes some time. So starting in Q4, we expect that to ramp up and then continue into next year.

And then in terms of the large customer we signed last year a $100 million deal, we did start taking a little bit of revenue in Q4. We expect that to continue to ramp up throughout the year. I will say there is some seasonality. We do have a little bit of work in the compute business that might be tied to, say, like a season or something like that, say, a sports season. So you may see a little bit of extra revenue in, say, Q4, and it dips a little bit in Q1. But generally speaking, you don't see big lumpiness, as you said, in the compute business.

Okay. And that makes sense. That makes sense. I was thinking sort of like Oracle, but they're bringing on these huge AI training data centers, which are just come all online, but that's not how your business is.

And I guess just one follow-up, not a little bit unrelated here, added to an accounting question. How much of that fourth quarter restructuring charge of $55 million was any of that in cash for this quarter? Or was the -- because the cash flow was a little weaker than I think people expected. And CapEx is higher so I get that. But was that...

Yes, good question. So most of the cash flow is a timing issue just in terms of timing of cash receipts and payments and we made some pretty big tax payments before the end of the year. So that skews the cash flow. But if you look at last year, I think it's relatively in line with last year. But in terms of the restructuring, that cash will go out in Q1. So the majority, a little over half was intangible assets, so there's no cash associated with that. Severance was a little less than half that will hit in Q1.

Okay. Great. And a lot going on here, but -- and I actually -- I definitely meant good when I said interesting.

The next question is from Will Power with Baird.

Okay. Great. Maybe just to switch gears to security. Great to see the continued Guardicore Segmentation API security strength and API, I guess, topping up $100 million. It'd be great just to get a better kind of outlook since for growth expectations on those 2 pieces in 2026, how that folds in? And then probably for you, Tom, it would be great to get your perspective just on how you're thinking about any potential AI risk kind of across your security portfolio, just given some of the market concerns out there? It seems like the businesses have been pretty resilient. But maybe you can just comments on what you're maybe seeing competitively from any other AI entrants or technologies in the marketplace.

Ed, why don't you take the first then I'll do the second.

Sure. Happy to. So yes, very happy with what we're seeing with Guardicore and API security. We had a really good, strong fourth quarter finish in terms of bookings. And the nice thing with both of these businesses is we're seeing a nice mix of new customers versus existing, especially with Guardicore. As a matter of fact, the majority of revenue is coming from new customers associated with Guardicore, which is great. And then with API, both actually are very low on a penetration rate within API security, less than 10% of our existing customers have purchased that. So there's an enormous amount of runway there. We're seeing a lot of big adoption across many, many different verticals, too. So it's not just a one vertical like financial services. It's really across everything.

So we expect, as we go into next year, very similar to last year. In terms of API and Guardicore now a little bit more scale, driving the majority of the growth. The other product lines, whether it's bot management and WAF continuing to grow, albeit slower and then service is continuing to grow as well. So we expect growth in most of those categories, maybe [indiscernible] tends to be a little bit more ventured but maybe that's not [indiscernible] more flattish. But we do expect growth across the board and this year to look pretty similar to last year with the majority coming from API and Guardicore.

Yes. And to your second question, that's a great question. We are not seeing risk from AI and do SaaS do-it-yourself kinds of things. One of the key reasons for that is for our services, security services, you really need to run it on the large distributed platform by and large. And one reason for that is if you try to sort of do it yourself and your data center in a few locations, you just get overwhelmed with the volume of the traffic. And you don't have any chance to really apply the security because you're flooded. And that's where Akamai's distributed platform makes the critical difference as we intercept all that traffic, the bad traffic, out where it starts, and we can do that at great scale. And so we're not -- I don't think -- we don't have that kind of exposure.

Now the good news is if the AI induced risk to SaaS as that materializes, that's a big tailwind for us on the compute side because these enterprises are going to need to run their models that are doing these SaaS tasks and generally, they're going to probably want to run them close to where their employees are, and that's a perfect application for our inference cloud. So on balance, if that really materializes, that's a tailwind for Akamai, I think, not a headwind.

Next question comes from Jackson Ader with KeyBanc Capital Markets.

This is Evan [indiscernible] on for Jackson Ader. Just curious on the compute side. What are you guys seeing as some of the main reasons for customers choosing Akamai over whether it's other hyperscalers or other competitors for compute workloads at the edge? And I know cost has been a key element you guys have called out in the past, but was just looking for some added color on how Akamai can continue to win some of these deals?

Yes. Great question. It's performance. It's scale. And yes, cost is generally lower. But just as an example, we talked about on the last call, the 3 big hyperscalers in the U.S. are all using our compute. And for them, it's not a cost issue because they have their own clouds, obviously, for them, it's performance issue because we can run their logic in a lot more locations than they can do themselves with their clouds. And so that results in better performance for them, they're closer to the users and better scale, especially if you're doing things around video that are a bit intensive. You need to do that in a much more distributed fashion.

And then for other customers, cost does come into play. As we talked about some of our customers getting really substantial savings as they move out of the major cloud providers, the hyperscalers to Akamai. In fact, Akamai achieve major savings as we moved out of the hyperscalers, a lot of our applications onto our own cloud. So better performance, better scalability and better cost in many cases.

And the next question comes from Patrick Colville with Scotiabank.

Just one for Dr. Tom, please. I guess I just want to go back to the Inference Cloud. I mean you talked earlier about some nice use cases for accelerated compute at the edge. And it seems like the comment spread is that latency is important for those use cases. But I guess my question is this. I mean, in the CPU world, edge compute was a good market, but it wasn't enormous. Most compute happened locally on device or at the hyperscale core. Why would accelerate compute be different that you're going to have this large and very exciting markets at the edge?

Yes, good question. And it's not just latency. Latency of course, matters, but it's scale. When you think about some of the AI applications, generative media, you're generating video, processing video. And just -- you don't have the capacity, the bandwidth at a core data center to be generating or processing millions of personalized videos concurrently. You got to do that in a distributed fashion. Just like anything with live sports or anything like that, it's got to be distributed. So it's not just latency. And increasingly, as we're seeing these applications, they are bandwidth intensive.

Also, when you talk about doing speech, when you're conversing with your avatar, it does need to be real time. You can't be going far away to a data center or it's not the same experience. Now in the past, the GPUs weren't fast enough to make that work. But now they are getting to that point where it is a few tens of milliseconds. And so the latency does matter more now.

And can I just ask a quick follow-up there on the Inference Cloud, again, actually. I mean, 2 parts. First one is, do you need to do any software updates in terms of the software that Akamai has for customers to run Inference Cloud? And then, I guess, the kind of -- the second part is in terms of Akamai's target customers here, it seems like the customer profile is slightly different to the existing customers. Am I interpreting that right that you will be able to sell this to existing customers, but also a new cohort and maybe even AI natives?

It's a broader customer pool. So our existing customer base, yes, they are good targets for us. But there's also, as we talked about, Ed mentioned, there's a lot of customers who are signing that weren't using Akamai before because maybe they didn't really have delivery needs or even web that firewall at any kind of scale. And so they're new to Akamai.

And in terms of software updates, we're always upgrading the software in our cloud platform, but it's nothing special per se with the GPUs. It works very much in the way that Akamai Cloud has worked, Linode has worked. We are selling an additional model, as Ed talked about, with clusters with a long-term contract in addition to the traditional model, which by the [indiscernible] by the token.

The next question is from Jonathan Ho with William Blair.

Congratulations on the large AI inferencing deal. I was wondering if you could give us a little bit more color in terms of what was unique about Akamai to cause the customer to maybe choose your solution over competitors? And if you could maybe give us a sense of philosophically, whether you're building out capacity to meet that demand? Or are you comfortable investing even above that demand as you're adding capacity?

Yes. It's what we've been talking about, it's really good performance, very reasonable cost. And I'd add for something that's this critical an application, trust matters. And we talked a little bit about that a few minutes ago. Akamai customers do trust us. We've really earned that with our delivery and security services, our reliability, our customer support. And for something this big and critical, I think that makes a big difference. So -- and we are needing to build out in this case. And that's part of the large investment that Ed talked about, we're greatly increasing the capacity of Inference Cloud.

As I mentioned, we pretty much sold out the 20 locations with the GPUs that we have deployed starting in the fall. And now we're going to increase that by about an order of magnitude and part of that will be used by this large customer that we talked about.

And the next question comes from Rudy Kessinger with D.A. Davidson.

Jonathan actually took the main one that I had. But on the $250 million, you're spending [indiscernible] Inference Cloud build-out. I guess by year-end this year, I mean, how many locations do you intend to have GPU capacity? And I believe that the initial announcement last quarter, it was like 17 or 19 locations or something. But how many do you intend to have that GPUs in by the end of this year?

Yes. I -- we're in about 20 now, and I don't expect that number to be a lot larger, but the locations were in themselves will be a lot larger, which enables us to add the model where we can sell clusters of GPUs.

And the next question comes from Mark Murphy with JPMorgan.

This is [indiscernible] on for Mark Murphy. Ed, I believe you mentioned that you're seeing deals in the pipeline coming from verticals that maybe weren't as prevalent before. Can you help us understand what those newer verticals are? And then are those coming more from the direct sell motion or from the channel?

Yes. So it's a little of both. We're getting some from the partners that we work with. We announced a relationship with NVIDIA. They refer customers over to us as an example. And in terms of the verticals, think of things like life sciences, manufacturing, health care, different types of industrials. Typically, generally don't have really big websites, but do spend an awful lot on compute and they're also good security customers as well. So direct motion is part of it. The direct is doing a good job of introducing this to all of our existing customers. I've gone on a couple of calls. And certainly, it's really going to have a lot of interest and customer feedback is that they believe we have a right to win here. It makes a lot of sense for us going here. There's an enormous amount of curiosity and we're doing a lot of proof of concepts. So good to see that the demand is coming from a variety of different sources.

