Please welcome Vice President of Strategic Finance, Treasury, and Investor Relations at Cloudflare, Phil Winslow.
Hello, everyone. That was a dramatic pause for the Safe Harbor statement.
Yes, sir.
All right. Welcome, everyone. Thank you for joining us today. I know it's been a volatile time in the markets, so we really do appreciate you taking some time out of the day to come understand the Cloudflare story a little bit better. I guess before we get started, I really want to set the stage for the rest of the afternoon. There's really an underlying theme that sort of connects all the presentations you're going to hear today, and that word is accelerating. We want you to walk away today with the full understanding of how we are accelerating innovation, accelerating our go-to-market transformation, and positioning ourselves to re-accelerate revenue with a keen focus still on maintaining the unit economics of our business. We are going to start with a fireside chat with our Co-founder and CEO, Matthew Prince.
You'll then hear from CJ Desai and Rita Kozlov on our product and AI strategies. Mark Anderson will then take the stage for an update on our go-to-market transformation. After that, you'll hear from Thomas Seifert, who will delve into the financials, followed by open Q&A with the audience with Matthew, Michelle Zatlyn, Thomas, Mark, and CJ. With that, please help me welcome Matthew Prince to the stage for a fireside chat. All right. Thank you, Matthew. All right. We're going to mix it up a little bit this year. Instead of a traditional presentation that you see from Matthew, we're going to do a fireside chat. When we were prepping for this, Matthew said, "I don't want to see the questions beforehand, but I never get the hard questions on the earnings call.
I want you to be, as Michelle put it, evil Phil. And so.
I wanted to get him devil horns.
Devil horns. Yeah. We'll call this ultra ego Ethel. We're going to do a little mix.
It's like the old banking.
The old banking, old version.
Oh, yeah.
My interdefuji. No, just kidding.
Not that either.
Exactly. We are going to do a mix. Somebody said some compliment sandwiches. We are going to mix it up, some things I think people would want just your views on, but also some of those tough questions that I get. Maybe we will start with one of those compliment sandwiches. Looking back at last year, what are you most proud of when you think about some of the accomplishments from Cloudflare in 2024 and how those position us for 2025? On the flip side, though, when you look back on 2024, what could Cloudflare have done better? Where did we not hit your expectations that you had at the beginning of the year? What are we going to do about that in 2025?
I think the bench that we've built in 2024, adding Mark Anderson to run go-to-market, adding Stephanie Cohen as our Chief Strategy Officer, adding CJ Desai, I mean, that's just, it's been so transformational to our business, those people coming in. Then the people that they've brought in below them, that's what I'm most proud of in 2024. I think 2024 actually was what I would say we could have done better. 2024 was sort of a year where actually the kind of innovation engine at Cloudflare, instead of focusing on new products, instead of focusing on getting more things out, we actually focused on kind of cleaning up some of the mess that we had made at the end of 2023. We had an outage at the end of 2023. We had some stability issues.
I think that that was a place where it actually felt like we slowed down a little bit in terms of innovation, which is something that's very unusual for us. It's fun. Next week is our security week. We have our developer week, which is coming up. What I'm excited for in 2025 is not only having, again, those new executives that have come in and really, I think, helped build kind of the bench and helped build the leadership team at Cloudflare, but I'm excited actually getting the innovation engine back up to its full steam again.
Great. You mentioned the security week. Let's focus on Act I and Act II, but from a security perspective, if you think about Thomas's section last year from Investor Day, we said, hey, two-thirds of our ACV was either application security, zero trust security, or network services, obviously a pretty significant increase over the previous several years. There's been a lot of talk about platformization in this industry. You may have heard that term before, but particularly in security. How do you see Cloudflare's competitive position evolving versus, let's say, those leading point product vendors, but also versus other competitors that use a similar platform branding?
Yeah. You know, I think that we started as a security company. The original idea of Cloudflare was how do you take a firewall and turn it into a service? We had to, in order to do that, build a network. Our goal in the beginning was just to get to sort of performance parity because we were adding a bump in the road and we needed to make sure we did not slow things down. We did that better than we expected. As a result, all of a sudden, we were like, well, it is not just more secure, but it is faster. From the very beginning, it was always and. It was always this and this, this and this. You got all of those benefits together. I know we have talked about sort of Act I, Act II, Act III.
Just like in a play, those are really temporal. It's when we built the various aspects. The real power of Cloudflare, and we say this over and over again, and I think it's actually one of the things that is the most underappreciated about the business. Every single server that is out that makes up our network is able to perform every single function at Cloudflare. As a result, if we get a server into an ISP, then that means that all of a sudden, in that location, we have the full range of Cloudflare's products. The flip side of that is, as a customer uses one individual product at Cloudflare, the cost is all on that first product.
As we add additional incremental products, since we're already processing the bits, the amount of incremental cost to us and the amount of sort of challenge to the customer at the end of the day, it doesn't get any more expensive. It doesn't get any harder to use that. I think that's what a true platform is. It's one of the reasons why I think we have been reticent to do anything in the big M&A space. We really do believe if we innovate and build on our own platform, and even the small acquisitions that we've done, almost every single one of them, the company is actually building on Cloudflare already. The integration, the technical integration was much, much simpler. I think that that's, it's not just security.
Everything that we do, security, performance, reliability, the developer platform, all of that stuff fits together because, again, every single server at Cloudflare can do our entire product suite.
Got it. Now, let's jump all the way to Act III. I think we're going to circle back to Act I and II. I was sort of the original Workers fanboy on Wall Street. This goes back to that original blog post in 2017. I'm like, that's that thing that I've been looking for. You convinced me. You'd say, hey, serverless computing just as a whole across just the whole cloud computing market is still relatively limited. Two questions here. What do you think has been the biggest bottleneck to just broader enterprise adoption of serverless platforms? What's going to change sort of adoption just of serverless in general, but also Workers specifically?
I think if you look at the evolution, I remember when I was in college, if you wanted to run a different application, you had to buy a different, literally physical server. Then someone like VMware comes along, you can take one server and make it into many and get higher utilization from that. The same application that used to run on a server, you could just kind of pick up and move to that VM that was running on individual servers that were there. When the sort of first generation of Hyperscalers came out, what they were doing is they were just taking either that VM or eventually containers. They were saying, instead of you having to own the equipment, we'll own it for you, and you lease it from us, and you could take that same application. There are a lot.
The vast majority of workloads that run in traditional Hyperscalers are these legacy applications that are running in something that would be very similar to what you would have done back in the early 1990s. Instead of you owning the box, it's a box that you rent from someone else. I think that is a business, but it's not how people, if they were able to start from scratch, how they would develop things. A lot of people don't want to start from scratch. I mean, there's a lot of legacy applications that are out there that are running on AWS or Google or various things. What's been powerful about the approach that we took is that we did start with a blank piece of paper. We said, if you were a developer, how would you want to develop applications going forward?
We did not build, honestly, we had no idea that this was going to be something that would be useful or interesting outside the walls of Cloudflare. Kent and John, Kent and John, our engineering team, and John is our CTO, we sketched out the idea for Workers because we were hitting bottlenecks for our own team being able to provide it. We had a little Mexican restaurant across from our office in San Francisco. I wish I had saved it, but on a paper placemat stained with salsa, we sketched out here is how Workers was going to work. We built it for ourselves. We built it for our own team.
We built it with this idea of if you could build something that just would scale infinitely, that would allow you to very carefully sandbox new code so that it did not impact other things that were running, what would that look like? The power of that is a lot of why Cloudflare is able to develop as fast as we can. It is why we are able to release as many products as we can. It is why we are able to catch, even in new markets we enter, the incumbent providers incredibly quickly and oftentimes overtake them in terms of performance and reliability and certainly on the cost structure of what we are doing. As we told this story to customers, because customers were like, how do you guys develop so fast? We tell them the story of workers, and they say, we want that too.
What you're seeing is when you get to start with a blank piece of paper, workers are the right way to do it. It's a totally different model than the traditional, let's go rent kind of a sliver of a server somewhere. I think what's been interesting is whether it was sort of crypto companies five years ago and sort of resurging again now, whether it's AI companies today, the killer apps for workers are the new apps that are out there. The amazing thing about that is that if you fast forward, that becomes the underlying platform for a lot of the things that are going to be disruptive innovation that are coming out.
The challenge is that we can't just go to United Airlines and say, go pick up your Oracle database and instead of running it on your equipment, run it on our equipment. I think that it is slower kind of as it gets started, but it is much more durable over the long term. You don't get that kind of these giant lift and shift deals that you have with other things because people have to either rewrite their applications or rewrite parts of their applications. In, again, the cases when there's new innovation, they have to start with a blank piece of paper. When they do, and it's why if you look at the AI universe today, the vast majority of those companies are using Cloudflare, one of the most common cloud platforms that's out there.
I think that's because if you're building for the future, if you're building a new thing for the future, of course, you're going to start with a better developer platform.
In fact, the number is 1,900 AI companies that we can target and see right now are on Cloudflare.
Yeah.
