Excellent. So thank you, everyone, for joining us this afternoon. My name is Keith Weiss. I run the U.S. software equity research franchise here at Morgan Stanley and really pleased to have with us from Cloudflare CFO Thomas Seifert. Thomas, thank you for joining us again. Always a pleasure. So we were on the same stage a year ago, and we were talking about how 2025 was going to be a year of inflection for Cloudflare. And it was a pretty spectacular year. There was activities that took place that weren't really on the bingo card, if you will, of investors. You signed your largest ever contract, $130 million contract, seven years in duration, the highest ACV deal you've ever signed at $42.5 million, RPO growing 48% during the year, 30% revenue growth accelerated in each of the last three quarters. The net revenue retention rate improved from 111% in Q1 to 120% in Q4. Number of paying customers up 40% year-on-year, and the list goes on. That's, I mean, by every measure, a spectacular year. So maybe just to start out the conversation, what clicked into place? I mean, it's not like you weren't performing well before, but it definitely feels like you were correct, right? There was an inflection that took place. What were the elements that came, like, really into fruition that enabled those types of results?

I would say last year was the result of mainly two things. The first one was the compounding effects of a sales transformation that has been going on for more than two years now, and all the things falling in place, moving up the enterprise stack, making good progress with large customers. the last and and and the larger the customer cohort for us that means uh above one million or five million dollars of acv that the more acceleration we saw productivity on the sales force pool of funds deal so the sales transformation really compounding and on on on all the vectors that have been touched i think that that was one of their the big contributors and i think the second part is that you know it becomes clearer and clearer that we are not selling just a product but we we put a platform in place that helps our customers solve significant and challenging problems and the invest the the innovation around that offering that platform is really what what differentiates us from from many in in in the market and i i think this could not have been clearer uh last year about all the the events around ai and the products and tools we are putting in place in order to to support that transformation that all came together nicely so knock on wood um i think those are the main two reasons why why last year was such a special year

got it so as as we head into uh 2026 are you going to give us a another uh prognostation of of an inflection point into 2026, or maybe a fairer question for you. If it was sales transformation and really stretching the benefits of the platform in 2025, any additional elements that we should be looking on and turning on in the story in 2026?

I think we'll continue to make progress around those fronts. And then on top of that, I think we all currently see that AI is replatforming the Internet in a way how code is written, how people interact. And I think this will be one of the significant tailwinds for Cloudflare in this year and moving forward, to be very honest.

I want to start and dig in on sort of the product side of the equation, which is always a little bit dicey with the CFO. but we'll give it a shot. You're pretty good with the prize. So we came out with a framework on team software at Morgan Stanley. We called it our best athletes framework. And the elements of the best athletes framework was in a time of rapid change within the overall environment of rapid innovation, we're looking to companies that have speed, meaning fast innovation. They have strength, meaning good kind of market positioning. They have good execution capability. That's the skill. And then flexibility, right? The ability to change the business models to match that scenario. And I think one of the most impressive things that we've seen over the past year is that product innovation really meeting the period, right? And you can see that across multiple parts of the platform. You guys often talk to, like, Acts 1, 2, 3, and 4. What's so interesting, though, is that it's not like Act I and II goes away when Act III and IV comes online, but you build upon it. So what is it that enables you guys to be that nimble and to be that sort of fast ahead of the right trend? So if we take Act 3, the workers and the developer platform, quickly workers became the developer platform of choice underneath a lot of these cogeneration tools, underneath a lot of those platforms. What enables you guys to get into that positioning to be in the right place at the right time when these innovations are taking place?