And [indiscernible] then I think at least 3 named wins versus hyperscalers now it's across CIS and security. You guys have always found success there, but do you see any changes in the competitive dynamics there? Is that improving for you guys versus the hyperscalers?

You cut out on the first part of the question, the competitive dynamic in what area?

Against the hyperscalers.

So what we've competed with the hyperscalers in delivery and security for over a decade, I don't see any fundamental change there. We compete very successfully against them. In fact, 2 of the 3 big hyperscalers are large Akamai customers for delivery and security. And of course, now we're adding compute into the mix and already all 3 are using us for our compute capabilities. And again, there, it's not an issue of cost for them. It's an issue of better performance, at least in part because of our distributed nature. We can get their compute logic closer to their users where they want it.

Yes. One thing I'd add [indiscernible] just would add, it's not necessarily that the only way we win is by taking business away from them. In a lot of cases, we're seeing new workloads, especially as inference becomes a much bigger part of the equation in AI, a very good spot to go to when customers have challenges where either latency needs to be very, very low and you need to be super close. We've seen some customers tell us that even being in a different state in the U.S. gives them too much latency. They need to be within couple of hundred miles, which is different than what you've typically seen [indiscernible] the CDN world. So it's not a question of a zero-sum game where we win they lose, it's we do some from time to time, take some workloads. We do go head-to-head in competition where we go in a [indiscernible] perform better, et cetera. So the market is just growing so fast that there's plenty of room here for us. I think we're starting to demonstrate that we're becoming a real player here.

And the next question comes from Jeff Van Rhee with Craig-Hallum.

This is Vijay Homan on for Jeff. Just one for me. I know you mentioned the impact of AI on the cloud segment. I was hoping you could just expand on the impacts of AI on security and delivery revenue maybe to the extent that that's driving traffic and how it's changing the demand of your customers for your services?

Yes. So there's a variety of impacts with AI on security. One of them is that AI really helps enable the attacker. And so we're seeing much larger bot nets out there because the attacker can use AI to take over a lot more devices, they can use the AI to train malware to get around known defenses, and so you see more penetrations. You've seen the AI with deep fakes you couldn't possibly know or fake. So in a lot of ways, it's making the attack environment much harder to defend against. Also, as enterprises adopt a lot of AI apps and agents, that's a whole new attack surface. And you need special defenses, like, for example, our new firewall for AI. Also today, enterprises are in a tough shape. They don't even know all the shadow AI they have. And so we have new capabilities there with our API security to extend it, to identify the AI applications they have exposed. So you need to know what AI you've got out there and you need to defend it with special firewall capabilities, which we do.

So I think AI is having and will continue to have a positive impact for our security business in terms of our revenue even though the attack landscape is nastier that in some ways, it's more need for Akamai services. In terms of delivery, we are, of course, seeing a rise in the scraper bots. And so if left undefended, that would create a need for more traffic. Now for our customers through our bot management solutions, we actually help them to deflect a lot of the scraper bots, give them visibility into what the various bots are, what they're doing. And then our customers decide, okay, which ones do they want to block, which ones do they want to do special things for. So I'd say on balance, yes, probably a traffic increase to an extent. But again, there, it's more -- creating more of a need for our bot management so that our customers can handle the various scraper bots in the way that makes sense for their business.

And this does conclude today's question-and-answer session as well as today's conference. Thank you for attending today's presentation. You may now disconnect your lines.

All right, almost lunchtime. We're super excited to have the CEO of Akamai, Tom Leighton. Tom, thank you for joining us at Nasdaq London and once again.

Thank you.

I appreciate it, Tom. We're going to focus a lot of conversations I think what the the key investor excitement is around, what you guys are doing in compute specifically with GPU inference. Before we get through, though, maybe just sort of a level setting question in terms of as you look 2025 year to-date. How would you characterize the IT spend environment across the different areas of business? And what about 2025 has surprised you either to the upside or the downside?

You mean spend that we're spending in IT or that our customers are spending?

It's customers.

Customer. Yes. So obviously, security is still very important on people's minds. The attack rates have gone up, especially now with Gen AI gives power to the attackers and a lot of adoption of Gen AI capabilities in enterprises, and there's a whole new attack landscape. You got to defend that. And on the compute side, a lot of interest in the new applications that are becoming possible, agents to do all sorts of things. And so I think rapid adoption of AI, and that's a tailwind for Akamai.

Awesome. So let's talk about the opportunity with the Akamai Inference Cloud. So Akamai Inference Cloud has launched. It's getting a lot of investor excitement. Can you talk about how the partnership with NVIDIA came together? And why do you feel Akamai can be a player in the GPU inference market, given a lot of competition with the hyperscalers and the neo cloud. So just maybe just give us a history of like how you guys thought about entering this part of compute space.

Yes, there's a lot of questions in there. We've always had a good relationship with NVIDIA. I remember way back in the day when there was interest in deploying GPUs in our edge platform to help create gaming as a service that never really took off because of the financials. It's better to have the user spend, the power in the home to power the console, to buy the the compute device, to pay for the colo in the home than doing it in the cloud. But now with the rapid rise in Gen AI and agents and Inferencing, that makes a lot of sense to do on the edge.

I think there's a lot of excitement there. Certainly, in our customer base about what's the web going to become, what's going to be possible? And just -- take an example, say, in commerce. We've got customers that, boy, wouldn't it be nice if you -- when you bought a article of clothing that it fit and you didn't have to return it. And so you can scan your body, the e-commerce company will know your shape, they'll know the various articles are clothing, do they run large, small, whatever. And so that when you do buy something, a very good chance it will fit.

Well, in addition, you'd sort of like to know how do you look in that article of clothing. And so today, we have some customers that will show you a picture of that. We've got customers that want to show you a video of you wearing the dress and then in different environments. Pretty soon, I think you'll see your personal concierge, will be an Avatar or it will look like a human being. It will talk to you. It will make suggestions of what you might look good in and show you. The whole experience in commerce could well change.

You look at buying tickets for things, concierge service for that, what you might like to go see, getting everything all scheduled. There would be agents just all over the place. Each agent doing a specific task, medium-sized model, and they'll all be talking to each other and coordinating against data personalized to you. And so the whole experience and how the web works, the protocols, probably going to be different. And we're excited because that's a perfect area for us. You're going to want those applications to be really fast, reliable, scalable, secure.

The way we have built our platform to do that and a key aspect is to be distributed. So the compute is being done close to the user. If your a particular agent is generating a personalized video, no way that scales in a central data center, just like streaming way back in the day, no way you're going to deliver all those streams from a few locations and big data centers, can't work.

You're going to want that to be done distributed. And if you're buying something, you're going to want low latency. Now -- before now, when you're running the models, they were slow. They were big. As you get the lighter weight models on things like the NVIDIA 6000s, they're fast. And so you do want to be close. So you don't have the bottleneck being the back and forth between the user and the model.

So I think in some sense, the web changes, but what you need in the fundamentals very much fundamentals that Akamai has helped the web scale from the beginning. They're delivering. And then, of course, it's got to be secure.

In some sense you're talking about kind of a continuation of hyper personalization, but doing it in real time, and that's kind of real-time inference capability and things like video -- real-time video, that's a whole new world.

Yes. I think you'll even see it in things that are what used to or what are today one-to-many video that the version of the game I see is going to be different than the game you see because maybe I like this team and you like that team. We've got -- I was talking to a customer just a couple of days ago in New York with exercise stuff. And you're in the class doing your thing, and you want to personalize in terms of your language so that you can enter other markets. Same instructor, but talks in your language and accent, I guess the whole -- everything I think is going to change in a way that is -- we're in a very good position to help enable.

Yes. That's super interesting. So let's talk a little bit about the underlying -- at least the hypothesis around unit economics of the Akamai Inference Cloud. So on the earnings call, you described the current unit economics as $1 CapEx basically equals $1 revenue. Can you unpack that for us? I mean, what gives you the confidence that these inference workloads will have the sufficient lifetime value to become a profitable, not just a high-growth business?

Yes. I think the same math looks like it's holding for the newer GPUs, they cost more, they take more power, but they generate more revenue in ballpark, the same ROI. And I think a good question when you're buying the GPUs is are they going to last? And what you've seen so far is the GPUs being used in these gigantic farms for a gigantic training. And there, every year, you want to buy the next version because you're trying to do more variables, you're trying to get well, can you do generalized intelligence or something. That's not the game we're in.

We're in supporting applications for customers. And the nice thing about that is that, hey, the 6,000 is good at generating a video, personalized video on the fly and it's still going to do that in 3 or 4 years, right? It's not an issue that we're going to need a bigger one to generate that video for you. So the specific tasks you're doing we believe that equipment will be good for us over the long haul because we're not in the big core training, taking the model to the next level to play chess better or to solve super hard math problems.

That's a pretty clear distinction I'm glad you made that. And so you talked about like the technical capabilities of the chips and the hardware. In terms of the workloads because they're applications essentially, do you also feel that, that application and those workloads are going to be around 4 or 5 years? Or do these workloads kind of shift from one cloud to the next as they kind of shop around for the best unit cost and just get your perspective on.

I think the applications once they're here, will be around. I think cost does matter to our customer base. And it's an area where we have good competitive advantages. Many of our customers that had been using a hyperscaler that have moved to us, we're helping them save money at the same time. And of course, with our distributed architecture, we're very efficient at moving data around. We do more of that than anybody. And so we're in a good position, I think, to offer competitive pricing.

Great. In terms of the CapEx build-out to pursue the Inference opportunity, how do you show that any incremental CapEx investment will ultimately see the traffic right? Today, it sounds pretty obvious, pretty easy. You build it kind of they will come, that's kind of the stage of the market. As you think about building a durable business, how do you guys think about the CapEx build-out and the revenue coming online?