Let's talk about competitive moats now in cloud computing. When I think back to sort of the centralized hyperscale cloud, AWS established a competitive moat when there was sort of a critical mass of data stored in S3 and whereupon just data gravity took over. Satya Nadella, just in an interview last month, said, hey, the speed of light is the speed of light. We've got to serve the whole world on having an inference fleet everywhere in the world.
It felt like Satya was interviewing for my job.
I know. I was going to say, I was like, I've heard that before, the speed of light from somewhere. It's like there's no more swallow of the speed of light. Let's say the shift to inference continues that we've been seeing and inference and agentic workflows need to become more distributed. What's Cloudflare's competitive moat for AI inference and agents?
We are the most connected network in the world. That connectivity is what will drive the real advantages over time. That connectivity, yes, gives you better performance, but it gives you better cost structure. It gives you better ability to do actual compliance as different governments around the world put different rules in place. That connectivity is very, very, very difficult to replicate. Again, we have done a bunch of things in order to get into every far corner of the universe. It is not clear that you can take, again, a traditional Hyperscaler model and actually distribute it that way.
Because, again, while what Satya said makes sense in terms of it isn't the way that you build a traditional VM or container-based system, what's powerful about Cloudflare is that we can, because, again, every server runs every single thing, including every one of our customers' pieces of code, then that allows us or is capable of running any of those. That allows us to actually move code to wherever it makes the most sense in our system. If you're building to containers, if you're building to VMs, you can't do that. It's too heavyweight to be moving things around that way. You actually, I think Satya is right that that's what you need. You need that hyperconnectivity. I don't think that you're going to be able to see the Hyperscalers actually deliver that because architecturally, they'd have to rebuild fundamentally what they're doing.
Whereas we built an app, we built the network, and then we built a developer platform that took advantage of the benefits of that network. We did not build it to say, oh, how can we go win a bunch of business running Oracle databases? We said, how can we build stuff that makes sense for us? That turns out, as more and more people are realizing, to be the type of platform that you need to build what the future is demanding.
Yeah. I use a line often read by the promoter of the Grateful Dead. To paraphrase him, it's like, other people might do what we do, but nobody does what we do the way we do it.
Yes. Yeah.
OK, next question.
Is this an evil fill question?
This is evil fill. OK, so back to evil fill now. Microsoft reported to have $13 billion in annualized AI revenue, and that was primarily inference and post-training workloads, as well as their Copilot revenue. They also said in fiscal 2025 that they were on track to spend $80 billion to build out AI-enabled data centers. Let's say Cloudflare gets to $5 billion and beyond in revenue. We've also experienced outsized growth in the developer platform because of AI inference and agents. Is network CapEx still in the low double digits as a % of revenue? Or as one investor put it to Thomas and me, is that all Cloudflare is going to need when Microsoft and AWS are literally buying hundreds of thousands of GPUs each year?
Again, it comes back to fundamentally what you're selling. If you're Microsoft, what you're doing is you are saying, I'm going to rent you a sliver of a computer. Then it's up to you, dear customer, to figure out how to get the most utilization out of that. I spend a lot of time talking to customers. If you talk to them and you say, OK, what's your actual utilization that you're getting from the CPUs, the GPUs that you rent? In CPU land, it's around 20%. In GPU land, for most customers, it's sub 10% in terms of that utilization. From the very beginning, and again, 100% utilization is almost impossible, but 80% utilization is very, very, very possible.
From the very beginning, when we spend a dollar of CapEx, because it is our responsibility in order to get that utilization up, as opposed to the customers, we can actually get more work done for every dollar of CapEx we spend. That is advantage number one. Advantage number two is that the very nature of how Cloudflare works is because every server runs everything, and we can move workloads to wherever there is capacity. There is always available capacity which is out there. Again, it is just a trade-off of performance might suffer slightly, but again, it is fine.
There's plenty of capacity which is there, which means that what we have been able to do across the business from the very beginning was spread load out across the network, see where there's demand, get paid for that demand, and then invest behind the demand as opposed to investing in front of the demand. If your business is selling a server, you've got to buy the server in order to sell it, right? Absolutely, you've got to invest ahead of kind of what your anticipated demand is. If tomorrow the next DeepSeek or whatever comes along and makes all of a sudden AI cost a tenth as much, you could get trapped with a whole bunch of CapEx spend that all of a sudden isn't actually going to be developed. I don't think that's what's going to happen.
I do think that as we make AI more efficient, there's just going to be more ways that we use it, and it'll soak up that demand. In our case, we're not selling a server. We are selling getting work done. That means that we can sell that getting work done. We can deploy it across our network, and then we can invest behind it. CapEx as a percentage of revenue, I think it's going to hold really steady over time. I mean, again, it might move up or down quarter to quarter, but over time, I still feel, even with this new opportunity, I still feel really comfortable with what our long-term model is for CapEx as a percentage of revenue. Does that mean that CapEx will go up? Yeah, but so will revenue. I think that those things will hold in place.
That's over and over and over again, the thing which differentiates Cloudflare is we're able to invest behind the demand as opposed to having to invest in front of the demand. That's because the business that we have is fundamentally different than we have to buy something, keep it in inventory, and then sell it out.
Got it. All right. Let's shift gears, but stay on the AI topic a little bit. In your founder's letter that Michelle and you wrote in September, then again on the Q4 call and your comments, you said that the current business model of the web, the quote was, is eroding because of AI. And you're not just CEO of Cloudflare and co-founder, but you and your wife also own a content creator. It's like in the Park City Record newspaper.
That is so exact.
This is for both your hats that you wear. Can you walk us through first how the internet business model is breaking, like you said? Two, dig into what you meant by post-search web. And then third, help everyone understand Cloudflare's potential role in helping figure out that business model.
Here is the challenge. We have really good data on how people and businesses interact online. If you go back 10 years, for every visitor that Google would send you, they would crawl two pages. The ratio was two crawls to one visitor. If you think of the value as a content creator, there are lots of ways that you can derive value. Some of them are monetary. You can sell subscriptions. You can put ads up and get paid for that. You can also just derive value out of just the ego of, I am a huge fan of John Candy. I have documented everything about John Candy. I am not personally, but you could imagine this. Somebody creates a John Candy definitive blog.
They may not monetize it in any way, but they look at their traffic numbers and they're like, wow, it's really amazing. Millions of people are coming to read my content. They get sort of an ego value out of that. That is 10 years ago. Two to one, the ratio. Two crawls to every visitor. What is it today at Google? The answer is six to one. Why has it changed? Because 10 years ago, if you searched like, how old was John Candy when he died, it would give you 10 links. One of them might be to the John Candy official blog fan club. You click on that, you read the article, and it's somewhere in there. It says it. Today, Google will pop up with an answer box that says, John Candy was however old he was when he passed away.
You do not click on the link. Today, it is six crawls for every visitor that they send. The rate of crawling at Google has stayed exactly the same over that period of time. What has changed is the user experience, which is there. If you talk to publishers, if you talk to anyone in the creative space, and you say, how much harder has your job gotten over the last 10 years in terms of deriving value from whatever you create? I have had a bunch of these meetings over the last 18 months. Almost universally, they are like about three times harder than it used to be. That lines up. As we all know, publishing is struggling today. Original content creation is struggling today at a six to one ratio. That is the good news. If you look at OpenAI, it is 250 to one.
If you look at Anthropic, it's 6,000 to one. If publishing is struggling at six to one, I think it's dead at 250 to one, and it's forgotten about at 6,000 to one. The problem with that is we all suffer, right? There isn't original content which is being created because, again, however you want to value that is gone away. Over the last two years, a bunch of our customers are publishers. They have said, historically, we've thought of the enemy as being the Chinese hacker or whatever. That's why we use Cloudflare to protect it. Increasingly, we're concerned that there are these AI companies who are taking our data and not giving us anything back in return. I think that that's just an existential problem for the internet.
If you talk to Sam at OpenAI, if you talk to the leading AI companies, I mean, OpenAI has said from the beginning, we should pay for content. Sam does not want to look like a sucker because he does not want to have to pay when nobody else out there is paying. Someone needs to help publishers actually control effectively the supply. My kind of utopian vision of the future, I do not know that we will get here, but humans should get content for free, and robots should have to pay a lot for it, right? Or should at least share in the upside of whatever these AI companies are doing. We announced last September that we were going to, first of all, give any content creators the tools to be able to understand who is taking their content.
Second, we were going to create the ability for them to say, you're allowed to have it because I've done a deal with you, but you're not, and give that control back. Third, I think over time, there needs to be a rate card where publishers can just say, for the new AI company that comes out, if you want to take my content, here's how much you pay for it every time you scrape it. Here's how much you give me every time your content is used in a result and deliver that back in a way. That has to be the future of what is going on because, again, we're not in a world where you run a search and you get back 10 blue links anymore. You're now going to run a search and get an answer back.
The fuel that powers those AI engines is original content. If original content goes away, the AI engines go away. I do think that we've got to figure out a way in order to make sure that original content creators are able to get value in this new AI-driven web. I think Cloudflare's in a really unique position because so many of the AI companies use us. So many of the original content creators use us. If we can help bring them together in a way which is actually sustainable, I think that's important for the future of the internet.
