Cloudflare, Inc. (NET) Earnings Call Transcript & Summary

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. I'm 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, right? There was activities that took place that we weren't really on the bingo card, if you will, of investors. You saw on your largest ever contract, $130 million contract, 7 years in duration. The highest ACV deal you've ever signed at $42.5 million. RPO growing 40% during the year, 30% revenue growth accelerated in each of the last 3 quarters. The net revenue retention rate improved from 111% in Q1 to 120% in Q4. A number of paying customers, up 40% year-on-year, and the list goes on, right? 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're correct, right? There was an inflection that took place. What were the elements that came like really to fruition that enables 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 2 years now and all the things falling in place, moving up the enterprise stack, making good progress with large customers, the last -- and the larger the customer cohort for us, that means above $1 million or $5 million of ACV, the more acceleration we saw productivity on the sales force, pool of funds. So the sales transformation really compounding on all the vectors that have been touched. I think that there was one of the big contributors. And I think the second part is that it becomes clear and clear that we are not selling just a product, but we put a platform in place that helps our customers solve significant and challenging problems. And the innovation around that offering, that platform is really what differentiates us from many in the market. And I think this could not have been clearer last year, but all the events around AI and the products and tools we are putting in place in order to support that transformation. That came together nicely. So knock on wood. I think those are the main two reasons why last year was such a special year.

Got it. So as we head into 2026, are you going to give us another prognostication of an inflection point to 2026? Or maybe a fair question for you. 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 make -- we'll continue to make progress around those fronts. And then on top of that, I think we all currently see that AI is re-platforming the Internet in a way how code is written, how people interact. And I think this will be one of their significant tailwinds for Cloudflare in this year and moving forward, we being honest again.

Yes. 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. I'll be pretty good with the front. So we came out with a framework on Teams software at Morgan Stanley. We called it our best athletes framework right? And the elements of that best athletes framework was in a time of rapid change within the overall environment of rapid innovation. We're looking to companies that are -- have speed, meaning fast innovation. They have strength, meaning good kind of market positioning. They have good execution capabilities. 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 Act 1, 2, 3 and 4. What's so interesting though is that it's -- it's not like Act 1 and 2 goes away when at 3 and 4 comes online, so you build upon it. So -- what is it like that enables you guys to be that nimble and to be that sort of fast ahead of the trend? So if we take Act 3, the workers and the developer platform, Quickly, workers came the developer platform of choice underneath a lot of these cogeneration tools under leaving 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.

I always go back and say when investors or -- start to work on us and 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 25 countries and interconnected today with about north of 13,000 networks. So that is the infrastructure today of 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 works now in Act 3, and whether it will be what we say pay to scroll products or whatever how we monetize content in Act 4 is 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 talked about Workers. Workers 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 now we're in a world of agentic 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 need API. The CPU orchestrates. The GPU compute. The API fetches the data. And in cloud and this happens all in one place, on one server in a world where now between 20% to 30% of our 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.

Got it. It has a lot of visibility by having some of that perspective. And a network that was built for flexibility, right, that one unified network that everything built up, and it's all software that gets built up. There's no bolt-ons as 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 ask developers like where do they want to develop their applications -- the first 3 on the list is exactly who think would it's Amazon, Microsoft and Google because they're huge and the behemoth. N4 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 -- like 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've built, the ease of use of the platform and the languages that you put into the answer 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 4.5 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 work TV over the last 2 months, it's going almost vertical. A big part of that is literally agent AI agents downloading, programming themselves still leading when they are done. And -- so this comes together now in a really unique way.

Yes. And this is a really interesting point. And I was talking with the [ Sachit Singh ], who covers a lot of infrastructure names for us. And we're talking after the data bricks 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 meetable to the agents. So does that change the way that you approach the Workers opportunity? Does that change sort of how -- sort of what you need to build out in there to make yourself be sort of platform of choice, not just for the human developers, but for the agents?

Very much. I mean we started really early, over a year ago or 1.5 years ago to put this boat in place. This is what we called it, around the developer activity. Their person in charge, Aly, actually the woman who developed Workers 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 how do we have -- 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.

Yes. Got it. 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 how do we cater to the developer ecosystem in terms of the offerings we have, the products we have, how do we get more dynamics in our go-to-market place and how do we build this marketplace for developer tools and products that you're trying 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.