Yes. So we build out CapEx for what we think we'll be consuming in the near future. Initially after Linode, we did do big tranches that is in use. And so we're in the mode now that we build out now based on what we think we're going to sell and use 3 to 6 months down the road, a related issue is the data center deals that we do. And so for the bigger data center footprints, those will be long-term contracts and will often build in room for growth. And we pay as we go. We do get hit with the accounting charge for the linearization of that so that the accounting says that we're spending more than we're actually paying in the early years. So we do get hit with some of that.

In terms of the Akamai Inference Cloud footprint, I think you're currently in 17 locations. What does that look like over the next year in terms of expanding upon -- building upon those 17 geo-locations?

Yes. So there'll be more cities we're currently building out in India and Southeast Asia, and they'll also get deeper. So there'll be bigger footprints. And so in some of these cities, we're looking at 10-megawatts data centers, which is larger than generally we've been. We're in over 4,000 POPs. And obviously, the vast majority of those are not GPU equipped POPs, they're not to think about the power. In fact, we don't even pay for the power in the vast majority of those locations. But for the GPU footprint, there will be some cities that will be bigger with more concentrated GPU.

And then in terms of like pricing and how the pricing model works for Inference. Can you just sort of lay out or sort of outline what that looks like? It could be -- is it pretty straightforward to sort of GPU per hour type pricing? Or do customers are going to make some sort of commitment and burn down? How does that...

Yes. So both. So far, it's been per consumption basis, VM, hours could be tokens in terms of GPUs. Also for GPUs, there are some cases where a customer may want a tranche. And then we would sell, okay, this many GPUs over some number of years that are committed.

Great. The Akamai Inferences Cloud is part of a broader business called Cloud Infrastructure Services. And that's been kind of the engine powering, I guess, the best growth at the company. Cloud Infrastructure Services, I think, accelerated in Q3. And the team has been signaling that could accelerate meaningfully going forward. What drove that acceleration in Q3? And to what extent is this being driven by a handful of customers? Are you satisfied with the breadth of the growth coming out of CIS?

Yes, great question. The biggest customers are generally the biggest spenders. So we've had a bunch of growth driven that way. I think we've been very favorably impressed and surprised at the breadth at the same time. And part of that is our field is now selling it broadly. Last year was really the first year of broad sales. We have a specialist team that we're growing. There's more market awareness that, hey, Akamai as a cloud company, what you're doing with the hyperscalers you can do with Akamai. Of course, now we have Inference Cloud, which gives us another capability in terms of market awareness and capability to sell.

So it's both. We've got some very large customers that are growing. The hyperscalers now are all cloud customers at Akamai and a lot of the more good enterprises, not having to be a giant company, but a solid enterprise.

You could take back to the origins of Akamai's entry into the Cloud business, through this acquisition of Linode, which got you into kind of classic centralized public cloud offering, was there any synergies with like Linode and their data center footprint? And then as you expanded that with the core Akamai network? Or was it kind of a complete -- completely different compute stack? And was there any sort of integration efforts associated with that?

Initially different compute stack, not distributed, not enterprise-grade, small and medium business, but really good stuff, very developer-friendly, very popular with developers. And so what we've done is get it to be enterprise-grade. So a huge amount of work on reliability, various certifications, more functionality that major enterprises need, also making it more distributed. So we're in over 3 dozen cities now with full stack compute and storage. A new fabric in these cities is to make it be a lot more reliable. And we have taken the capabilities and put it into our distributed platform, and you see that with our managed container service, which is a software. We take the software that we are running in those core data centers for containers and deploy it into our edge platform.

And in fact, one of the hyperscalers is using that today because we can get their containers, their business logic closer to end users. We have a lot more cities than they can get with their hyperscaler approach, which we compete with.

So yes, Linode was different, but we are using those capabilities to fully integrate it with our platform. Another thing we'll be doing is making the container deployment be serverless. Today, we're serverless with JavaScript, but that function as a service, which is more session-based but getting containers to be spun up automatically, so you don't have to worry about preprovisioning, like today, you do with the cloud and the hyperscalers.

When you think about that specific element about managed Kubernetes Service, running more serverless more broadly, that being unlocked, you got GPU and Inference Cloud, how much of like getting into cloud-native workloads, how much of a demand boost or demand driver can it be for CIS going forward?

I think a lot. That's where -- I think a lot of the future demand is going to be and that excitement is -- Yes, that's, I think, very important to be able to do that.

Is that driving a lot of the customer decisions and the sort of contract commitments that you're winning today?

I think it's driving customer interest because a lot of that is still work to be done. We just started Inference Cloud. We're not fully serverless now with containers, but that road map is exciting for customers. And I think it's just, hey, customers are getting aware that we have these capabilities. It performs very well. It's more distributed than the hyperscalers, so it gives better performance. And for a lot of applications, we're a lot less expensive than hyperscalers. If you're moving a lot of data around, which big media is, if you have really chatty applications, a lot of hits, which commerce does, we're less expensive.

This is brought up a question in my mind. I just hosted the MongoDB management team. And they made the point that in 2025, all 3 major hyperscalers had some sort of outage. And when I think about Akamai's offering, kind of the distributed nature of the 4,000 points of presence, what you guys got the 36 locations with Lenode. Does the Akamai Cloud start to become part of the resiliency strategy for particular workloads given that we've seen kind of outages across the hyperscalers this year?

I think that makes a lot of sense. And I think you'll see the outages probably increase. Just the nature of the beast that as companies get older, your people are single points of failure. And if they move to another company, you've got code bases that nobody at the company really understands it. And that happened in one of the recent cloud outages. It's an area we worry a lot about at Akamai.

Also change management. I'd say the hyperscalers have done a pretty good job there. They do go down from time to time. It's an area where we make huge investment at Akamai. The last time we had an event like that would have been 4.5 years ago. And our goal is to really be Five 9s. And we changed that. One of the biggest banks here in the world based here, the regulators measure it and Five 9s is the standard, and we -- they achieved it using Akamai.

And that's 10 minutes of downtime in 2 years across everything, including attacks or you made a change that failed something up. So a huge effort. And I give the hyperscalers a pretty good score even though they've had outages. Some of our other competitors, very poor. They used to go down every quarter, now you see it happen multiple times in a quarter with big outages. And that's not something you change overnight. It's -- it has to be a design focus and a lot of effort put in to make that happen, not an accident.

The conversation that you and I have been having over the last 15 minutes has been focused on the Compute business and specifically the Cloud Infrastructure Services piece, where that's where the Inference Cloud is, that's where the public cloud business with with Linode is, there is a whole portion of compute revenue that's non-CIS. Can you talk a little bit about what's in this business? And is there a situation because I think it's kind of like a drag on growth in terms of the overall Compute Business growing mid-teens. Is there a world where that business gets sold or it moves into discontinued operations so that investors kind of see the better growth that you're seeing specifically from CIS?

Yes, that's a great question. That business is the stuff that we were doing that was compute related before Linode, basically. And so there were special compute related functions we would do for customers. For example, queuing, customers wanted that, so we did it. It wasn't a real focus area for us. And over time, other companies were created, that's all they did. And so in particular, with queuing, we ended up making a relationship with one of those companies where we've migrated our queuing solution to theirs, which is better, saves us a lot of money because we don't have to worry about keeping a legacy product going.

And also, they become a customer of our cloud. And so as their business grows, it grows on Akamai and helps us. We did that with our media services live, something else that we didn't want to keep doing. We've done it with our Video Manager Service. And these these things, a bunch of those were in what we call other cloud applications. They are the legacy cloud.

And we've got -- there's net storage in there. There's some other things, some of them are growing generally, they're not focus areas for us. And so it's not something you would sell that business. We got to keep our customers happy. But if the right deal comes along where we can save money, help grow our Cloud business by getting more partners on it. Yes, we would do that.

I think going forward, we've said that business is going to -- other cloud apps will be flattish roughly. All the investment is in CIS and that's where all the growth and potential is. We'll probably still report both, although we may -- when we talk about compute just focus on CIS next year since that's what's important. We'll disclose, I think, OCA CC What it was. And of course, as CIS gets bigger, if you were to combine them get this piece pretty static, this piece is getting very large. It won't be -- won't matter.

Over time, it will mix down. So it's made the conversation on to security, which is -- generates the majority of Akamai's revenue. You've been in cybersecurity for over a decade now, built a fantastic business, a $2-plus billion run rate business. Growth has dipped below 10% on a constant currency basis. When you take a step back, how much of an investment priority is the Security business for Akamai, given all the opportunities around public cloud, Akamai Inference Cloud, CIS, just in terms of like where the incremental dollars are going? How would you sort of stack rank security versus the rest of the business?

Security is very important. It's the majority of our revenue is an important growth driver for us. As we look forward, our goal is to try to do around 10%, including M&A. We're always looking for the right addition in terms of M&A. We've been very happy with Guardicore and Noname, they're both doing great. We've looked at 30% to 35% ARR growth this year, which is great to see, and I think a lot of runway there. Separate from M&A, we've launched our firewall for AI. I think protecting AI is really important, and they'll be -- that will become, I think, an important market where we want to be a player. So yes, it's important, a big area of investment for us. And I would say security and CIS. Those are the big investment areas.

100%. When you think about the composition of the security business, you have your traditional web security business business that you've been in for a very long time. And then you have the growth engines of that business. You mentioned Guardicore and your API Security business, which is Noname, the AI Firewall. Is there a time where there -- is there like a stabilization potential with kind of the core web security business? And is there a time where that -- the dollar growth from Guardicore core plus API Security -- is there like kind of an inflection point where those scale and we see better growth out of the security business? Or does M&A really need to be part of the equation to sustain that double-digit growth?