Got it. All right. Last question in two parts. Good fill, evil fill. You have talked about playing the long game. It is a general theme at Cloudflare. Cloudflare is also about to celebrate its 15th birthday this September. What do you think about Cloudflare on its 20th birthday? What are you most excited about in terms of potential of a company? Conversely, though, evil fill, what keeps you up at night?
You know, I think that, let's start with what keeps me up at night because I think that that then feeds the answer to the second. The internet's under attack, and it has been for a while now. The analogy that I've used sometimes is that the first 40 years of the internet up until 2016 was a little bit like episode four of Star Wars, which is a terrible movie if you actually go back and watch it. I mean, the plot is incredibly naive. Random farm kid discovers this miraculous power, which is essentially governed by this group of weirdos. And then basically through thinking nice thoughts and closing his eyes, is able to destroy the most powerful kind of weapon and sort of object of control in the universe, and then everybody celebrates. I mean, really naive and simple movie at some level.
Yet that is the first 40 years of the internet, except replace the Force with the internet, replace Jedi with developers and network administrators, replace the Death Star with what were the traditional kind of sources of power throughout human history, which is government, religion, education, family, and media. The internet disrupted, blew all five of those things up in major ways. Again, we can debate whether that was good or bad, and there are pros and cons. That is the story of the first 40 years of the internet. In 2016, things changed. While episode four is a terrible movie, episode five, The Empire Strikes Back, is actually a really great movie. That is what we are living in right now. Like the AT-ATs have landed on Hoth, and we are Luke kind of hiding in the, I do not know, the weird ice cave thing.
I think it's going to be a real challenge. What's been hard is regardless of what you think of them, I've always thought of kind of Cloudflare as like the younger plucky brother with five big brothers. Those five big brothers are like Google, Meta, Amazon, Microsoft, Apple. We could kind of pick fights, and then they'd help us kind of finish them. One of the most depressing calls that I was on over the last little bit was there was a legal ruling that was kind of just bad for the internet. We and Google and a bunch of others were kind of involved. We called Google up, and we were like, hey, let's go. We know how to take this on. Let's go fight this fight. Come on, big brother. They said, you're right, but we're tired.
By the way, we're an AI company now. We're not up for another fight to preserve the internet. We're hearing something similar from all five of those companies these days. What I worry about is that if they're not the ones fighting for this thing, this amazing thing that has given a lot of us a career, let us build Cloudflare, I'm not sure who is. I think a lot of that's going to fall on us. I think a lot of the work that we do behind the scenes to really fight for the internet is critical because I think a lot of the historic defenders of the internet just aren't up for that fight anymore for lots of different reasons. That's what keeps me up at night.
When you see what Russia is trying to do to recreate sort of the controls that China has, Iran is doing the same thing. Turkey is doing the same thing. India is doing some things that are very, very similar. Brazil is doing some things very similar. Like someone's got to push back on that and say there's actually real value in an open, connected internet, which is there. That's what worries me. What I think is super exciting, selfishly, is I think over time, the internet still wins. I think the people who are defending it today, while it feels like a very lonely job, are going to be in an incredible place to kind of win the future.
Time and time and time again, we're seeing from our customers, we're seeing from policymakers, regulators that they really do respect how principled we've been around this area. They respect the opportunity that we have. I think that not only if you talk to employees at Cloudflare, yeah, they're super excited that we're on a path to more revenue and more customers and those things, but they're also really driven by that mission. I think the fact that we have been able to build a great business, but build a business that actually really matters in the world, is something that allows us to get the best talent. It allows us to have the success that we've had. That's what I'm most excited about over the next five years.
All right. Thank you, Matthew. That is the end of our fireside chat. We will be back up here at the end of the day for open Q&A from the audience.
Thank you very much.
I'm very excited to introduce our next presenter. This is a man I've known for many, many years, really marveled at his track record of really delivering innovation at scale. The key word there is at scale. Couldn't be more excited to have him inside here at Cloudflare to do that once again. CJ Desai .
Hello, everyone, and good afternoon. I'm delighted to be here. I'm seeing some familiar faces, so thank you for coming. The question that I get asked the most is, why Cloudflare? Among many choices that I had, which Matthew has shared in the earnings call, fundamentally, it boils down to I am absolutely a believer in platform strategy. Platform strategy, being a platform company, and making sure that the platform is a single platform with a single architecture on which you can build multiple products is definitely super attractive. With the experience I had in products and engineering at multiple companies, starting actually with Oracle, was always about the platform and building applications. We built financial applications when I was at Oracle.
Here, this one programmable network, which is very composable, which I'll go into the details, was definitely super attractive because what it fundamentally does is that you have a massive TAM, right? Network is getting disrupted. Most customers that I've spoken to after joining Cloudflare have one or the other network modernization project going on. In that, we have a role to play. We have the ability to execute, and we can absolutely win as customers are modernizing network. It is a massive TAM. Second, I think Matthew touched on it, super innovative company. Releases in my past job, our releases were every three months or so. Here, when I speak to the engineering team, product teams, we can do daily releases. We can do weekly releases, and the code just gets deployed on the global network. Super innovative culture, very customer-driven.
We get the feedback, and we figure out how to solve the problem. Innovation is the hallmark of this company. That is the number two. I think on the potential, which I will touch on, when I look at the industries, when I look at the geographies, when I look at public sector versus commercial, there are a lot of places where Cloudflare has absolute right to play and being the fundamental network infrastructure for those companies. Those are the three main reasons. I would say at the end of the day, the mission that Matthew just touched on to help build a better internet is something that is appealing to me.
I was at the University of Illinois when the browser was created, which transformed the worldwide web with HTML standards and others, and accelerated the internet adoption from just people who are focused on research and academics to the population around the world. Now, one programmable network and one composable network. Let's just go into the details. I know some of the things got touched on in the earlier session. Having every Cloudflare service that can run on every server and that can go to every single data center that we run, which is close to 325 plus cities, that is phenomenal. If you think about it, very few architectures in the world are like that, that, hey, you create a service that can run on every server that is out there.
Every single, basically, data center, and we do not run in public cloud, is created by Cloudflare and operated by Cloudflare and globally deployed. I mean, that's just amazing. When you think about it, if we decide to create a product feature or if we decide to create a new product, which is a service because we have software as a service, and that gets deployed in minutes, seconds, depending on the type of feature and so on globally, right? You all have seen a version of this slide. Having 325 plus global locations, so it's 335. The big change, including the progress that the team has made in the first quarter of this year, is that more than half now also have GPU capacity to run the AI workloads, right? I mean, that's phenomenal. 190 plus cities can run GPU, as in AI inference.
50 milliseconds is something that we always talk about, that from every eyeball, you can have a very, very fast performance. The capacity of the network approaching now close to 350 chips per second. That is at a high level. From my standpoint, Cloudflare Network is the product. Cloudflare Network is the platform. Yes, you build multiple software layers on top of it. This is the competitive advantage. Given it has been architected this way, it is very, very meaningful to our customers. When I think about 2025, we are adding capacity in India for a variety of reasons. We are also doubling our compute capacity, so CPU 1.5x across multiple regions, and adding three new sites.
We are preparing the network at the fundamental level to plan for our $5 billion-plus revenue growth over years so that the network is resilient because all of our customers, whether they're free, self-service, or enterprise, completely rely on this global network. Now, how is this network built? When I speak to customers, I boil it down to a few simple things. Number one, security. Of course, security is the forefront. That's our beginning of the company. We need to provide security, whether it's for external users of our customers or employees or contractors. That's number one. Number two, I would say, is connectivity. Matthew touched on it. This connectivity at a global level while providing local controls is pretty phenomenal.
The flexibility of this network, right, where the developer tools, where you can build specific application on Cloudflare platform, like how we build it internally, also gives this power to our customers. The reason this is effective is, one, cost. Total cost of ownership reduces when our customers go from one product to two products to three products to multiple products or services they run. We provide centralized policy management. You are not going through various dashboards, various tools, various reporting. All your security policies are in a single place, depending on the type of users. We try to provide customer experience really, really well from a user experience perspective. I was talking to a Fortune 50 company, and the CTO really wanted to understand the portfolio of Cloudflare. While we were going through it, he said, CJ, just one thing I wanted to know.
I am a user of Cloudflare for my home network and was very proud to show me what he had built internally on his home network and said, we are going to open doors for you here in the enterprise because I'm a big fan of the product. Having our users, which are self-service or free customers, has allowed us to create a really configurable platform that can be managed by individuals or by enterprises. The focus on innovation, when I think about it, you see the journey and highly innovative, so many products. I think the final list is 60 plus products that we can sell to our customers. We provide that both via self-service as well as from the enterprise contract perspective. The innovation just continues.
Next week, as Matthew shared, is a security week where we are going to outline additional innovations, whether it's for AI and a few other things that you should watch out for. Really proud of this innovative culture where new products with a few engineers, because the platform fundamentals, whether it's user experience, whether it's the programmability of the network, is provided in the platform, which allows the incremental cost of innovation much lower versus a point solution. Super innovative company, Cloudflare is, and the team has done an incredible job across our X to continue to push the innovation to our customers. Now, one of the things you'll say, okay, you have the incremental cost to create a new product or a new feature because the Cloudflare platform provides so many fundamental capabilities. Unit economics are better from an R&D perspective.