You know, I always go back and say to investors or start to work on us and say, if you really want to understand what we are and what we do, you have to go back and understand the architecture of the network. This is really the highest competitive amount we have. The network today, after 15 years, is now in 330 cities, in many cities like San Francisco in multiple locations, in 125 countries, and interconnected today with about north of 13,000 networks. So that is the infrastructure today, off-the-shelf hardware. every product and every service we run, whether it's application services in Act 1, SASE and Zero Trust in Act 2, whether it is workers now in Act 3, and whether it will be, you know, what we say, pay-to-scroll products or whatever, how we monetize content in Act 4, it's running on that one network. Every product and every service we offer on every server in every location. And this is a flywheel that is feeding itself. So, you know, you talked about Workus. Workus was originally not a product that was supposed to be sold. It was a tool that helped us to innovate our products fast enough, how we write code and how we can deploy code with the push of the button to region Earth all around the globe. And this flywheel, I think, is, based on this infrastructure, is allowing us to stay ahead of this innovation game. And, you know, now we're in a world of authentic AI flows, where, looking back now, it seems the infrastructure of the network was built exactly for that use case. What do you need in order to build an agent? You need a CPU, you need a GPU, you need an API. The CPU orchestrates, the GPU computes, the API fetches the data. And in Cloudflare, this happens all in one place on one server in a world where now between 20% to 30% of all websites and all traffic is moving through the network. So the insight you have into what market needs and where markets are going is driving that flywheel to an extraordinary extent.

You have a lot of visibility by having that perspective, and a network that was built for flexibility, that one unified network that everything builds up. It's all software that gets built up. There's no bolt-ons, as Matt's told us, over the decades. But on the other side, you've been able to develop an extreme level of developer affinity. On the most recent conference call, you talked about over 4.5 million human developers. One of the things that really struck me, I was looking at a Stack Overflow survey, and they asked developers, like, where do they want to develop their applications? The first three on the list is exactly who you think it would be. It's Amazon, Microsoft, and Google, because they're huge and they're behemoths. Number four is Cloudflare, right? It's become the place, not what your boss is telling you to develop on, but the place where developers want to go to develop. How do you gain that developer affinity? How do you make yourself such an attractive platform for those developers?

I think a couple of things come together. The tools you build, the ease of use of the platform, the languages that you put into the hands of the developers to work with it, the cost of the product, the performance of the product. So they all have been coming together. The interesting part now is we talked about four and a half million human developers on the platform. If you look at the agents on top of that, it's staggering. And if you look at downloads of Worker KV over the last two months, you know, it's going almost vertical. A big part of that is literally agentic AI agents downloading, programming themselves, deleting when they are done. And so this comes together now in a really unique way.

This is a really interesting point. I was talking with Senjit Singh, who covers a lot of infrastructure names for us, and we're talking after the Databricks presentation. And one of the themes we're starting to hear more and more is the tooling doesn't necessarily have to be built out just for or can't just be built out for humans. Now you have to build a set of tooling, a set of infrastructure that's very amenable to the agents. So does that change the way that you approach the workers' opportunity? Does that change sort of what you need to build out in there to make yourself the sort of platform of choice, not just for the human developers, but for the agents?

Very much. I mean, you know, we started really early, over a year ago now, a year and a half ago, to put a speedboat in place. This is what we called it, around the developer activity. Their person in charge, Ali, was actually the woman who developed Verka as a product in order to get our arms around what go-to-market do we need for enterprises, what features do we need, how do we get to the large deals, how do we keep the human developer count growing, and what do we have to do from an offering, from a feature and a performance perspective to address what you said. How do we make ourselves discoverable for agents that are doing development work, writing code on our network?

And you guys recently acquired Replicate and Astro to strengthen that developer platform. Can you talk to us about what these acquisitions brought into the equation and how it fits into that broader vision of workers that we're talking about?

I mean, you know, how do we cater to their developer ecosystem in terms of the offerings we have, the products we have? How do we get more dynamics in our go-to marketplace? And how do we build this marketplace for developer tools and products that, you know, that you are drawn to work with us in our platform? I think both acquisitions squarely fit into that motion, and you'll see us do more activity in this direction moving forward.