Yes. So the majority of the revenue, as you noted, is in the traditional businesses, Web App firewall, Bot management to stack on top of that, stopping DDoS attacks. And that is more mature. We're the market leader by far. It's growing, but it's not going to grow as fast as it did. And then you've got the newer businesses where we have market leadership position growing very rapidly. I think M&A is an important part of future. Adding new capabilities, the security landscape changing very rapidly, AI should accelerate that. And so I think acquisition, yes, is important going forward. And the key is to get the right value. We're not going to pay crazy amounts of money for something. We're very careful with what we buy.

The API Security business, you mentioned, across the $100 million run rate, what do you think -- what does the TAM sort of look like in API security? And what do you see the potential penetration of API Security within the Akamai customer base?

Yes. API security is great because there's a lot of room still in our customer base, growing very rapidly, but also outside the customer base, you don't have to have been a traditional CDN or WAF buyer to need API security. And it also has growth potential in terms of AI. So already, the latest version of the product will find your shadow AI and pretty much every enterprise has shadow AI today. And you're going to need to identify that and protect it. So it couples very nicely with our firewall for AI.

Let's talk a little bit about Guardicore, which has been 1 of the major drivers of security for the last few years. What's the runway for growth there? And to what extent is Guardicore emerging as a land product for new customers? Or is it primarily an installed-based there?

It's both. Good runway. It's something we do cross-sell into our base, but Guardicore, in some sense, even less related to CDN or WAF. It's behind the firewall solution. any enterprise really, especially if they're worried about data ex filtration or ransomware, we need Guardicore. And so we have had a lot of new customers to Akamai through Guardicore sales.

If we think about going into -- running up on about 2 minutes left, as we think about exiting 2025, and then you've got Q4 to go, so I don't want to get necessarily get ahead of that. But is it you think about like your priorities for the team in 2026? What are your hopes and aspirations in terms of whether its investment priorities, growth objectives, what is 2026 going to be about for Akamai?

Yes. It's about stronger growth about rapid adoption of Cloud Infrastructure Services, particularly with Inference Cloud, which is a new capability. continuing the strong growth in our security capabilities. Also, as you talked about adding a lot of new customers because you think about the rapidly growing products, API security, Guardicore core segmentation, cloud infrastructure services, those can be sold to a lot of customers that aren't Akamai customers today that don't worry about CDN, maybe don't worry about Web App Firewall in the sense that our traditional base does.

And so we have already started a program to greatly increase and [indiscernible] Hunting new customers at Akamai. So cross-sell is great. We're doing that, but we've got a lot of opportunity to increase our overall base, and that will be a big focus area next year.

Can you speak to like the sales motion and the -- maybe even the partner strategy around that to land -- potentially multiple products that are traditionally outside of the traditional Akamai customer base?

Yes. partners even more important, especially those products I talk about often partner-led. So that's good. They're partner friendly, which is good. So yes, I think it's -- we're in a good position now, and we're making the investments and the associated compensation structure to realize that.

In terms of the partners, should we think about these as GSIs, are these the hyperscalers, which partners are these?

They're not the hyperscalers, but yes, so GSIs, the big firms that you would think of, sometimes carriers good partners. The hyperscalers are customers, and we compete with them. Sometimes our products are listed on their marketplace, but we generally are not going to market with them.

We've gone through basically 30 minutes without mentioning the Delivery business, but maybe just one in terms of kind of the state of the Delivery business post couple of years of consolidation. What are your sort of underlying assumptions for that business in terms of pricing and the competitive dynamics? And what are your hopes for that segment of business? We just come down as a percentage of as I think it's sort of under 30% of the business today. But kind of what's the state of the Delivery business as we head into 2026?

Yes, a lot better than it was in the last couple of years. And you're right, there's been a lot of consolidation, probably at least 5 competitors are gone because they were selling below their cost. That is still happening, but less than it was before because there's fewer of them. It's still competitive. Traffic growth overall is better. The pricing declines, which we control to some extent for us, less than they were before. We do turn away some business which hurts on the traffic side, but it's better on the pricing side. So we're looking towards close to stabilization.

It's declining -- revenue is declining low to mid-single digits, probably be that way for a little while. We'd like to get it even -- overall, over time, like to actually grow it. It's not -- we've cut back a lot on the investment there, both in terms of the CapEx, half where it was before, and we're careful about the kinds of big users that we take on to make sure it's really profitable and driving cash flow for us.

Well, Tom, thank you for spending 30 minutes with Nasdaq London. Enjoy the conversation. All the best in 2026.

Thank you.

Good day, and welcome to the Third Quarter 2025 Akamai Technologies, Inc. Earnings Conference Call. [Operator Instructions] Please note, today's event is being recorded.

I would now like to turn the conference over to Mark Stoutenberg, Head of Investor Relations. Please go ahead, sir.

Good afternoon, everyone, and thank you for joining Akamai's Third Quarter 2025 Earnings Call. Speaking today will be Tom Leighton, Akamai's Chief Executive Officer; and Ed McGowan, Akamai's Chief Financial Officer.

Please note that today's comments include forward-looking statements, including those regarding revenue and earnings guidance. These forward-looking statements are based on current expectations and assumptions that are subject to certain risks and uncertainties and involve a number of factors that could cause actual results to differ materially from those expressed or implied. The factors include, but are not limited to, any impact from macroeconomic trends, the integration of any acquisition, geopolitical developments and other risk factors identified in our filings with the SEC. The statements included on today's call represent the company's views on November 6, 2025, and we assume no obligation to update any of these forward-looking statements.

As a reminder, we'll be referring to certain non-GAAP financial metrics during today's call. A detailed reconciliation of GAAP to non-GAAP metrics can be found under the financial portion of the Investor Relations section of akamai.com.

With that, I'll now hand the call off to our CEO, Dr. Tom Leighton.

Thanks, Mark. I'm pleased to report that Akamai had a strong third quarter with results coming in above expectations for revenue, margin and earnings per share. Revenue grew to $1.055 billion, up 5% year-over-year as reported and up 4% in constant currency. Non-GAAP operating margins improved to 31%, and non-GAAP earnings per share was $1.86, up 17% year-over-year as reported and in constant currency. Our business performed well across the spectrum of our portfolio with accelerating momentum for our Cloud Infrastructure Services or CIS, continued strong demand for our high-growth security products and continued stabilization of our delivery revenue.

We're especially pleased with the greater recognition of the strength of our distributed platform and differentiated strategy among customers and industry analysts. For investors who may be less familiar with how we have transformed Akamai's business model from the CDN pioneer to a leader in cloud security and distributed cloud computing, I encourage you to read the new report on Akamai from IDC, titled, Akamai: Navigating the Cloud Frontier - A Transformation from CDN to Distributed Cloud Provider. It offers an objective third-party perspective of how Akamai has evolved our strategy and our key differentiators in security and distributed cloud computing for AI inferencing at the edge. You can find the report on our website.

As a great proof point for the advantages offered by Akamai's uniquely distributed compute capabilities, the top three cloud providers in the U.S. are all now using Akamai Cloud Infrastructure Services. And in Q3, one of them signed an expanded multiyear renewal that solidifies Akamai's position to be their premier distributed cloud computing provider. This customer has a dominant position at the core of the Internet, and uses our widely distributed managed container service to get their business logic closer to end users for superior performance.

Our revenue for cloud infrastructure services in Q3 was $81 million, up 39% year-over-year as reported and in constant currency. That's an acceleration from the 30% growth rate we had in Q2. We signed many new and expanded contracts for our Cloud Infrastructure Services in Q3, including with a major global appliance and consumer electronics manufacturer in South Korea, a multinational financial services company in Singapore, a leading U.S. developer of analytics software, a U.S.-based supply chain planning software vendor, a European cybersecurity provider, a major U.S. airline, a leading American video game company, a leading media and entertainment company in India, and one of the largest media companies in the world. Also, a multinational gaming company in Japan contracted for our Cloud Infrastructure Services as part of a larger $37 million 2-year renewal for an array of Akamai products and services.

Last week I was with our team at the AI Industry Conference, NVIDIA GTC, where Akamai took a major step towards the future with the launch of Akamai Inference Cloud, our platform to support the growing demand to scale AI Inference on the Internet. With the rise of AI, the Internet is undergoing a fundamental shift in architecture. The Internet we're building today is driven by AI, where human intelligence is supported and augmented by intelligent systems powered by AI Inference.

Akamai is positioned to power inference the way we power the web by bringing inference physically close to users. This will enable faster performance and global scale to support intelligent applications worldwide.

When the web was first taking hold, the need for performance and scale is what catalyzed Akamai's founding. Akamai helped to end what was known as the World Wide Wait, enabling the Internet to scale to provide real-time services to billions of people around the world. Subsequently, we introduced web security as a cloud service, enabling the web to be used safely for myriad critical applications such as banking and commerce. We see the same need for performance, scale and security playing out again with AI inference today. By combining highly scaled GPU and compute capacity with Akamai's unparalleled global reach and security at the edge, Akamai Inference Cloud enables intelligence to run instantly, securely and exactly where it's needed, right next to the user, agent or device. This is how Akamai can power the new generation of AI applications, conversational, personalized and Agentic. All designed to scale in real time to meet unprecedented demand.

As we look at AI investment cycles, we see the market at a transition point. Until now, the AI story has largely focused on training, the initial creation of AI models from massive amounts of underlying data. To train foundation models, AI pioneers have relied on hyperscale clouds and their centralized data centers with their enormous concentrations of compute, power and capital. We believe that AI Inference or the execution of queries against a trained model is the new frontier, one that requires purpose-built infrastructure to enable distributed low-latency, globally scalable inference at the edge, with response times measured in a few tens of milliseconds.