Why the spaces that we go after, whether it is on our Act 2 with Cloudflare One, or there are areas within our Act 1 as well on application security, what you will see from Cloudflare is that we will start small, we will start fast, and then we will build deep functionality within our X, right? Because our platform allows us to do that. What you will see is if you look at a 2x2, initially, we may become a niche player. Then we may move to the shift to the right and then to the top. The analyst firms that we work with recognize that. As you know, customers look at these reports and we say, okay, now we are playing in area X, say, phishing protection or so on, email phishing protection and so on.
We will systematically improve over time by adding very, very methodical feature prioritization and ensuring that that product becomes best in class with a wide adoption. If I have to simplify Cloudflare platform, I would say three areas that we are very focused on. The area number one, which is our Act 1, providing connectivity and security to external users, right? That in itself, there is still a lot of TAM to be had. There is still a lot of traditional competitors who play in that space. We can continue to grow that business. That fundamentally, that market in itself is also growing. You come to Act 2, providing security for employees and contractors, which we call Cloudflare One. Here, we started a few years ago. Thomas will share some of the numbers from last year. This is a big market.
This is a big market. The penetration from other players is still not very high in the enterprise or in public sector. That is the second piece, providing security and connectivity for internal users and contractors. Number three, that Matthew touched on early on, is the developers and what can we do for them as they build new applications, whether they are AI applications or applications that take advantage of our connectivity and this global network. When you look at these three acts, right, and there are lots of details below it, just in existing Cloudflare customer base, if we had sold all the products, that in itself is a $7 billion TAM, right?
Because the land and expand motion with whether we start with Act 1 or whether we start with Act 2 or sometimes we start with Act 3 allows us then to get the permission from the customer and their trust to expand our portfolio, which is very, very wide. That is what is super exciting when I think about it, is that the day before yesterday, I spoke to a Fortune 10 customer who is not a Cloudflare customer. I said, okay, here is the portfolio. Here are the things we have in area 1 or area 2, as in Act 1, Act 2. The customer said, CJ, we did not know that your portfolio has become this wide for the enterprise. There are these three or five vendors, as in point solution vendors that we are using today, which we could potentially replace with Cloudflare.
Let's get started on one or two of those use cases. That's what this incredible opportunity, whether it's financial services, healthcare companies, or retail companies, Cloudflare has. Now, when you think about the landscape and when you speak to these enterprise customers or even mid-market customers, it's a fairly complex world out there. The world around 2008 or 2009, when the whole public cloud migration started, right, with AWS on the leading edge, this is the second half of the first decade of this century. That's when some public cloud movement started, moving some VMs, some workloads, and so on. You look at the second decade, between 2010 to 2020, and you had other Azure, GCP, and other got mainstream from a cloud perspective, including some in Asia.
When you think about an enterprise customer who is very focused on digitization, who is very focused on figuring out where they can leverage public cloud for whatever reason, I heard from customers that, hey, we have a philosophy that 75% of our workloads should run in public cloud, okay? There are some customers that I speak to which fundamentally say, we don't want to be dependent on just one public cloud. So we currently are using three for different types of workloads besides our on-prem footprint. Multi-cloud is a very acute phenomenon that customers deal with today, and it's 2025 that they are dealing with this massive landscape from on-prem and public cloud perspective. Now you add software as a service, multiple software as a service that our enterprise customers use. You think about the threat landscape. You think about the network architecture.
What you find, if they are able to draw a diagram like this, there is aging hardware. There is so much complexity in managing this network. The demand digitization is pushing on these customers that they have to modernize their network. While they modernize their network, as the user demand grows, they also have to make sure that it is secure, right? That creates fundamentally opportunity for Cloudflare. From a Cloudflare perspective, if we have to simplify this, right, you have external users, you have employees, you have contractors, you potentially have part-time workers, and you want a connected network that is secure at the end of the day, highly available, that customers can rely to run their network infrastructure.
From our perspective, this is a North Star that we can go to these large enterprise customers or enterprise customers and say, here is Cloudflare. Here is what it does. You can replace multiple point solutions by using this so that lowering the cost or TCO, better experience for your end users while you remain secure, right? That's how I look at it. Now, as some of you know, or most of you know, we have customers globally. One of the things that really impressed me a few weeks ago when I was in Asia, that we also have substantial brand names within our Asia-Pacific region customers that are doing really, really well, that they rely on Cloudflare. One of the large crypto customers told me that, CJ, we use one Hyperscaler for our compute and storage, and then we use Cloudflare for network and security.
The thing he said is that you are the digital front door for all of our 100 million monthly active users, and we rely on you for security. We do not rely on you for rely on Hyperscaler for us. I mean, that's a very big statement coming from a company in Asia-Pacific that is doing really, really well. Between the public sector and Fortune 500 and others, and of course, digital natives and AI natives, the customer base is very, very wide. Now, when I talked about the potential, the thing that I was trying to really understand within my first few weeks here was the penetration of Cloudflare, and I see this as a massive opportunity, is fairly low compared to the TAM, right? That's why Mark Anderson was hired to build an amazing enterprise go-to-market team. I'm delighted to be working with Mark.
He's the best. When I think about this, you look at this opportunity, and then you can add retail, or you can add public sector, you can add global public sector, and you will see there is a huge headroom for our Act 1, Act 2, and Act 3 products for Cloudflare. I mean, this is similar to when I joined the last employer I was with, which was not penetrated as much. Over time, we really, really focused on these types of customers and got a much higher market share. This is a massive opportunity from my perspective. Now, organization needs, they vary. Sometimes when I meet these large enterprise customers, they say, CJ, security for external users, we want to replace some of the legacy incumbents, and that's where we want to use Cloudflare.
Sometimes when I meet the customers, they'll say, hey, the first-generation vendor, when it comes to our employees and contractors and making sure they have secure access, we may want to replace that with Cloudflare. Sometimes certain AI native companies will say, hey, can you tell us more about the developer platform and what we can build? Sometimes it's a combination of all three. One of the large customers that I spoke to last week basically told me, can we use this thing from your Act 1, this thing from Act 2, and from Act 3, can we build an app that is solving for some of our B2B traffic, which is phenomenal, right? I personally feel that the portfolio is wide, but also really, really good in terms of what we propose to our customers.
Now, any modernization project, people will say, we want to go with SaaS architecture. We want to go with cloud architecture. Typically what you will see is that they may buy a point product and then eventually feel like the project was successful and it got some returns. Cloudflare positioning with the platform and the products we have, that once you have first product in, and as Matthew said, that might have a higher barrier to entry, but once you are in and you deploy it and you configure it, to have the incremental products, whether it's to replace your existing DDoS mitigation solution or whether it's something related to application firewall or API Shield, whatever the case might be, you start seeing the benefits because you have a center of excellence related to Cloudflare.
You can just say, let me turn this on, let me turn this on, and continue to use the products while I take care to get rid of some of the traditional legacy solution. I've been spending a lot of time really understanding customers use Cloudflare for what and why, right? Canva is a great company. You see a lot of them on billboards, even in Manhattan. They are also on the forefront of AI. If you go to their website and if you're a creator, it will say, hey, we'll help you create a very easy text campaign or some images or some video using AI. They also, similar to Cloudflare, have free customers, self-service business customers, and of course, enterprise customers. They are using us across all three acts, right? That company is a big company, right?
It's a digital native, but it is a big company that is taking on a very competitive market and doing really, really well. If you think about them, they are using us across three acts. When I think about this brand, as in Porsche, they initially started in 2022 with 3,000 websites that are dealer-facing for sales and service and converting them to use Cloudflare for security. Because of European Union regulation and some data localization requirement, they used our worker platform to create apps specifically for that region, right? This is the real advantage of Cloudflare, that while we have a global network, we will still comply with data regulation and other things if necessary and demanded by a great customer like this. Now, Matthew talked about the mission.
All the employees of Cloudflare, when you ask them, because I have asked them when I joined, I said, hey, why are you at Cloudflare? What keeps you here? Everybody is aligned to help build better internet. They will say that, okay, CJ, I'm excited because one of the engineers in Germany I spoke to yesterday said, I'm excited because we help sometimes with governance on the internet. Sometimes we make sure that, as you see here, that advocating for network neutrality. You can see some of the bullet points here, but Cloudflare does really, really care about making sure that we are helping build better internet because internet is the foundation for every business. Network is the foundation for every business. We want to make sure that it's open, it's connected, it's secure, and it's available.
That is a fundamental thing that we stand for, and it's a very purpose-driven company. Now, 2025, one of the questions I get asked, okay, CJ, you have been here. What do you think? Where do we need to go? I wanted to share just two or three lessons from my past, okay? The one thing that I learned at ServiceNow was the land and expand motion. What does that mean really for product and engineering? What does that really mean? You have to really, really focus that customers get value out of whichever product you land with, as in first product that they buy. That is absolutely applicable here as Mark Anderson and the team try to run the motion of land and expand within the go-to-market.