And can we just touch briefly on sort of the defensibility? There's AI labs that are focused a lot on the – I mean, it starts with code development and agenda code development, but they're broadening their scope, right? Is there a dividing line that you feel comfortable? where there's a defensibility of sort of the workers' platform versus where this expanding scope of what we're seeing in the AI labs and their focus on that whole development lifecycle is going to hit?

I think it's a combination of many things. I think it starts with us being that control plane where agents have to move through. Whether it's a human interaction or an agent of interaction, it needs to be secured, it needs to be transported. And it moved to our network today with, you know, 13,000 interconnects and 20 to 30% of the web traffic behind us. So there's that. I think the architecture of the networks and how workers architect it on top of that in terms of no containers but isolates. You know, we talked about how agent workflows are architected. You need a CPU to orchestrate. You need a GPU to compute. you need an api to get data in in in our world you know the you have to wait for the traffic for the api to deliver data or not in in in our architecture of isolates this wait period can be important can be really expensive when you work on container platforms in our case the spin up and more importantly the spin down time when you're waiting for the data to come back is is really fast goes to zero with that the offering of of how we are able to to to price you know not for time not for capacity but for a task completed makes us really unique so i think the combination of the ecosystem the tools the architecture of the network and then the architecture of workers itself running on that network is a pretty substantial competitive amount that is hard to

overcome. It comes back to that strength that being in front of 20% of the internet, but really the infrastructure of being in front of that 20% of the internet with a secure platform should be a durable advantage. And that kind of sort of segues into act four of what you guys have been talking about in terms of the agentic web. Can you just maybe as an overview describe to us, what's the opportunity for Cloudflare in there? We all understand a lot of agents are being built on workers. That's part of Act 3. There are more and more agentive activity taking place. But what is the new monetizable opportunity for Cloudflare

in this Act 4? Well, in the near term it is actually that opportunity in getting ready to monetize a world of agentic AI workflows and content is actually driving business into the older acts first. So, you know, today about, I would say, 80% of the 50 top AI native companies is behind our network. And they need to secure, they need to mitigate traffic and requests. So that's thriving business. We see tailwinds from verticals that we weren't traditionally strong in, especially media content creative publishers that now want their content to behind a network where they can control how it's monetized so in in the first step it's there you know the opportunity of act four is literally driving uh business in act one and act two in the first place and we've seen this as part of the acceleration in the in the last year and we'll see more in um in in during 26. so act four then moves us into the world how is content monetized how do we sit between frontier models and content and give both sides the the opportunity to monetize there will be agentic ai workflows that do business on each other's behalf or on your behalf with microtransactions So how do we build the trails in order to facilitate that? That's why you hear us talking and partnering with pretty much every large payment provider, whether it's the credit card companies, whether it's PayPal or Coinbase. We talked about putting our own stablecoin in place in order to facilitate that. How that is going to play out in detail. I think there are a lot of questions still that need to be answered. But we are making really good progress building an ecosystem and building the rails that need to enable that.

Got it. Excellent. You guys talked about AI traffic already ramping up on the platform. I think you made a comment about in January of 2026, weekly AI agent requests more than doubled on your network. And a lot of investors, Claudebot was a good case in point, had made the connection of, like, this is a lot of API calls, right, which is kind of act one in application services. This is a lot of network traffic to be secured. Where are we in terms of sort of, yes, there's a doubling, but from a very small base, where are we on that evolution of when this agentic traffic gets to a volume that it's material for your business, that it becomes a more material driver of acts one and acts two?

So, I think we're in the early innings. Today, I would say slightly north of 50% of the traffic in our network is API-driven, but most of it is human-initiated. As you said, in the first six weeks of this year, we saw agent-driven traffic really go through the roof, double literally over a very short period of time. You know, from how we handle a request, it really doesn't matter whether it's a human request or a machine-driven request. And so how that will – and most of our requests are priced in terms of – Phil always makes a good comparison – in T-shirt sizes. So you as a customer, you buy an M-T-shirt that gives you 40 million requests, and as you approach the cap, you buy the next size T-shirts. So there is a lag and a step up before you see revenue materialize. So it's a lagging, I think, indicator. We're coming from a small base, but the slope of the increase is staggering. I think if you asked me six months ago if that is even close to possible, I would have said no, but it's staggering.