As AI systems are adopted at scale, we expect the growth of Inference will drive enormous demand to this new intelligent layer of the Internet. And we're not the only ones who see it coming. Fortune Business Insights noted in a report on the rising global AI Inference market that due to rising demand for real-time low-latency AI processing near data sources Edge Inference leads the market and is projected to grow at the highest CAGR of all AI inference models. And as NVIDIA's Founder and CEO, Jensen Huang said: When we launched Akamai Inference Cloud at GTC, "Inference has become the most compute-intensive phase of AI, demanding real-time reasoning at planetary scale. Together, NVIDIA and Akamai are moving Inference closer to users everywhere, delivering faster, more scalable generative AI, and unlocking the next generation of intelligent applications."

Akamai Inference Cloud brings together Akamai's globally distributed architecture and expertise with NVIDIA's Blackwell AI infrastructure to provide the computing needed to unlock AI's true potential. The service is available today with 17 locations around the world, and we're building out more points of presence as customer demand grows.

One of our initial customers, Monks, the European digital-first marketing technology services and consulting company said "with Akamai Inference Cloud, we will accelerate the delivery of key capabilities, including identifying players and plays and delivering tactical insights to coaches while the game is still happening. That's only possible by distributing advanced GPUs to the edge, and we believe it will transform how we approach sports broadcasting and immersive fan experiences."

Another customer, Harmonic, whose technology helps to distribute video content for television and the Internet said, "Akamai Inference Cloud will allow us to run larger parameter, more capable models locally, expanding the number of functions we can deliver cost effectively within the same compute instance to deliver fast response times, sophisticated personalization and more enriching video content."

At Akamai, we believe the technology ecosystem that enables the AI revolution will require multiple providers of AI infrastructure, the hyperscalers, NVIDIA and also Akamai with our unique distributed capabilities and our unparalleled expertise at the edge. We are very excited about what lies ahead.

The edge, of course, is also where Akamai deploys our security solutions. And as customers speak with us about their plans for AI inferencing, they tell us they see valuable synergy between Akamai's security and delivery product lines and our cloud computing capabilities. and how they trust Akamai to help make the Internet faster, more reliable and secure for their businesses.

Akamai security growth in Q3 continued to be driven by strong demand for our market-leading segmentation solution and by rapidly growing customer adoption of our API security solutions. Combined, these high-growth security products grew revenue 35% year-over-year as reported and 34% in constant currency. Our segmentation wins in Q3 included a $3 million expansion contract to give one of North America's largest health care technology companies the visibility and control they didn't have before. A $1 million contract with a European insurance group that is also a net new Akamai customer, a multiyear contract with a large insurance company in Korea. And an expansion contract with a large bank in Mexico to extend their initial deployment across their operations in Latin America.

In Q3, we also continued to see growing interest in our market-leading API security solution. As organizations shift towards an API-first strategy and expand their use of AI applications that rely on APIs in a fundamental way, adopt tools that enable them to discover and monitor deployed APIs and to manage risk. Comply with stricter data protection regulations, especially in Europe, and respond to public reports of API-related breaches that have raised awareness and increase the stakes for financial and reputational loss.

Akamai API security wins in Q3 included a $7 million contract for API security with one of Europe's most important banks. As part of a $31 million multiyear commitment for security and compute. Adoption of API security as part of a $20 million expansion contract with one of the world's largest airlines. An expansion of API security as part of a $42 million contract across the breadth of our portfolio with one of the world's largest software companies. A $2.6 million contract with one of the largest life insurance groups in Asia. The displacement of a competitor at a FinTech payments provider in Brazil. And we also signed 7-figure contracts for API security with 2 of the 10 largest financial institutions in the U.S. and one of the largest banks in Canada.

We're also pleased to note that for the sixth straight year, Akamai has been named Customers' Choice in Gartner's latest Voice of the Customer report for cloud web application and API protection. Akamai was also recognized as Customers' Choice in Gartner's Voice of the Customer report on online fraud detection.

Before I turn the call over to Ed, I'd like to express my gratitude to our employees for their great work in Q3, and for making Akamai a great place to work. In recognition of your efforts, Glassdoor named Akamai as one of their top 50 best led companies of 2025.

Now I'll turn the call over to Ed to say more on our Q3 results and our outlook for Q4 and the year. Ed?

Thank you, Tom. I'm pleased to report that we delivered strong third quarter results with total revenue of $1.055 billion, up 5% year-over-year as reported and 4% in constant currency. We also had another quarter of very strong bottom line performance with non-GAAP EPS of $1.86 per share, up 17% year-over-year as reported and in constant currency. Our strong EPS results were driven by higher-than-expected revenue and strong execution across the board.

Moving now to revenue. Compute revenue, which is comprised of the fast-growing Cloud Infrastructure Service or CIS solutions that Tom mentioned earlier, and our other cloud applications, or OCA was $180 million, up 8% year-over-year as reported and up 7% in constant currency. As a reminder, in Q3 2024, we recorded a $7 million onetime benefit related to the release of some deferred revenue in conjunction with the expiration of a long-term legacy compute contract. This revenue was part of our other cloud application products, and it had a 5 percentage point impact on a year-over-year total compute revenue growth rate.

Total compute revenue was driven by continued strength in CIS. For Q3 2025, CIS revenue was $81 million, accelerating to 39% growth year-over-year as reported and in constant currency, a nice step up from approximately 30% growth last quarter. As a result, we continue to expect CIS ARR year-over-year growth in the range of 40% to 45% in constant currency at year-end.

Security revenue was $568 million, up 10% year-over-year as reported and 9% in constant currency. Revenue from our high-growth security products, by that, I mean, API security and Zero Trust Enterprise Security was $77 million, an increase of 35% year-over-year, and 34% in constant currency.

Given the continued strength in our API security business, we now expect to exit 2025 with a run rate of approximately $100 million for that product line on both an as-reported and constant currency basis. Finally, we continue to expect the combined ARR for our high-growth security solutions to increase by 30% to 35% year-over-year in constant currency for 2025.

Moving to delivery. Revenue was $306 million, down 4% year-over-year as reported and in constant currency. This result was slightly better than expected, marking another quarter of improved trends in our delivery business.

International revenue was $525 million, up 9% year-over-year or up 8% in constant currency, representing 50% of our total revenue in Q3.

Foreign exchange fluctuations had a positive impact on revenue of $4 million on a sequential basis and positive $8 million on a year-over-year basis.

Moving to profitability. In Q3, we generated non-GAAP net income of $269 million or $1.86 of earnings per diluted share, up an impressive 17% year-over-year as reported and in constant currency, and $0.20 above the high end of our guidance range. Finally, our Q3 CapEx was $224 million or 21% of revenue as we continue to invest in our fast-growing CIS business.

Moving to cash and our capital allocation strategy. As of September 30, our cash and cash equivalents and marketable securities totaled approximately $1.8 billion. During the third quarter, we did not repurchase any shares. As a reminder, year-to-date, we spent $800 million to buy back approximately 10 million shares, marking the largest annual buyback in our history.

As it relates to our use of capital, our intentions remain the same to continue buying back shares over time, to offset dilution from employee equity programs, and to be opportunistic in both M&A and share repurchases when market and business conditions warrant.

Before I provide our Q4 and full year 2025 guidance, I want to touch on some housekeeping items. First, as in prior year, seasonality plays a significant role in determining our financial performance for the fourth quarter. Typically, we see higher than normal traffic from our large media customers and a pickup in seasonal online retail activity from our e-commerce customers. However, both are difficult to predict.

Second, regarding Q4 operating expenses, as is typical for our business, we expect it to be higher than in Q3. The main driver is the seasonal jump in sales commissions as our most successful reps achieve their annual quota accelerators. Finally, on July 4, 2025, the One Big Beautiful Bill Act was signed into law, introducing significant provisions, such as the permanent extension of certain expiring Tax Cuts and Jobs Act provisions, international tax framework modifications and it restored some business tax benefits. This new legislation, however, has not had a material impact on our tax rate in 2025.

With those factors in mind, I'll move to our Q4 guidance. For Q4, we are projecting revenue in the range of $1.065 billion to $1.085 billion, up 4% to 6% as reported and up 3% to 5% in constant currency over Q4 2024. At current spot rates, foreign exchange fluctuations are expected to have a negative $5 million impact on Q4 compared to Q3 levels, and a positive $11 million impact year-over-year. And for the full year, at current spot rates, our guidance assumes foreign exchange will have a positive $13 million impact on revenue in 2025 on a year-over-year basis. At these revenue levels, we expect cash gross margins of approximately 72% to 73%. Q4 non-GAAP operating expenses are projected to be $322 million to $331 million. We expect Q4 EBITDA margin to be approximately 42% to 43%. We expect non-GAAP depreciation expense to be between $143 million to $145 million, and we expect non-GAAP operating margin of approximately 28% to 30% for Q4.

Moving on to CapEx. We expect to spend approximately $171 million to $181 million. This represents approximately 16% of our total projected revenue.

Based on our expectations for revenue and costs, we expect Q4 non-GAAP EPS in the range of $1.65 to $1.85. This non-GAAP guidance assumes taxes of $57 million to $60 million based on an estimated quarterly non-GAAP tax rate of approximately 18% to 19%. It also reflects a fully diluted share count of approximately 147 million shares.

Our complete guidance for full year 2025 is available in today's press release, but let me walk you through the highlights for now. For the full year, we expect total revenue to grow 4% to 5% in constant currency. Non-GAAP operating margin of approximately 29% to 30%, and EPS in the range of $6.93 to $7.13.

In closing, we continue to be very pleased with the performance of our Cloud Infrastructure Service and high-growth security solutions. We are very excited about the potential of the Akamai Inference Cloud as we extend AI to the edge.