Number two, a few years at Symantec in the first decade of this century, being there for the customer when they need us, if there is a security incident or if there is a network that they want to rely on. Just a couple of days ago, a large customer had an issue with their incumbent. They switched over to Cloudflare. The way the team helped them out while they were under attack is super, super meaningful and creates a massive customer loyalty. Number three, what I learned was how does the go-to-market team work really well with engineering as we scale our enterprise motion? We have such a massive TAM in enterprise and large enterprise, including public sector, that if the teams really partner well together, and that's why I'm excited to be working with Mark Anderson and the teams, magic really happens.
These three are very applicable when I think about Cloudflare and what my previous experience can help as we scale this company. What's the plan for 2025? First of all, it always starts with the network. Network has to be super resilient. Oh, by the way, while it is being resilient, is it really going to be there to scale for $5 billion-plus revenue? When I say $5 billion-plus, our free customer base is growing, our enterprise customer base is growing, our self-service customer base is growing. We have to make sure that network stays resilient and we continue to add the capacity, as I touched on early on, on the network layer. Second, the foundational platform.
One of the biggest asks from the enterprise customers is, CJ, we are complex and we need Cloudflare's help to ensure that the platform really works within our infrastructure. Whether it's identity and access management, role-based access controls, and many other things that enterprise customers demand, even from an integration perspective, we are investing in it and we are investing and making progress so that enterprise customers, large enterprise customers, feel that we are serving them at the highest levels. Most of them tell me, you have a great piece of technology in this area or you have great technology in this area, but we really, really want you to give us very robust enterprise-level controls. The good news is we know how to do that and we can give them.
When I think about the portfolio, making sure that we have the fastest network, we have the most secure network, whether it's for external users or internal users, that is something table stakes and we'll continue to make progress there. It's a race with no finish line. Number two, on Cloudflare One, when I think about this portfolio and the massive TAM it has, we wanted to focus in certain areas. And one of the areas that we are focused on with these large enterprises, or you think about retail locations, bank branches, and so on, is coffee shop networking. Making it simpler rather than using a variety of point solution for a remote user versus a branch location versus a corporate head office, can you just use Cloudflare? It will be fast and it will be secure, right? I mean, that's it.
That's the simple message that will provide secure access, will provide secure web gateway. Whether it's coffee shop networking, by aligning fundamentally principles between network and security, you get the best of both worlds when it comes to your employee base, right? That's how I would classify on our Act 2. You'll hear from Rita on Act 3, a lot of innovation that is going on just right now. Last point on customer support, we have massive customer base. We are expanding our enterprise customer base. How can we serve them even better than we are today? We are going to focus on providing a great customer experience for all of our customers. That's the work both Mark Anderson's team and my team are doing jointly to do that. My final remark is where I started.
I joined this special company because it has platform architecture, massive TAM, and potential in the enterprise and public sector. That really excites me. Very innovative company. Just by focusing on certain foundations while staying innovative across Act 1, Act 2, and Act 3, I will do my small part in building a great iconic company for the future to come. Thank you very much. I am going to invite Rita to talk about our AI strategy. Welcome
. Thank you. Thank you, CJ. Hello, everyone. My name is Rita Kozlov. I am the VP of Product for Cloudflare's developer platform. It is the year 2025, so obviously we are going to talk about AI. I promise I will mention the hot word right now, agents as well. Before we get into all of that, I would like to start with Cloudflare's vision for developers, which I think is really inspiring.
Cloudflare's vision for developers, in a nutshell, is to make it possible for every single developer out there to bring their ideas to life and to make that as easy as possible from the moment that they write their first line of code to deploying it to the first user and the millions of users to come after that. Here's the thing. Making something easy is actually, in and of itself, really hard. To do this, when we started on the developer platform eight years ago, we had to take a really big, really bold, and even non-consensus bet. That was to build our architecture for the platform on top of something called isolates. Now, Matthew talked about this before. All code running on the server up until this point had been running on a virtual machine or container.
The problem with that model is that it makes it fundamentally impossible to abstract away infrastructure in a way that really allows developers not to worry about it. We know this actually because Cloudflare's entire first act, when you think about it, has been about bridging that gap between the way that the centralized cloud running VMs and containers works and the way that users actually interact with the internet. Now, with isolates, isolates compared to VMs or containers are really, really lightweight, which means that they can spin up really, really quickly and reliably, and they give you that platform that you need in order to be able to run fast. When we made this decision, we knew that we were making a trade-off, that it would be harder to lift and shift existing applications onto Cloudflare's platform.
When you're making a bet, you want to bet on the future and not on the past. We wanted to build the best platform for building greenfield applications. What's really incredible is that today, if you're a developer, Cloudflare offers you an entire full-stack platform of offerings, everything from compute to storage to AI and media, literally all of the different pieces that you need in order to build an application. Every single line of code that you write gets instantly deployed to region Earth, which means that you never have to think about scaling it. It's just fast, it's performant, it's reliable, all of these things from the very beginning. Now, a few clever developers clued on to Cloudflare being their secret weapon for running ahead of everyone really, really early on.
For those developers, I have a little bit of bad news. It is that the secret is out. This platform is not so secret anymore. Today, there are over 3 million developers that are building on Cloudflare. We are continuing to see that number grow and accelerate. Part of this is due to this large paradigm shift that we are seeing with AI. We started to talk about the shift a year ago and how it represented something as big as the cloud, mobile, and social before it. That statement is truer today than it actually was a year ago. I realize that I am talking to a finance audience and you all get held accountable to your predictions all the time. I thought I would give you a chance to hold me accountable to some of my predictions from a year ago.
A year ago, we talked about how 44% of developers, no surprise, were already using AI to help them write code. Analysts were predicting that by the year 2030, about 50% of knowledge workers would be using AI to help them augment their day-to-day tasks. I remember seeing this figure and thinking, they're being a bit conservative. Guess what? They were. Today, not in five years, but already we've surpassed that figure with 75% and not 50% of knowledge workers already using AI. The number of developers, that almost doubled as well from 44% to 76%. There was another actually more interesting prediction that we made. It was that workloads in AI were going to shift from training to inference. Guess what? We're starting to see that play out in real time as well.
We're seeing that with OpenAI reporting that their new O1 model is now spending more and more time in inference than in pre and post-training. We're seeing a similar thing with DeepSeek, who is optimizing training so much that they're starting to spend, again, more of their time in inference. This is all a year ago. This is all a recap. Let's talk about what's coming next. While we're still in the early innings of that migration from training to inference, the next shift that's going to happen is going to be around inference being focused rather than on augmentation to inference being focused on automation. When you hear the word agents, that's exactly what people are talking about. I know that there's a lot of jargon that's flying around right now, and I think people are using the word agent very loosely.
I wanted to share my framework for thinking about the different types of AI. The first type of AI is predictive AI. It's been around for a long time. It's often referred to as machine learning. Cloudflare has been using predictive AI since the very, very beginning of the company. A really canonical example here is taking something like all past internet traffic and looking at it in order to be able to predict whether a new piece of traffic is a DDoS attack or someone just having a Black Friday sale. Now, generative AI is the new type of AI that's come in over the past couple of years. As the name would suggest, it allows you to generate something that didn't quite exist before. That can be a piece of code, some text, an image, a video.
Generative AI has largely been focused on making us more efficient through augmentation. Rather than having to manually, artisanally handcraft an email, I can now ask AI to assist me with that. Now, agentic AI is focused not on just the augmentation piece, but the full end-to-end automation of a task. Let's run through an example. I think everyone here is familiar with a UI that looks something like this. Maybe you've asked it for some help with a task recently. If I'm a junior salesperson and I need to follow up with a customer, I might ask it to help me write an email so I can follow up more quickly. With automation or with an agent, AI cannot just help me write the email, but it can help me automate the entire end-to-end flow of what you might ask a junior salesperson to do.
If I went to a conference last week and I met a whole bunch of people, I can ask an agent to generate me a list of all of the people that I talked to, then to write me that email. It is not going to stop there. It will let me review that email if I want to, but then it will actually send it on my behalf. When a customer follows up, it can either notify me or even write back to the customer. How do agents work? Agents are really made up of three core pieces. There is the AI piece. That is the thinking of the operation. There is the workflows piece. That is going to be the executive arm that helps the AI then go and make sure that those things all happen.
There are the APIs or the tools that actually go and make those actions. Now let's take a look at this through a developer's perspective. What would it take for someone to build an agent? First, you need a way to get user input. More and more, we're seeing agents that are communicating over voice. For that, you would need something that talks over WebRTC, which is the Internet Voice Protocol. You need to convert that from speech to text in order to execute those next steps. You might also use a chat UI. That's a very common pattern. You need somewhere to host it. From there, you need somewhat of a gateway layer so you can cache responses.