And if you think about the unit economics, if you will, of the AI and autonomous agent traffic, it sounds like you guys are neutral, right? It doesn't matter to you whether it's a human-initiated or an agent-initiated. Do you still feel comfortable with the economics of that core application services business from that rising traffic spike?

Yes, we are. I mean, this is the beauty of the efficiency and the flexibility of that network that it's able to digest incredible amounts of fluctuations and upsides. I mean, go back to COVID, where traffic literally increased 60 percent quarter over quarter, and we didn't even see a flinch in our network, neither from a performance nor from a capacity, even more important, nor from a cross-margin perspective. So it's able to digest incredible amounts of swings of traffic. We are highly comfortable that that is what the network is built for and that we can digest that. You have good visibility in what the trends are. We guided CapEx for this year in the widest range we ever guided it, 12% to 15%. And the highest on the upside, 15%. Not so much because we need more capacity. It's much more a reflection of some of the supply chain pressures we see from a pricing perspective, especially on the memory side. But again, it gives you an idea how flexible the network is. One of the things that always has driven this efficiency in the business model based on the architecture is that we are in this unique position to invest behind demand. demand. So you're never in a position where you say, I have to invest $100 million in GPU capacity, hoping that traffic will come. If for whatever reason, traffic would explode across the locations in San Francisco, we'd probably be able to double the capacity across all San Francisco locations, generate revenue in these sites before we even pay for the hardware that we put in place. And that means that you're always in this comfortable position that you can follow demand and you're never really investing ahead of it. That is also why over the last year and a half, we spent so much time on really understanding what inference workloads are, how they vary. Inference task is not like inference task. and how we optimize our hardware stack for that so we can abstract the need for the customer to decide, I need this capacity, I need this GPU, by just saying we'll compute it for you, we'll find the cheapest, the most performant way to approach this. And again, it comes back to the architecture of the network. This is why we are so CapEx efficient, and that's why the absorbability of the network to deal with gigantic fluctuations in traffic is so significant.

I want to switch gears a little bit and talk about network services and in particular Zero Trust and Cloudflare One. This has been, I think, a building momentum within the Cloudflare story. This is, I think, the part that most aligns with sort of the improving sales productivity. But it's also one of the more competitive environments. There's a lot of areas that Cloudflare is almost sort of a class of one when we're talking about these application services or workers. But we've talked to several SASE vendors at this conference. How does Cloudflare look to differentiate in that environment where there's more of a competitive environment, there's more of a sort of direct vision from the end user, the CISO, of what they're looking for in these solutions?

Yeah, I mean, we are really happy with our SaaS development for a couple of reasons. First of all, from a feature performance perspective, we launch products early, and then they feature out over time, and we call this moving up and to the right in whatever garden or forest or quadrant, and we are right there now. What we needed most now over this year in terms of continuing momentum is building out our go-to-market path, partner and channel. And that was needed because, you know, from a go-to-market perspective, there was not really a lot of value to be delivered for the partners on application services. The products are really easy to install. It's really even for, you know, Morgan Stanley would be behind our network within hours, not within days. But for Zero Trust, we needed to build out a partner channel strategy. which Mark Anderson successfully did. Our partner-driven revenue share in the fourth quarter was the highest ever. It was 29%, which is really a good proxy for our progress on zero trust because this is what we mainly use the channel for. If you look at our competitors, their channel shares in terms of revenue are significantly higher, so we have lots of opportunity to continue to grow. and our win rates are are really really strong so you know when we are in in consideration the likelihood that we win is is really high so we are happy with the momentum and of course sassy for us is not just a product or suite of products it's part of a platform and as part of this platform it it is a significant value distribution we have not been tired to read down the wins we had on even very large and uh customers in traditional verticals even over the last earnings calls so good progress last point um it's probably our strongest margin product because it takes advantage of the infrastructure that has been built with what you would call act one you know the size of the pipes that transport the data is sized besides on act one for data are moving out to the eyeballs, and now you fill it with traffic coming back with Sassy products, so it's a high-margin product, so we have ample room to compete.