With that, I'll wrap things up. And Tom and I are happy to take your questions. Operator?

[Operator Instructions] Today's first question comes from Mike Cikos at Needham & Company.

Congrats on the solid quarter and the execution you're talking to. I just wanted to come back to the segment commentary real quick and just to make sure that I'm understanding the guidance on the various components properly. For security or compute, are we at this point reiterating the growth that we had spoken to last quarter? I think you were looking for computing the magnitude of 15%, call it, security around 10%. Can you provide any final parameters there? And then I do have a follow-up.

Mike, this is Ed. Yes. So with security, we're calling for about 10%. We don't give guidance by quarter. So for the full year, we obviously reiterated our ARR guidance. So that hasn't changed. And as we said before, with compute, will be maybe a touch under 15% for the year as some of the bigger contracts ramped up a little bit later in the year than we had expected, but definitely picking up momentum in CIS though.

Great. And one more for Tom, but just to add before I leave it there. Are there any [ tea leaves ] or guardrails you can give us as we're thinking about '26, just given how '25 has played out at this point?

No. So we'll give you guidance in '26 in the February call. The only thing I would say is that the momentum that we're seeing in CIS, especially with the new AI Inference Cloud, we're getting a tremendous amount of activity and demand for that. So there's a very good chance we could accelerate growth in our CIS business next year.

That's excellent. And that actually feeds right into my next question beautifully then. For Tom, I know you spent a good chunk of your prepared comments on Akamai Inference Cloud, which has to be intentional, right? But you spoke to a couple of different initial customers, Harmonic being one of them. You spoke to the 17 locations, and this being available on a global basis. Can you just help us think about where we are in that inference curve, right? We've been in frontier pioneer model world for a couple of years now. It feels like things are happening at the inference level, but what's the boots on the ground view that you have today?

Yes, first inning of a very exciting game. There's just so many customers that want to adopt AI applications and do inference for -- applications that need to be fast for real users. An example would be a commerce company, and they want to present the right products, personalized images to the user and then if the user is interested in something, they really want to show not only the product, but the user wearing the product and not just in a picture, but in a video. And in the video, they want to show the user in a context and environment that makes sense for the user. We've got customers with AI-powered toys. Customers with -- that want to do robotics and they need really fast control and response. And so they need to do the compute, the inference close to the end user at the end.

Media Workflow, people are going to see different versions of games, different versions of movies. It will all be personalized. And this needs something like the NVIDIA Blackwell 6000, which is very powerful, but also well suited to run at our edge. And that's why the partnership with NVIDIA makes so much sense. So we're in the first inning. We're just launching the service in the last couple of weeks, but we're seeing a very strong amount of interest among our customer base. This is exactly what they want.

Our next question today comes from John DiFucci with Guggenheim Securities.

This is Lawrence Vensko on for John DiFucci. You've communicated your initiatives related to the go-to-market strategy with the hiring of hunters and experienced specialists to support new business sales and security and compute. Can you just walk us through how you're thinking about the hiring of additional reps for the remainder of this year? And how should we think about the ramp period of these sales reps before they become fully productive?

Yes. We've made a lot of progress in the go-to-market transformation, and that will be continuing through the early part of next year. I would say by the time we get into Q2, a lot of the transformation will be done. There will be continued hiring from now through at least the first half of next year to have increased sales capacity for hunting, both in security and obviously also in compute. So that's an ongoing effort, but the transformation part will be largely done by the beginning -- towards the beginning of next year.

And our next question comes from Rishi Jaluria with RBC.

Wonderful. Great to hear. So a lot of the momentum, especially on the compute side. Maybe two questions from me. I wanted to first start with -- you talked, Tom, about some of the deals that you're signing with the hyperscalers and where they're using you for more edge inferencing. At the same time though, as we know from all the hyperscalers, everyone is suffering through capacity constraints right now and everyone is trying to find any resources wherever they can as we think about just advancing AI on both the training and inferencing/reasoning side. So maybe help me understand what gives you confidence that as we get through this capacity constrained era, regardless of how long it lasts, that Akamai will still be well positioned to benefit from that. And that's not something that some of the hyperscalers or partners might think about bringing in-house as the opportunity materializes itself, especially with edge inferencing and more capacity gets built out? And then I've got a quick follow-up.

Yes. Akamai has a unique platform. Nobody is like us. We have over 4,000 points of presence. We're in -- already in 700-plus cities. And that's unique. And we already support our Edge Worker solution, our function of the service in every location. In fact, one of the hyperscalers is using us for ad market, to get their ad logic a lot closer to users, another hyperscaler using us for API orchestration again, needs to be fast and close to users. We talked about our managed container service. A third hyperscaler is using our managed container service for their media workflow, and again, because it's closer to users. So it's not a capacity constraint issue. It's not that the hyperscalers have run out of capacity. Certainly, that's not the case. It's because our platform is different and we can make their logic, their compute logic run faster for users because it's closer.

And obviously, we compete with their cloud businesses to win that business, but it's because of our better performance. And the good news is we don't have to go deploy 4,000 new reasons or start new deployments in 700 cities because we already have this distributed platform. We're augmenting several of the regions that we've talked about with the new Blackwell 6000, and that's pretty exciting. And we will continue to do that as we scale that business.

Yes. Understood. No, that's helpful. And then maybe if I then shift to -- just thinking about the API Security business. Obviously, you've done well with that. In addition, just as we think about kind of the evolving paradigm in applications, right, shifting to agents and multi-agentic systems. And obviously, we're seeing the rise of things like MCP and A2A as a way to bring those altogether. Are there opportunities to take what you're doing in API security today and start to extend it now into these new kind of Agentic protocols?

Absolutely, and we're already doing that. For example, of API Security and detecting the APIs, detecting the new AI agents and apps that the shadow AI that enterprises have running and don't know. And then, of course, to protect that with our new AI firewall service. And this is one of the reasons that we're so excited about our API security business. So not just to protect the legacy APIs, but the new Agentic web and to protect -- and to identify and protect all those new inferencing applications.

And our next question today comes from James Fish with Piper Sandler.

Interesting stuff on the compute side. But how should we think about the required CapEx for inferencing here and relationship you have now with NVIDIA? Has that evolved over the last year or so? And for you on that point, how should we think about compute gross margins here moving forward over the next year to 2?

Jim, good questions. So we've done an initial deployment with NVIDIA. We're in about 17 cities or so, and we've got that up and running. That was in our Q4 CapEx. It will be informed by demand. And what we're seeing is initially some customers coming to us with some pretty large request. So the CapEx will very closely be followed by revenue, and it will be informed by demand. So we had given you sort of a metric of about $1 of CapEx as a $1 of revenue. It's about the same here. It could be a little better depending on utilization. As far as margins go, it will be very similar. I would expect obviously, as we do a deployment, let's say, we get some orders for $50 million, $100 million, whatever it may be, and you're putting out a little bit more CapEx for further down the line -- pipeline stuff.

It'll be a little inefficient at first. But once it gets to scale, I would expect to see similar gross margins to what we see in compute and maybe even a little bit better just given the scarcity in the marketplace. So you could potentially have slightly better margins. But we also expect to get pretty good operating leverage and scale out of it as well. So as we get -- as we scale up, I expect us to get -- help the overall gross margin of the company as we get to scale.

Got it. And going back to the go-to-market side of things. This quarter did seem to have a lot of callouts around, I'll say, cross-sell or packaging across all 3 segments. Is there a way to understand what incentives you're giving to your sales team? Or what percentage of the business is whatever you want to call it, ELAs, others call it pool of funds. What percentage of customers are using all 3 segments at this point?

Yes. Good question. So I would say one of the things we did and what we're seeing in the field this go around here this year is we incented longer-term deals. So we're definitely seeing an elongation of our sales cycle, which is good -- not sales cycle, but our customer average length, which is good. So we're seeing longer-term deals, larger deals, et cetera. And you'll see that reflect in the RPO metric that will come out tomorrow in the queue. So that's one of the drivers.

We're also seeing larger deals, not necessarily pool of funds or ELAs. We do have a few of those, but those are necessarily drivers. We are seeing the adoption of compute definitely pick up in the installed base. So I don't have a metric to give you today on what the percentages that use all three, there's a decent amount of customers. I'd say we have some compute-only customers, but largely speaking, most of the customers are either a security customer or a compute customer. About 74%, 75% of our customers are both security and delivery customers. So that's been pretty consistent over the years, maybe come up a little bit. But yes, we don't have those -- necessarily those pool of funds deals or ELAs where you've got a big commitment without product specifically addressed or volume against a particular product line.

And our next question today comes from Gabriela Borges with Goldman Sachs.

I wanted to follow up exactly where the prior commentary left off. So in terms of the potential phases of inference deals, Ed, I think you mentioned off hand there in the order of magnitude, $50 million to $100 million. I'd just love to hear a little bit more about that. Do you think that the pipeline can support or does the pipeline already show deals of that sort of order of magnitude because I could see how that could be a meaningful contributor to your growth algorithm into next year?

Yes. What I was saying there was, let's say, if you had a large -- because we have had some -- interested in some fairly large deals. If you had a deal of that size that you would buy more CapEx than that. So let's say, if I got it in order for 50 to 100 or something along those lines, I'd probably be buying more CapEx because there's more demand to follow. But the point I was trying to make there is that we are seeing some customers reach out to us for some pretty sizable deal sizes in terms of larger than what we typically see in compute today. So it would be -- I'd say, just in general, just going to be larger deals to start for sure.

Yes, super interesting. The follow-up I have, either for yourself or for Tom is on the delivery business. You're putting up the stabilization and the growth rates that you've been talking about for a couple of quarters now. Remind us, is there idiosyncratic opportunities when the cohorts come up for refresh, depending on the year that help influence those pricing dynamics and remind us why pricing is more stable now than it has been in the past?