If a query comes in that you've seen before, you want to be able to respond fast as opposed to doing the whole thing all over again. You want to be able to observe the AI, make sure that it gets better over time as opposed to worse. You need access to the actual large language model, the LLM that's going to generate all the content, that's going to come up with a plan, that's going to constantly be doing the reassessing of it. The LLM needs then access to an orchestration layer, kind of like the workflow. For the orchestration layer, you need a unique combination of compute that can actually do the task, but you also need some storage in order to keep track of all of the tasks that have been completed and make sure you're executing on the next one.
You also need access to a whole bunch of different tools. That can be a browser. If I go back to the example of a salesperson sending an email, if that email needs to be personalized, you might want to go to your customer's website, make sure you're including relevant information from there. You need an API for sending the actual email. You need maybe to connect to an internal service to pull up any open support cases to make sure that you're addressing them in the email. You need a vector database to hold all of the domain knowledge that's related to your own industry. Finally, sometimes AI is going to need to ask a human for help. This is commonly known as a human-in-the-loop architecture.
Now, when we talk about building out all of the primitives and all of the pieces that developers need to build a full-stack application, agents are a perfect example of that type of application. Today, Cloudflare provides all of the different pieces that you need in one place in order to build an agent. We offer everything from Real Time, which was previously known as our Calls product, to take in that voice input. We have Pages that can help you with the UI. We have our AI Gateway that can help you with caching responses and with constantly evaluating and observing your AI. We have Workers AI for hosting the models themselves, whether that's the LLM or the voice-to-speech, the voice-to-text piece.
We have all of these different tools for your agent to be able to connect to, whether it's a browser, an API, an internal service behind Zero Trust, or a vector database. We have actually the perfect pieces for that orchestration layer. If in the past you've seen us talk about a primitive that might have seemed a bit more obscure, like Durable Objects, it is actually such a good building block for exactly an agent because it takes those two things, compute and storage, and puts them together into one piece. Now, I'm not going to take credit for having invented agents. Though funnily enough, we did consider agents as a name for a Durable Objects product that would have been very prescient. We did inadvertently build the best platform for building agents. It is not just about having all of these different pieces.
You might be thinking, well, don't cloud providers have hundreds of different services? Surely you could spin something like this up on one of them, right? What makes Cloudflare unique? Why is Cloudflare the best platform for building agents? This comes down to three key reasons. The first is the cost and scalability that are always going to be intertwined with each other. The second is performance. The third is developer experience. Let's get into it. I'll start with the cost and scalability piece. Within that, I was talking before about how agents are made up of AI, workflows, and APIs. Let's start with that AI piece. Now, Matthew was talking before about how GPU utilization tends to be really low for companies. It can float at around 30%. For 30% of organizations, it's around 15%. Why is it so low?
If you were to run a workload on a Hyperscaler, the very first thing that you need to do is pre-provision all the resources you're going to need. Now, for training, this is very easy because you have all of the variables ahead of time. You know how large a data set you're looking at. You know how large a model you're going to train. And you're fully in control of the start and the stop buttons. Inference, on the other hand, is incredibly unpredictable. This is because inference relies on human behavior. That's going to rely on what time it is in the day and what people are doing. Many of our customers are also launching inference products for the very first time. They have no frame of reference to look at for how a product is going to do.
What they have to do is hope for the best and provision their capacity right around that top line. Regardless of how they do, that's just how much capacity they need to have in order for the best-case scenario. What ends up happening in reality is that, again, that traffic is going to continue up and down. With any other cloud provider, you are paying for that 70% of resources to, in reality, be sitting there completely idle. With Cloudflare, we took a serverless approach to inference, which means that we can spin up those resources as you need and scale up, and that we're only going to charge you based on the usage that you end up having. All of this blue area is what the cost savings look like for our customers.
We are not doing this just because we are being nice, although this is a nice benefit for our customers. It also helps us get better utilization out of our resources because it means that when one customer is having that trough in traffic, we are able to schedule another workload on those exact same machines. This is just the AI piece. Now, let's talk about the workflows piece. I was talking before about how the orchestration layer needs to connect to all of these different pieces outside of it. All of these are, again, outside of your control frequently. It can be an LLM that is going to take several seconds to think. It can be a human that could take minutes, hours, days to respond. On any Hyperscaler out there, you are paying for that entire duration of the request.
This is known as wall clock time because if you were to look at the clock on the wall, from the moment that an operation starts to the moment that it ends, you're paying for that entirety of time. I talked in the beginning about the big bet that we took with isolates. This is where it really pays off because containers are so lightweight that we're able to spin them down when we're waiting on something else to happen and then spin them back up just as quickly. This is something containers can't do. They're just not going to spin up quickly enough. You have to leave them on the entire time. We're able to pass those cost savings onto our customers.
I'll use an analogy that I think every single person in this room is familiar with, which is if you've ever been in a cab to JFK and you're sitting in that stop-and-go traffic, there is nothing more unnerving than watching that meter tick up, even though you're not moving. This is what it's like to be paying a Hyperscaler for any of these workloads. Now, if you can imagine that you would only pay for the bits where you're actually getting closer to your destination, that's what it's like to be building this on Cloudflare. Cost and scalability was the first piece. Now, let's talk about performance. Keep in mind that for agents to really take off, performance needs to be really, really fast because our expectation of agents is going to be the same as it is of humans.
Cloudflare is in the perfect Goldilocks zone of being able to service these use cases because we're able to get as close to the user as possible, but with the entire power of a GPU or an entire data center behind it, as opposed to devices that are always going to be just too small to be able to run the types of workloads you need for an agent. It is not just the ability to get close to the user. It is about that end-to-end experience of an agent. This is where we built a really connected and a really smart network that's able to move workloads around based on where it's going to make the most sense for the entire flow to run.
Sometimes that does mean that you're going to have the AI piece, the API, all of these things running in the same data center, which is going to lead to the best end-to-end performance. Now, the third reason Cloudflare is the best place for building agents, and this is the one that's closest to me because I started my career in software engineering. It's the developer experience. I've built applications on the Hyperscalers. The thing that's so painful about it is that you spend about 50% of your time on things that, again, are not getting you any closer to your destination. Actually, in this environment where things are moving so quickly with AI, every single day there's a new innovation. You just don't have time to waste on all of that.
Whereas, again, on Cloudflare, because you get all of these benefits out of the box, you do not have to spend time on them. That is time back that you get as a developer to actually build the best application possible. By being able to combine AI, compute, storage, our network, and then obsessing over the developer experience, we can create experiences that really resonate with developers. In fact, recently, a customer was telling me about how ever since they moved their development from a Hyperscaler to us, they have a new problem where their developers are going so fast that their product managers are having a hard time keeping up with them. That is a world-class problem to have. It is not just about having the pieces too. It is about making sure that they all connect and integrate with each other in a really seamless way.
This is what we spend a lot of time thinking about in optimizing developers' workflows. Just a couple of weeks ago, we announced our agents framework that allows developers to spin up an architecture that's quite complex like this all in a single command. That is what we talk about when we talk about developer experience. Why is Cloudflare the best place for building agents? Sun Microsystems had it better than we could have. They were just a few decades early. It is because the network is the computer. For agents, you need a really smart computer. You need a really interconnected computer. You need a really private computer, secure computer. Cloudflare's network offers all of these different things. We know that this is resonating for developers because developers, sometimes they tweet about it.
More importantly, when developers are really excited about a technology, what they build is just really incredible. Today, already over 1,300 AI companies are building on top of Cloudflare. I'll share just a couple of the use cases with you today. I am really, really inspired by the innovation that's coming out of developers. One use case is in finance. It's an agent that helps out with actually due diligence. It will go and research everything about a company, including the funding that they've raised, the founders, the team size, their budgets, all of that, and help you come up with the next decisions to make. There's even a company that's actually in the medical space that's helping build an agent for clinical trials.
It will go and analyze and summarize all of the data that's been collected so far and help doctors with those next steps, just as you would expect of a lab assistant. These are just the tip of the iceberg. We're seeing so much innovation coming out of developers every single day. Before I go, I wanted to leave you with one final thought. We talked in the beginning about the big bet that Cloudflare took with structuring our platform on top of isolates. At the time, we got a lot of skepticism. People were thinking, well, it's a pretty big trade-off that you can't take applications and lift and shift them. Yes, you can be the best in Greenfield. Is that market even large enough? Aren't the incumbents too big?
This reminds me actually of something that happened in another industry where technology completely transformed the landscape. Fifteen years ago, you might have asked similar questions about photos. Cameras had existed for over a century, and the players were pretty well established. A new technology came along, and we started carrying around a smartphone in our pockets at all times. This completely changed the trajectory of the industry. All of a sudden, in the matter of a couple of days, the same number of photos was being taken as would have taken years to take before that. We're in a similar inflection point right now with AI, with AI driving an explosion of new applications that are being built. In fact, over the next five years, more code will be written than has been written over the course of the entire history of software.
For those applications to be successful, you need a platform that's going to make it as easy to deploy that code as it is to write that code. That's exactly what we've built. The Greenfield opportunity is bigger than it's ever been with AI and now agents. I can't imagine a company better positioned to win all of that opportunity than Cloudflare. It's an exciting time to be a developer. I'm really honored to get to work on this every day. With that, I'll pass it over to Mark.