Got it. 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 co-development, agentic co-development, but they're broadening their scope, right? Is there a dividing line that you feel comfortable there's a defensibility of sort of the Workers platform versus where this expanding scope of what we're seeing the AI labs and their focus on that whole development life cycle 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, interaction needs to be secured. It needs to be transported. And the move to our network today with 13,000 interconnects and 20% to 30% of the traffic behind us. So there's that. I think the architecture of the networks and how Workers architected on top of that in terms of no containers, but isolates. We talked about how agents workflows are architected. You need a CPU to orchestrate. You need a GPU to compute. You need an API to get data. In our world, you have to wait for the traffic for the API to deliver data or not. In our architecture of isolate, 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 downtime when you're waiting for the data to come back is really fast. It goes to zero. With that, the offering of how we are able to price, 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.

So it comes back to that strength that being in front of 20% of the Internet, but really the infrastructure of being 20% of the Internet with a secure platform should be a durable advantage. And that kind of sort of segues into Act 4 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 a there more and more attentive activity taking place. But what is the -- like the new monetizable opportunity for Cloudflare in this Act 4?

In the near term, it's actually that opportunity and getting ready to monetize a world of agentic AI workflows and content is actually driving business into the older Act first. So today, about I would say, 80% of the 50 top AI native companies is behind our network and that they need to secure. They need to mitigate traffic and request. So that's driving business. We see tailwinds from verticals that we weren't traditionally strong. And especially media content creators, publishers that now want their content to behind a network where they can control how it's monetized. So in the first time, the opportunity of Act 4 is literally driving business in Act 1 and 2 in the first place. And we've seen this as part of the acceleration in the last year. And we'll see more in -- during '26. So it then moves us into the world, how is content monetized, how do we sit between frontier models and content and give both sides the opportunity to monetize. There will be agentic AI workflows that do business on each other's behalf or on new behalf with micro transactions. So how do we build rails 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 invoice. We talked about putting our own stable coin 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 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 1 in application services. There's 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 Act 1 and Act 2?

So I think we're in the early innings. Today, I would say, slightly north of 50% of the traffic in our in our network is API-driven, but most of it is human initiated. As you said, in the first 6 weeks of this year, we saw agent-driven traffic really go through the roof double literally over a very short period of time. From how we handle a request, it really doesn't matter whether it's a human request or machine-driven request. And so how that will -- and most of our requests are priced in terms of [indiscernible] always makes a good comparison in T-shirt sizes. So you're the customer you buy an M T-shirt that gives you [ 40 million ] you request. And as you approach the cap, you buy the next size defer. So there is a lag and a step-up before you see revenue materialize. So it's a lagging, I think, indicator. We are coming from a smaller base, but the slope of the increase is staggering. I think if you ask me 6 months ago, if that is even close to possible, I would have said no, but staggering.

Got it. And the -- if we 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, 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 -- and the efficiency and the flexibility of that network that it's able to digest incredible amounts of fluctuations and upsides. And I mean, go back to COVID where traffic literally increased 60% quarter-over-quarter and we didn't even see a flinch in our network, neither from a performance nor from a capacity, even one part not from a gross 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. We have good visibility on what the trends are. We guided for this year in the widest range we ever guided at 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 price 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. So you're never in a position we say, I have to invest $100 million in GPU capacity hoping that traffic will come. If for whatever reason, traffic would explode in 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 1.5 years, we spent so much time on really understanding what inference virtuals are, how they vary. Inference task is not like inference tasks and how we optimize our hardware stack for that. So we can abstract the need for the customer to decide I need this capacity you need this GPU, but I think we'll compute it for you 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 and traffic is so significant.

Got it. 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. I think the part that most aligns with sort of the improving sales productivity. But just also one of the more competitive environment. So there's a lot of areas that Cafe is almost sort of a class of one, right, when we're talking about these application services or Workers. But we've talked to several SASE vendors, right, at this conference. How does Cloudflare look to differentiate in that environment where there's like more of a competitive environment? There's more of a sort of direct vision from the end user of the CISO of what they're looking for in these solutions.