So it seems to me that having Mark Anderson engage a partner channel, as well as the evolution of the products, just gives you more credibility as a security vendor, right? It's fundamentally a little bit of a different buyer from Act 1 and definitely from Act 3. What are the customer conversations today like? Are you still convincing them that Cloudflare is a real security vendor, or are we past that point and it's really about how broad can we go with the Cloudflare 1 platform?

I think we have moved far beyond that already. I mean, and you see that, you know, the largest customer cohorts are the fastest growing customer cohorts. If you just go back to the earnings calls, most of the customers that we read down there are coming from traditional verticals, critical infrastructure verticals, the progress we are making with federal customers, not only in the U.S., but also outside with friendly. So I think we are beyond that. I think the biggest change in the go-to-market transformation where Mark Anderson really made a difference is much more in terms of ingraining this enterprise DNA into the sales force and into the cloud fair culture. We've been quite open, transparent that we were always a product and innovation-driven go-to-market engine, and it needed this DNA change. And then last but not least, what I said before, you know, combining that with a highly effective channel strategy that those two parts were missing. I think he did an incredible job, you know, instilling this into their into the company. And I think we are in the customer conversations. We are beyond what you said. You know, we don't we don't have to convince people anymore that we are a serious player.

Right. So Mark's been on board for about two years. engaged sort of a channel strategy, but to your point, created much more of an enterprise sales motion. You put that together with the breadth of the solution portfolio, and I'm assuming it's those two things coming together that enable something like a $42.5 million ACV deal. So maybe we could just touch on what does that look like? What is the customer acquiring from Cloudflare when they're putting that much budget annually behind your company and behind your platform?

well you know first of all i i i interpret that as a vote of confidence right you know you commit a lot of dollars to to a supplier across a very broad range of products a pool of fun steel in this case really means that you can buy every product that is on on on on the offering list you you you negotiate a rate card for every product that we have and then you decide um how you use our we talk of course to each other but you know how you consume um what it does for us it's it makes this product adoption and expansion rather frictionless right because literally once you are on our network with the product every other product and service is it's very much a mouse click away you know this is all it needs so so having this rate card already negotiated the commitment in place it makes a rather frictionless process in terms of expanding and expanding across our product portfolio.

Got it. So pool of funds contracts have pretty quickly ramped up. I think you talked about it as about 20% of your ACV. From an investor standpoint, how should we think about the impacts of this new contracting methodology in terms of variability in the business model and and potential more volatility in kind of the revenue run rate?

Well, first of all, we should be all excited because, you know, it's large deals and it's committed revenue over a period of time. And it generated noise, especially in the beginning, especially when an existing customer migrated from a traditional contract into a pool of funds contract to make this more visible. Well, you know, say you're a customer, you're committed to $10 million a year. Now you sign a pool of funds contract for $40 million over or $45 million over two years. So, of course, you will get better unit pricing because you have a significantly larger commitment. So in the first two quarters, you know, while you consume as much as you did before, maybe even slightly more because the unit prices are lower, you have to step. So you have a headwind to DNR for a quarter or two. It makes forecasting a little bit more tricky, but then you catch up and you move out of this. You have this Nike swoosh from a revenue perspective, and this is what we have seen last year. Of course, this is only a headwind that you see when you migrate an existing customer into a pool of funds contract for a new customer. That is just DNR expansion moving forward. So it makes my life a little bit, or my team's life, a little bit more difficult from an accounting, billing, especially forecasting perspective. But since it's committed dollars over the length of the contract, you know, this is all good news. And, of course, both eyes have to get used to how you handle that. And I think we are through this now. The forecasting has become much more stable. If you look across all the pool of funds, steels that are in play, we are slightly ahead of linearity. So, you know, there's good management also from a customer consumption perspective. So, that's become a rather beneficial instrument to especially grow our large customer cohorts even faster. And, you know, one of the reasons why DNR improved so significantly over last year.