All right. Good afternoon, everybody. Thanks so much for being here. My name is Mark Anderson. I'm the President of Revenue here at Cloudflare. Excited to talk about what's going on in go-to-market.
My first 13 months on the job, I've probably done about 250,000 air miles meeting with customers, prospects, partners all around the world in every geographic theater and our people. At this point, a lot of these large customer meetings were with C-level executives at traditional enterprise companies, companies that maybe were a little outside of the realm of who we sold to in the past at Cloudflare: banks, insurance companies, governments, and whatnot. The conversations were remarkably consistent, which was eye-opening to me. We've all felt in this industry since the end of the financial crisis, IT budgets have gone crazy, right? They've doubled from 2010 to 2011 for IT. Infosec budgets almost tripled in that time. More and more people being hired, more and more apps, more and more infrastructure. It was almost like easy money.
When your budgets are going up every year, even the most progressive CIOs and CISOs were pretty less safe there about tackling technology debt. The very best of them kind of kicked the can a little farther down the road. However, customers in the last two years that I've talked to are reporting—not everyone, but a lot of them are reporting budgets are flat to down. I think this is a massive forcing function in our industry right now. When budgets are going up every year, we all witnessed a low sense of urgency on prosecuting that technology debt. What do I mean by technology debt? It's the sprawl of vendors, infrastructure applications that every large enterprise has tolerated for decades. In fact, most of the big banks that some of you all work for here still have IBM mainframes that they're running critical applications on.
With budgets constraining, companies are being forced to confront that technology debt faster than I've ever seen it in 30 years in tech. I think the pressure is on. This is going to drive what I think is going to be a massive consolidation faster than ever before to platform players and fewer vendors, fewer neck-to-choke, as some of my favorite customers like to say. At the same time, Gartner is telling us that companies are looking to get higher revenues from their IT and Infosec investments, especially around generative AI, not just tech companies, but every company. Years ago, companies were afraid of being Airbnb'd or Amazon'd in the cloud era. Now, I think companies are worrying about their competitor down the road doing a better job with AI and taking better care of their customers, maybe putting them out of business.
The problem is, however, 70% of large-scale tech programs fail to meet timeline, budget, and scope expectations. This has to change. This technology sprawl across legacy networks and point products highlights the mess that these executives must face. This requires a complex balancing act to address five critical and often competing IT priorities: user experience, resiliency, security, agility, and cost. Modern users are increasingly expecting seamless and intuitive digital experiences, yet delivering these at scale constrains system resiliency and really drives up cost. At the same time, this growing sophistication of cybersecurity threats demands robust defenses, often constraining operational agility and impacting the overall user experience. Now, Cloudflare defined the notion of the connectivity cloud as a way for customers to connect, protect, accelerate, and build their businesses so that they can be more agile while retaining control.
As companies move aggressively to a more digital native environment, being able to connect your infrastructure, your SaaS providers, your developers, your employees, and customers just becomes mega imperative. Cloudflare, from my perspective, is the only cloud-native organically built network out there in the modern world. Customers everywhere that I talk to, they get this massive differentiation. When I stepped off the board 13 months ago, I was super excited at the opportunity here at Cloudflare. At last year's investor day, I stated that I really believe the market is coming to where we are already. When I talk to customers, I see the light bulbs going off more and more every day. The potential in front of us, as you heard from Matthew, as you heard from Rita and CJ, I think it's massive.
Customers are just waking up to the fact that they need Cloudflare now more than ever. Let's talk about what we're doing to capture the significant opportunity that's ahead of us. Stepping back for a moment, again, while I was on the board back in 2023, we knew we needed to start transforming our go-to-market operations. We knew that the mountain that we were about to climb kind of felt very far away. We navigated a lot of change and made a lot of progress in 2023 and 2024. I'm not going to be satisfied until we've built a world-class go-to-market organization. That's how much of my career success has come, is building these high-performing teams. In 2025, I'm focused on four incremental key actions that are going to help us scale and enhance our go-to-market momentum to take Cloudflare to $5 billion and beyond.
Our momentum to bring the right AEs and sales pod team members to customer teams globally is going full steam ahead. This ensures that we engage with senior executive relationships at the biggest spending prospects and customers, regardless of segment, regardless of vertical, regardless of geo. To drive this momentum and focus on developer products, I brought back the amazing Ali Cabral to Cloudflare. We have created a dedicated speedboat organization that you are going to hear from Ali directly talking about our developer go-to-market motions. I will bring her up specifically to get into the details. I will tell you, 60 days into this move, we have driven a ton of excitement in the market and are seeing a lot of opportunity here. We also made significant progress with our partner-first motion that I started early last year. First, let me go back to business-to-business go-to-market university.
From last year, you remember my equation for momentum is the product of mass times velocity. In sales, momentum is ACV is the product of sales capacity times productivity. Now, during the fourth quarter, we again delivered double-digit year-over-year improvements in sales productivity. This continues the upward trend from the trough that we saw early last year, where productivity levels were consistently achieved in the 2021 and 2022 cohorts. We're focused on continuing to improve this positive trajectory, especially while onboarding many new senior go-to-market resources that have proven track records. They've built relationships and trust with large enterprise, large vendors out there all around the world.
The drive that I'm seeing in the market to consolidate vendors to fewer, more capable platform players, combined with the vast lineup of innovation that we have here at Cloudflare, leaves me confident that we can ramp AE productivity up for a very long time. Now, going back to these 400-plus meetings that I've had over the past 13 months, interestingly, many of these CIOs, CISOs, and CTOs, they had little to no contact with Cloudflare. I was often the first executive that they met with. Our success with these large customers was often done at the project level, maybe a WAF deal that we worked on. We worked with a senior manager. Her job was to land this project on time, in scope, on budget. It's not necessarily the strategic level. That has changed a lot.
Just on the go-to-market team, we brought in stage-appropriate talent to build relationships and earn that really important trust. In 2024 alone, we hired over 250 AEs, but we also managed out about 100 of them, people that just did not have the right competencies or stage experiences at the next few legs of our journey. Really important for us to have the very best team out there. We expect that the capacity of these ramped AEs will accelerate each quarter in 2025. Second point, Gartner predicts that enterprise IT investment trends will heavily focus on AI, particularly generative AI and agentic AI, with a strong emphasis on data-driven decision-making, cybersecurity, cloud infrastructure modernization, and platform engineering to unify and automate operations across many different environments. Now, by understanding these digital imperatives, our AEs can deliver solutions or sales play combinations against them.
In other words, sell a combination of individual products to provide a unified solution for customers. I say this to customers all the time. For Cloudflare, the network is the platform. Third, we launched this developer speedboat a few months ago. Instead of hearing from me, I'd like to invite Ali Cabral, who heads up our developer go-to-market speedboat team, to walk you through what we're doing here.
Thank you, Mark. If I look familiar to any of you, it's because I was on this stage two years ago talking about our product strategy for the developer platform as a VP of Product for that business. Now I'm back to manage the developer go-to-market team for those same products. Taking all of that deep, entrenched knowledge on our products and customers from four years of running that team, now applying that to our go-to-market strategy.
Now, look, we built this developer platform under constraints. For Act One, we had extreme needs around scale, reliability, performance that we could not compromise. Two, our Act Two business needed to deliver quickly a best-in-class product to meet an evolving security landscape. We needed that developer velocity. These constraints really powered our technology strategy for the Act Three business. I am excited about how that technology strategy is meeting the moment in this market that we are all seeing us in, right? There is a huge industry shift underneath us, and that industry shift is AI. We really were truly built for this moment. Okay. When I was in Rita's shoes, we had built this exciting technology, but we were in search of a big, motivating shift in the way developers were building and what they were building. In other words, we were looking for that killer app.
AI is that moment. Our compounding effect of all of our businesses amplifies the opportunity we have in front of us. Nothing is more motivating to me than capturing that opportunity. How are we going to capitalize on this moment? We've shifted our internal teams to respond and adjust to the market movements. To grow our developer business, Mark made the decision that the time is now to build this speedboat around the developer go-to-market function that combines sales, marketing, developer relations, our growth program, and our startup program, all empowered to maniacally focus on that developer persona. We learn quickly. We adjust quickly. We operate independently while building bridges to the rest of the business. This speedboat bridges the gap between our product-led growth products and our strategic sales team.
Often, for the developer business, that pipeline, that qualified pipeline, lies in early interest and experimental usage that often can be missed if you're not focused on it maniacally. Our developer go-to-market team will amplify that early interest and that early consumption while seamlessly converting that interest into strategic large conversations. To expand the strategic focus in this area, we're going to have a list of targeted accounts that we focus on that really, really are designed to be maniacally focused on that Workers opportunity. We'll build deep, entrenched partnerships with them by doing unscalable things, like mutual roadmap development. We'll do specialized hackathons. We'll do focused events together. This is going to allow us to build a developer business that's truly grounded in our customers' needs and allow us to refine a repeatable approach to large and strategic deals.