Yes. I mean we are really happy with our asset development for a couple of reasons. First of all, from a feature performance perspective, we launched products early and then the feature out over time, and we call this moving up into the right and whatever Gartner-Forrester quadrant, and we are right there now. what we needed most now in over the last year in terms of continuing momentum is building out our go-to-market partner and channel. And that was needed because 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 Morgan Stanley would be behind our network within hours, not within days. So -- 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. This is what we mainly use the Channel 4. If you look at our competitors, their general shares in terms of revenue are significantly higher. So we have lots of opportunity to continue to grow. And our win rates are really, really strong. So when we are in consideration the likelihood that we win is really high. So we are happy with the momentum. And of course, SASE for us is not just a product or a suite of products. It's part of a platform. And as part of this platform, it is a significant value distribution. We have not been tired to read down the wins we had on even very large customers in traditional verticals even over the last earnings calls. So good progress. Last point. It's probably our strongest margin product because it takes advantage of the infrastructure that has been built with what you would call Act 1, the size of the pipes that transport the data is sized on Act 1 for data moving out to the eyeballs. Now you fill it with traffic coming back with SASE product. So it's a high-margin product. So we have ample of room to compete.

Excellent. So it seems to me that the -- 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 step 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 One platform?

I think we have moved far beyond that already. I mean -- and you see that the largest customer cohorts of the fastest customer -- 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 on 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 the difference is much more in terms of ingraining this enterprise DNA into the sales force and into the cloud fare culture. We've been quite open, transparent that we were always a product and innovation driven go-to-market engine, and it needed the DNA change. And then last but not least, what I said before, combining that with a highly effective channel strategy, those 2 parts we're missing. I think he did an incredible job instilling this into the company. And I think we are in the customer conversations we are beyond what you said. We don't have to convince people anymore that you are a serious player.

Right. So Mark's been on board for about 2 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 enables something like a $42.5 million ACV deal. So maybe we could just touch on what does that look like? Like what is the customer acquiring from Cloudflare when they're putting that much budget annually behind your company, behind your platform?

Well, first of all, I interpret that it's a vote of confidence, right? You commit a lot of dollars to a supplier across a very broad range of products. A pool of fund steel in this case really means that you can buy every product that is on the offering list. You negotiate a rate card for every product that we have. And then you decide how you use -- we talk, of course, to each other, but how you consume. What it does for us is it makes this product adoption expansion rather frictionless, right? Because literally, once you are in our network with the product, every other product and services, is very much a mouse click away. This is all in me. 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 is 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 potential more volatility in kind of the revenue run rate?

Well, first of all, we should be all excited because it's large deals and 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. Say you're a customer, you're committed to $10 million a year. Now you signed a pool of funds contract for $40 million over -- or $45 million over 2 years. So of course, you will get better unit pricing because you have a significantly larger commitment. So in the first 2 quarters, 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 2 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 at my teams like a little bit more difficult from a accounting building, especially forecasting perspective. But since it's committed dollars over the length of the contract, this is all good news. And of course, both sides 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 at across all the pool of funds deals that are in play, we are slightly ahead of linearity. So 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 one of the reasons why DNR improved so significantly over last year.

Got it. Got it. I want to turn to gross margins. As the CFO, I have to ask you some margin is as well. So you guys are still within that 75% to 77% long-term target model. But we have seen some impacts on the gross margins. And it seems like a lot of this is you shift from free to paid customers. you dig into why that an impact and sort of how -- like 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?

Yes. Maybe just for background. So 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 if you even go back to [indiscernible] the reasons why we have free customers 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 ice say, co-locate us and hear the benefits. So we have many free customers. The cost to serve these customers, we put into sales and marketing there. 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 move from sales and marketing into cost of revenue. so have 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 2 quarters is a significant shift of free customers wanting to give us money. So it's actually really, really positive. And we -- if you look at our paying customer count, it has exploded. Some of them are really micro customers. They might be just $1 above the threshold, but it means that all the costs move from sales to marketing into gross margins on that is accounting-wise it pressures gross margin overall, the unit economics they hold. 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 [indiscernible] is still below corporate average. We've made significant improvements over the last year in terms of utilization, but it's under the average gross margin. So there's slight pressure. However, the overall unit economics for this product are really, really good because significantly cheaper to sell that product. So if you look at across all 4 Acts, they have very different gross margin profiles, but very consistent unit economic margins. So we -- one of the lessons we learned 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 gross 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.

On the AI front -- on the AI front, it sounds like everyone's focused on what the AI is going to [indiscernible].

A couple of ways. And I think if you go back to our plus from -- last week, we wrote a big piece of software in a week with one developer 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 4 to monetize that. So clearly -- and then 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 for get dollars for a moment to optimize our hardware stack, better than anybody else in that direction of how we run workloads on our network. That's the reason why our CapEx is so low. So I think there are 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 agent AI is such a multiplier to our infrastructure.

Outstanding. It's a great note to end on. Thomas, thank you so much for joining here.

Thank you, Keith. Always. Thank you.