Got it. Got it. I want to turn to gross margins. As a CFO, I have to ask you some margin questions as well. So you guys are still within that 75% to 77% long-term target model. But we have seen some impacts on gross margins. And it seems like a lot of this is mixed shifts from free to paid customers. Can you dig into why is that an impact and sort of what's the duration of that impact? And is there anything else that we should be aware of from the gross margin perspective within that time frame?

Yeah, maybe just for background. So, you know, we have hundreds of thousands of paying customers, but we also have millions of free customers. And that is an incredible value to the business model because, you know, if you even go back to OS1 and read the reasons why we have free customers, It gives us threat intelligence diversity. It allows us to consolidate huge amounts of data that makes us an attractive partner. If we go to an ISP and say, co-locate us, and here are the benefits. So we have many free customers. The cost to serve these customers, we put into sales and marketing. We always have. That is how we handle it. Now, if such a free customer now decides he or she wants to become a paying customer, the cost to serve the customer moves from sales and marketing into cost of revenue, so it has an impact on margin for the overall P&L. It doesn't really matter, but you have a shift in margin. What we've seen over the last two quarters is a significant shift of free customers wanting to give us money. So it's actually really, really positive. And if you look at our paying customer account, it has exploded. Some of them are really micro-customers. They might be just a dollar above the threshold, but it means that all the costs move from sales to marketing into cross-margin. Accounting-wise, it pressures cross-margin. Overall, the unit economics stay whole. In addition to that, so I don't want to be flimsy about this, our developer business, as we talked about it before, is really taking off. The growth rates are significant. Developer product is still below corporate average. We've made significant improvements over the last year in terms of utilization, but it's under the average cost margin, so there's slight pressure. However, the overall unit economics for this product are really, really good because it's significant and cheaper to sell that product. So if you look across all four acts, they have very different cross-margin profiles, but very consistent unit economic margins. So one of the lessons we learn now, and as we go into our investor day later this year, we'll talk more about both sides of that equation, because if we just signal and continue to signal cross-margin performance, that would lead to a misallocation of capital. We don't want that, so we have to adjust. I'm going to try to squeeze in one question from the audience.

It sounds like...

Yeah, yeah. A couple of ways. And, you know, I think if you go back to our blog post from last week, you know, we wrote a big piece of software in a week with one developer and there, whatever, less than $1,000 of tokens. I think that gives you an idea. I think it monetizes in a couple of ways. First of all, it just increases traffic through the network in a very gigantic way. The traffic continues to be secured, needs to be moved, needs to be controlled. So we become this control plane for that traffic that is going to explode. These agents will do business with each other. We will find ways as part of Act 4 to monetize that, so clearly. And then, you know, the biggest advantage, I think, of the developer growth that I've seen so far is it gives us insight into how inference tasks are going to develop moving forward. So the variety of things that are happening is so impressive, but it gives us this unique opportunity, forget dollars for a moment, to optimize our hardware stack better than anybody else in the direction of how we run workloads on our network. That's the reason why our capex is so low. So I think there are more than one benefit to what you described. I think the flywheel around this is going to be a significant tailwind to our business. It's one of the reasons why we think Agenda AI is such a multiplier to our infrastructure.

That's a great note to end on. Thomas, thank you so much for joining us. Thank you, Keith. Always. Thank you.