Yes. So good question. We will call out generally when we have -- we've got today now six customers that are 1% or greater of our revenues. So we don't have a ton of concentration risk. But when those 6 or our top 10 as they come up for renewal in 1 quarter or within a 6-month period, that will definitely have an impact on revenue. We don't -- we didn't have that this year. We don't expect that next year. We just have the normal revenue renewal cycles that we typically have. We are pushing and incenting our sales reps for longer -- longer contract length and we are seeing that, which is good, including with some of our larger customers, which is good as well.

As far as the pricing dynamics go, we have seen some stabilization in the larger media deals where you see probably the biggest impact on revenue. Still some price pressure in sort of the larger base, if you will, but nothing that's out of the ordinary as a matter of fact, it's a little bit better than what we've seen historically. And in terms of what the drivers are of that, I think it's just -- there are fewer competitors in the marketplace. A lot of them have gone out of business or in some cases, just exited the business if they're an existing different company that's not just focused on delivery. So that it may have something to do with it as well.

And our next question comes from Frank Louthan with Raymond James.

So it's been a pretty good year for delivery overall. Anything really driving that and anything early that you see in the trends in Q3, to give us an indication of how that seasonality is shaping up. And then with the larger contract deals you're getting with customers, any risk to the sales cycle along dates?

Yes. Usually -- so on longer sales -- sorry, I keep saying cycles -- I mean longer contracts, no, it doesn't necessarily elongate the sales cycle with existing customers. With newer customers, obviously, your sales cycle is a bit longer, but not with existing customers, no. In terms of the dynamics of the business, Q3 typically is a little bit lighter on the seasonal quarter for delivery, just given the dynamics of content, sports seasons, vacations, that sort of stuff. So we saw kind of a normal pattern there. But traffic growth is probably the biggest driver coupled secondarily with the pricing declines have moderated consistently throughout the year. So it's really a combination of the two.

And our next question today comes from Sanjit Singh with Morgan Stanley.

Congrats on the solid results. I wanted to go back to the compute business, and it's great to see CIS accelerating. In terms of that other $100 million in that portion of the business, I know there's headwinds around the storage piece and some -- I think there were some video optimization close that you guys were transitioning to our partner. In terms of when does that potentially stop being a headwind to overall growth? Just to have an update on that portion of the reference.

Yes. Those would be in our other cloud compute applications segment. And that's not really the focus of the business. There's still some more that will happen there, but more or less that was flat quarter-over-quarter. It's not where we're investing, not where we'll see the growth. I think you want to focus on the CIS, the Cloud Infrastructure Services, that's where all the excitement is and where all the future potential growth is. And so the -- in fact, going forward, we may even just separate the two and report on CIS because OCA is pretty small, pretty stable. And the big driver of our future growth is Cloud Infrastructure Services. And that's where we have our products like Edge Workers for Function as a Service, and 4,000 POPs, manage container service that was being used by a hyperscaler for their media workflow and AI. So Akamai inference Cloud with huge potential growth. And that's all part of Cloud Infrastructure Services. So probably that's what you want to be thinking about when you think about Akamai Compute.

Yes. So -- and that's what we're looking at as well. I just want to get a sense of when -- maybe when some of the revenue headwinds start to fade. But fair enough. On the Inference Cloud opportunity, what do you sort of envision? Is there a security attach opportunity associated with Akamai Inference Cloud and love to see -- care about how you're thinking about the security attached to Akamai Inference.

Yes, absolutely. And that's a great question because as you're putting these applications and models out there, they got to be secured. In fact, they have even greater vulnerabilities than a normal application or API would have. And so you're going to need API security, you're going to need like an AI firewall. And those are things that we have market-leading solutions for it so that we can build that around your inference engines and your models that are on Optimize Cloud.

And our next question today comes from Fatima Boolani with Citi.

Tom, I wanted to ask you a big picture question. As we think about the mix of traffic and let's just call it your traffic estate that runs through your pipe. When you think about the mix between classic video delivery and OTT versus enterprise, but now maybe even AI and some consumerized AI applications potentially driving uplift there. I'm curious how you're thinking -- well, what the state of affairs is today from a traffic mix perspective with the rise of AI and bots. And how should that modulate over the next couple of years and thinking about how that dovetails into your delivery and CIS business. Would love your perspective on that. And then I have a follow-up for Ed, please.

Yes. Interesting question. Today, the vast majority of the traffic would be video, either on-demand or live and software downloads. Now as you think about the future, you're right, sites that -- by commerce sites that might not have had any video or much video in the past, they're already thinking about how to make the entire experience be immersive with video. So that as you go to a page, it starts with a video of you wearing something they think you'll like to buy. And that does increase the traffic. Also, even the traditional video, live sports. As we talked about, we already have a customer that -- the feed that I see when I'm watching the game is going to be different than the feed you see or things -- it will be customized to the person. And so AI is going to be used even to watch traditional video to give a better, more personalized experience.

And who knows if you ever get people wearing devices that give you augmented reality, well, that drives a lot of new traffic. So I do think that you're right that there is a reasonable prospect that as AI becomes more prevalent that it will impact the traditional delivery business in a way that will be favorable to Akamai.

And Ed, this is a back to basics question, piggybacking on some of what you discussed earlier. But could you just walk us through the relative high-level gross margin and gross profit profile of the business from a segmentation perspective. And really, the spirit of the question is you've seen very nice operating leverage in the business in spite of the fact that I would have thought that the delivery business is perhaps got the most inferior gross margin profile relative to the other segments or revenue segments rather. So I would love to kind of get a quick recap of the relative gross margin profiles and some of the puts and takes, especially as it relates to some of the CIS business and the CapEx requirements that you talked about earlier.

Yes, sure. So I'd say, we had shown some margin back a couple of years in terms of the 3 segments with security. And from a gross margin perspective, kind of high 80s. Compute was in the low 70s and delivery is probably high 60s, if I'm remembering it correctly. It's probably still in that sort of ballpark. And I think the business in general will run in the kind of low 70% gross margins might be down a point or 2 as we're building out scale for -- you got a little bit of a timing issue between the demand and the revenue and the build-out because you have 2 pieces of the build-out. You've got the colocation where you're -- especially with AI Inference will be buying more -- larger colo buys, so you have some unfavorable accounting potentially.

So you might have a little down trend in gross margin and compute for a bit. But we do think you can get to that sort of low 70s gross margin. We are seeing from a CapEx perspective, even with the AI Inference and what we're selling with the AI Inference Cloud, that dollar of revenue to dollar of CapEx is a reasonable proxy for what we're seeing. And like I said, it could be a little bit better with the GPU as a service because there is a scarcity there. So there's a pretty decent market price set for that today, but that could be a little bit better. So I do think that, overall, you get good leverage because our sales force, the way we're going to be designed, you've got some specialists, but it's reasonable in terms of the specialists and our sales force will be able to sell everything and bring specialists and so we should get good operating leverage across the entire business. So it's not just delivery that gives you good operating leverage, which it does, but it's all 3 products that will give you good operating leverage.

And our next question today comes from Jonathan Ho with William Blair & Company.

With CIS, are these mostly sort of new compute and capacity deals that you're winning? Or is there also sort of a growth coming from existing workloads that are being migrated onto CIS? And related to that, is there may be an opportunity to benefit from outages at AWS and Azure as customers look for greater resiliency?

It's a combination. But the outages you raise is a really important point. And people don't think about it very much. And this is important whether you're buying compute or delivery or even security. In fact, maybe even more important in security. Our goal at Akamai is to have five 9s of reliability. And that means that over 2 years, the site or the app is down less than 10 minutes collectively from all causes, including attacks. And we're achieving that today, and we know that because some of the world's major banks, the regulators, want to see Five9. And we put a tremendous amount of investment into our platform to make sure that when we update metadata or software or a customer updates their metadata, it doesn't have some unintended consequence that blows up the platform. And we see what happens with our competitors who don't make that investment. And for example, in security companies that we compete with where some of them, they're down for an hour every quarter, and that's just a disaster for a bank. They can't survive like that in today's world. So reliability is critical. And that's an area that really is a strength for Akamai. And so I think that's super important. That's a great call out.

And just maybe as a follow-up. When we think about your security business, can you talk a little bit about the penetration rates for opportunities such as segmentation and as well as how your AI security products are doing?

Yes. So there's a lot of room for growth, a segmentation and API security. And also, as we talked about earlier with API security, get a whole new market around the AI Inference engines and models and agents that are going to need special security. Obviously, it's early days on our specific protection for AI firewall, but very strong interest in the customer base, hundreds of prospects and lots of proofs of concept and starting to get the first customers now adopting the solution. And I think that will be a pretty quickly evolving landscape in terms of protecting the models, the inference engines and the agents.

And our next question today comes from Will Power at Baird.

Okay. Great. I guess two questions. Tom, maybe starting with you. I guess it would be great. I know you've touched on this a fair amount. As I think about the Akamai Inference Cloud and the NVIDIA partnership, I wonder if you could just speak to the medium, longer-term strategic importance of it. Is it principally access to black well GPUs? Or are there other parameters and facets of the partnership to be aware of anything on the go-to-market side, et cetera? Then I have a question for Ed.

I think the really important thing is it's providing very strong compute capabilities to support AI close to the users, the data, the devices, the robots and the agents, and that's what's unique about the offer. So you're combining Akamai's distributed platform, which is the most distributed by far in the world with leading GPU capability. The new 6,000 are very powerful. You can support models with hundreds of billions of variables. They can do very cool things that weren't possible before, literally like showing a user who comes to the site in the site in a video. Very cool stuff. And that's unique, I think, with Akamai. And we do have a good relationship with NVIDIA, and we are working together in terms of commercializing the capability with customers and the ecosystem.

Okay. Great. And I guess, Ed, nice upside on operating margin performance and it sounds like largely expecting that to continue into Q4. So I'd just be great to get any color as to kind of what's driving the upside relative to prior expectations. And kind of this 30% adjusted operating margin level, is that kind of the right framework to think about as we kind of move into next year?

Yes. So we'll give you guidance for next year in February. I'd say execution, I mentioned that at the top of the call. Some of that's driving additional revenue. Some of it is just -- even with our development efforts and the capitalization of labor is a good gauge for productivity. That was a bit higher this quarter. So we were more productive there. The team has done a nice job even in procurement with our -- some of our vendors in terms of getting better pricing. We're modernizing some of our back office, so we're able to retire some older systems and consolidate on to one. Oracle is a good example of where we're consolidating on to. We've got great execution with our team that's building out the network in terms of getting better pricing on co-location and bandwidth and that sort of stuff. So I'd say it's just good execution across the board.

And our next question today comes from Patrick Colville at Scotiabank.

I guess one quickly for Dr. Tom and then one to Ed, if possible. So Dr. Tom, I guess, my question for you in regards to the NVIDIA partnership, there was some releases last year and Akamai introducing NVIDIA GPUs out of the Edge. So just help us understand what is new this year in late 2025 versus last year with this NVIDIA partnership.

It's much deeper. And now we're deploying the Blackwell 6000, and that's a huge leap from where we were with the 4000 in terms of capabilities, and we're deploying them much more broadly, and a much stronger partnership and relationship. So it's a world of difference and that enables the Inference Cloud and all the applications that we talked about.

Okay. Crystal clear. And congratulations on that exciting partnership. Ed, if I may, just a numbers question. I mean, the first caller kind of asked you about your soft guidance for the different business lines. I think your answer, if I was not mistaken, was 10% for security, just under 15% for Compute. So if I understood that rightly, that implies that delivery probably takes a leg back down again in 4Q. Am I thinking about that the right way? And if so, is that conservatism? Or is there something that we should know about as to why growth in delivery would inflect back down?

Yes. We gave you a pretty wide range there, and delivery is hard to call in Q4. So I wouldn't read it as we're expecting anything negative necessarily with delivery. The seasonal aspects of the quarter don't really manifest themselves to around Thanksgiving. So it's hard to say what the retail season will look like and what the sort of end of year media season will look like as you get new devices online. So there's a variety of outcomes you can get this quarter.

And yes, what I said was we should get to about 10% for Security, a couple -- a little bit below 15% on Compute, but that actually bodes well for next year because it's really just a timing issue and then delivery is sort of the fill in. So depending on where you peg your model on that range, you'll get a variety of outcomes for delivery.

Thank you. And our next question today comes from Rudy Kessinger with D.A. Davidson.

This is Andres Miranda for Rudy. Coming back to the beginning of the call, you said that you have now the 3 larger cloud providers using CIS. That implies to us that, that wasn't the case before. Can you just confirm that? And if that's the case, did you just sign one this quarter? And what is the upside moving forward? How should we think about the 3 large cloud providers.

I didn't completely get the question. Can you repeat the question that you're asking?

Yes. So like at the beginning of the call you said you had the three larger cloud providers now using CIS. Does that imply that, that wasn't the case before? So did you just sign one of them in the quarter? Or what is the upside moving forward for CIS?

Good. We signed the third -- we just signed the third one. And the first one signed a much larger contract. So one got a lot bigger, and one is new. And so now we have all three. Is that what you're asking?

Yes. Thank you.

And our next question comes from Tomer Zilberman with Bank of America.

I actually want to go back to Security and talk about the demand you're seeing there. If I take the commentary from some of your peers, they're talking about an acceleration in the underlying demand environment, but when I look at your trends, you've kind of been growing 9% to 10% the last few quarters and you called out 10% for the year. So I just wanted to see kind of the puts and takes of what you're seeing in terms of demand and how you're thinking about that as you go into next year.

Yes. So I think what we're trying to highlight for you today is we're seeing significant demand for our API Security and our Guardicore Zero Trust platform security products. API Security, in particular, if you go back in time, we closed our acquisition of Noname back in June. So this is the first quarter of full organic year-on-year growth. And we more than doubled. So we're going to exit at about $100 million run rate. So we're seeing an amazing demand for API security and continued strong demand for Guardicore.

So -- and also you have -- as Tom talked about, with API Security and what's happening with the Agentic Web and the opportunity for future growth, if you kind of look out into the future, there's a lot of runway for API Security and the penetration rate inside of our base for both products is relatively low, and we've got a great track record of getting very high penetration. So we've got some products that are been around for a while that are still growing, but it's not growing as quickly. So I think you missed the opportunity, if you just focus on the overall number. You're really going to look at the two fast-growing products, the underlying demand for those products and also our position in the market, we're very well positioned in the market with leading market leading products.

And our next question comes from Jeff Van Rhee with Craig-Hallum Capital.

This is Daniel Hibshman on for Jeff Van Rhee. Congrats on the quarter. Maybe one just to open up on the delivery trajectory and what's being expected there. Now, is this a sort of line where as always the growth rate is kind of always be in constant flux just based on the market dynamics that year? Or do you think the trend line we're stabilizing around here kind of in the single-digit growth -- a single-digit decline is sort of a trend line to be expecting into the future? Just kind of long-term vision on how you expect that to behave secularly.

Yes. Good question. So there's always seasonality just in the trends on traffic on the Internet. Q4 is always a larger seasonal quarter. You've got the return of college football and the NFL and you've got new content for new TV shows and that sort of stuff. So that always drives more traffic at new devices coming online. You've got a big holiday shopping season. So that's very typical to see a jump in traffic there.

Typically, Q1 might be a little bit lower than what you saw in Q4. Q2 generally in line with Q1. And then Q3 tends to be a little bit seasonally softer just traditionally because of not as much sports going on. You don't have this -- you've got no new shows coming out and people are on vacation, not using the Internet as much. So I don't see any change in that sort of general trend on the Internet.

As Tom talked about, with these AI agent applications could create more video traffic, just more API traffic. So there's a possibility for a leg up there. We've had weak gaming over the last couple of years, some big titles are in the backlog. That actually helps console recycle refreshes help as well. So we're due for one of those in the next year or 2. So I think the demand trends could get better over time. But we've been pretty cautious with delivery because we -- occasionally, you can get surprised. But we're very pleased with what we're seeing, and we are kind of getting back to that where we had projected the business to be kind of flat to down single digits is certainly what we've delivered this year. And if the trends remain constant in terms of what we saw this year, it's possible we could do that again.

And then just one other for me on the share repurchases, just any methodology or strategy change to call out both in regard to the record buybacks we saw in the first half. And then it looks like -- I just -- out of curiosity was looking, it looks like this is the first quarter. Akamai hasn't bought back any shares since it looks like a quarter back in 2009. So typically just a very, very steady cadence. Is this a little bit of a shift to buying more opportunistically in more concentrated amounts? Or just any thoughts if there's any change in the thinking there?

No. As Ed talked about earlier, no change in our thinking or strategy. And in fact, we bought back more shares or spent more on shares this year than ever before. So yes, we didn't buy back shares in Q3, but we bought back a ton this year.

And our next question comes from Jackson Ader with KeyBanc Capital Markets.

The international versus the U.S. script has slipped a little bit here in the last couple of quarters. Just curious which segments were strong or maybe stronger than you expected in the international segment? And just traditionally, when we see international strength versus U.S., which segments should kind of be flagged in our minds?

Yes, good question. And you're right, it did slip a little bit. One thing that did aid to the U.S. growth last year or the '25 was some of the Edgio contracts were more concentrated in the U.S. So those are not starting to anniversary. So that's a piece of it. But internationally, I would say in APJ in particular, we are seeing that sales team really embraced compute in a way that they're sort of leading the charge as far as signing up a lot of large compute deals and embracing security as well. So it's kind of strength across the board there.

APJ, in general, has been very, very strong for us. EMEA has been strong. Some of the country growth that I've seen across Western Europe was a little bit surprising in terms of a bit healthier than we've normally seen. So I'd just say, across the board, international has always been good for us. We're seeing it sort of pick back up again. And I would say compute and APJ is probably the one thing that I would call out in terms of strength.

Okay. Great. And then a quick follow-up on the Inference Cloud. You mentioned, Tom, the fact that you're close to the robots, right? You're close to the user, which I understand. But I'm just curious, I mean, is there -- would there be other things outside of kind of what I think of as like almost the Internet of Things use case that might also trigger more inference at the edge. Like if a rise in small language models were to happen or diversity of silicon. Are there kind of headlines we should use as word association that would be positive for the Inference Cloud when we see them.

Yes. I think -- well, I don't know about catchwords to look at, but pretty much everything is going to be powered by AI. I think how the web works is going to change. Everything is going to be model-based, how you interact with it will be different. There'll be applications people haven't even dreamed about yet. And a lot of these, especially when there's a bot that's doing something important like car or humans interacting in some way, you're going to want the thing to be in real time. And by that, I mean within 10 small number of tens of milliseconds. And for that, you need to be close and you're not going to want to be going far away to a big data center with a giant model that's doing a zillion things and has a lot of capacity and infrastructure that really isn't relevant to your test. And so I do think you're going to see the proliferation of small- to medium-sized models that are operating close to the user with specific tasks in mind. And I think that's why you're starting to hear the industry leaders talk about Inference as being -- well, really the next big thing and ultimately larger than the giant models and the training and the core of the Internet. And that that's where the future of the Internet and the web goes.

Thank you. And that concludes today's question-and-answer session and today's conference call. We thank you all for attending today's presentation. You may now disconnect your lines, and have a wonderful day.