My team owns our developer go-to-market team, our talented go-to-market team, digestible playbooks and snackable talking points. We are ready to take all of the organic success we have already seen in the field and make that repeatable for our teams so that we can repeat those things again and again and again. We have the data. We have the products. We have the team. I could not be more excited to be back at Cloudflare to accelerate this growth and the developer platform and take it to new heights. With that, I am going to pass it back to Mark.
All right. It is so great to have you back. She has been shot out of a cannon since she got here. We had our three regional APAC, EMEA, and America sales kickoffs over the last month and a half. I think she was the top presenter there.
I also want to talk a bit about partners. They're playing an increasingly crucial role in Cloudflare's growth strategy. We can't build a better internet by ourselves. We've got to do it with partners, both technology partners and with distribution partners. These are important routes to market that we're big-time investing in in FY25. The ecosystem supports all of our customers, regardless of how they want to purchase or consume our innovation.
For example, solution providers who focus on resell and potentially wrap services around it, systems integrators very much focused on the delivery and ongoing implementation and ongoing management, managed service providers who offer Cloudflare technologies as services and implementation services, service providers who embed our offerings into their practice, and then technology partners where we can offer our customer solutions to enhance their tech stack, offering a better together alternative where the integration is slick and modern, and distributors, people that can provide scale for us, that can bring in their group of resellers to go out and represent Cloudflare. We need more feet on the street that have subject matter expertise, but they also have, in many cases, decades of success with customers and do a lot of business with them. Tom Evans, I hired him as our Chief Partner Officer almost a year ago.
We've met with well over 100 principals at many of the biggest partners that I had at Palo Alto Networks. When I joined Palo Alto Networks in the end of 2011, I went to a lot of my F5 resellers and said, "Hey, man, we're building a better mousetrap for security. We want to get you in on it." The interest was high, but there was a lot of fear from the mostly American company or American resellers that were afraid of what Cisco might do to them. All the European SIs and resellers were afraid of what Checkpoint might do to them. When I talked to them about Cloudflare this time, I called every CEO of every partner I've worked with and said, "Hey, man, not only is this a better mousetrap, this is the future." I got zero resistance and very high level of interest.
Wrapping up, I think we have a lot of opportunity ahead here. You've heard from Matthew, CJ, Rita, and myself that we all believe we're the right company at the right time to get customers to the next level. We came here because we love our mission. We love the business that Thomas and Matthew and Michelle and others have built. Being able to partner with world-class people like Steph from Goldman Sachs and my brother from Another Mother, CJ, on the product side, he is the best product leader I've ever known. I felt that way for a very long time. He's probably the most polished customer-facing product leader that I've worked with. I really think that as a team, we can do some incredible things together. In 2025, the bottom line is we will be world-class in the go-to-market organization.
With that, I'm going to pass it over to our CFO, Thomas.
Thank you, Mark. I'm Thomas Seifert. I'm the CFO. Welcome here. Hello, everyone. You might remember last year I had to hurry out to catch a wedding. While we have a wedding this year again, we did a better job planning for it. You have my undivided attention after this speech and Q&A too. It was interesting when all of you came in and an astonishing amount of you asked me, "How are you, Thomas?" I understand by now this is more than small talk. I know that is one of your KPIs, figuring out what the sentiment of the CFO is and is he optimistic enough. It made me think of yesterday. We were all sitting here going through a little dry run.
I was there in the corner, really proud of the team, how everything came together. It made me think of my first investor day. I looked at my script and it said, "Pause here, Thomas. Pause here. Pause here." I looked at my CEO and he had smile, smile, smile. I said to our head of IR, "Why does he get the smileys?" She said, "Because you're the CFO." This year, I think this script would have deserved a couple of smileys. I think we are in a year of inflection points. You saw this already in all the presentations before us where things fall in place and come together and why we are excited about what is ahead of us, why you hear us all talk about an acceleration that we see.
Before we go into this and to the meat of my presentation, I like the part where Mark went into some teaching exercise. I thought we got so many questions on pool of funds. I wanted to spend some minutes on what is the rationale behind them of the pool of funds and why do they impact our finances the way they do. Before we go to the really exciting part, a little bit of financial accounting 101. Quick summaries. They are good. Pool of funds are good. They allow us to accelerate adoption of our customers. Like so many good things in life, they come at a price. Consumption-based nature of these contracts can impact, unfortunately, the shape of our financial reporting. How do they do that? Enterprises are increasingly accustomed to purchasing cloud services through flexible consumption methods.
As Mark is successful and his team and as we migrate up the enterprise stack and our customers and our opportunities become bigger, this is how they want to purchase. As you can see from the graph on the right, we had a significant adoption of pool of fund deals over the last five to six quarters. However, with this increased adoption, the noise also increased compared to a purely readable model. I picked two cases, I think, that would demonstrate this in a meaningful way. This is one example of an existing customer, very large customer, very large technology company who had been on our platform for many, many years and had adopted quite nicely. In the first quarter of last year, they adopted a pool of funds contract, a significantly larger commitment on an annual basis.
Because of the significant commitment, of course, they are shifting down the pricing curve a little bit. When they started to transition in the first quarter, revenue dropped because they were on a lower pricing curve. As adoption then picked up again, revenue grew. The customer started using more of our services. Revenues accelerated and even, of course, exceeded our pre-pool of funds runways. This was an existing customer that switched from a readable contract into such a structure. We have other cases too. This is an example of a really fast-growing AI company. Signed on with us early, very small deal, $500,000. After two months, already figured out that their need on our platform would be significantly bigger. Signed a north of $7 million pool of funds deal for a year.
The service took a while to ramp, and we only recognize revenue when the service ramps. That is something we see more and more often. Our deals are becoming bigger and bigger. We are addressing more and more traditional enterprises. They are migrating off heavily installed and intertwined incumbents. From the signature to the start of the service takes a while, has impact on our revenue. Just two examples. You can see that pool of fund contracts are effectively ramp deals in terms of revenue recognition. Also, they start lower in the beginning and in the near term compared to the readable structures that you might be used to. In the end, they lead to higher revenue growth over time, and they lead to accelerated adoption.
We think the negative headwinds that we've been seeing from pool of funds deals—well, you see this already on the right side of the graph—are disappearing. We think by the second half of this year, 2025, the negative impact will have faded away. Okay? So much to the 101. Let's talk about why we are excited, how things are coming together. With that said, I want to frame the rest of my conversation here primarily around two things: how we look at growth in the near term and moving forward, what drives the acceleration, what are really the signals we've seen in detail when we were in our last earnings call, and then how does this relate to profitability? How are we going to manage the company moving forward from a profitability perspective? Starting first with growth.
I just said in the beginning, this seems to be the year where things are coming together, where technology and platform are accelerating, where an enterprise-ready team should take on the excitement that we see. As I mentioned on the fourth quarter earnings call, we have real visibility into the factors that are driving that acceleration. I want to take literally every one of these points on the slide and give you a bit more insight into the data that drives our confidence to accelerate and also to invest behind the data we are currently seeing. As you heard just from Mark, we are super encouraged by the improvement he and his team are making around productivity on the sales side. This continues now for a couple of quarters.
Not only is the productivity increasing, we see a significant shift in the attainment of the ramped AEs. Productivity and attainment are coming together. The whole curve has shifted to the right and has improved by over 10 percentage points for AEs that achieve north of 80% of their quota. First data point, super good progress on productivity. Attainment curve is shifting to the right. What we also see is all the efforts that you heard Mark, CJ, Ali, Rita talk about on the product side are generating momentum, significant momentum from a pipeline creation perspective. Our expanding product portfolio generates the customer value. You heard Mark talk about this. We literally exited 2024 with a carry-forward pipeline that was 40% over the previous year. Higher productivity, higher attainment being set by a stronger pipeline year over year.
As Mark mentioned, not only have we made significant progress from a productivity perspective, but also upleveling our go-to-market systems, the processes that go along with it, the folks that we put in place. All this action being in place now, the amount of AEs that are on enterprise and in seats is encouraging us. This will now allow us not only to feed revenue growth productivity driven like what we did last year. This year will be the year where we can accelerate hiring and net capacity. Sales capacity is increasing. Remember Mark's chic formula. This will drive ACV acceleration moving forward. We continue to accelerate AE ramped AE hirings throughout this year. The factor that I like a lot, and you hear Matthew talk about this when he says we stack S curves on the product side on top of each other.
On the customer side, we see something similar happen. We are stacking customer cohorts of larger and larger customers on top of each other. We have and we continue to have incredible success in the small and medium-sized businesses that we serve coming from our free origin. We are stacking now larger and larger customers on top of that. There are very few tech companies in the world that serve such a broad continuum. That was one of the bear cases always against us. I think we put this to rest, if not last year, then this year. The amount of $1 million customers that we have, we seem to be in an inflection point. The dot is sticking up. We came to the revenue run rate we have with significantly less $1 million customers than our peers.
Only 1% of our total paying customers are represented by them. It is an incredible white space in terms of opportunity in front of us with the work Mark has done, with the network Mark brings, with the network CJ brings to us and the team he hired. We think we are on an inflection point in terms of how the curve is going to behave moving forward. Why do we think that? Because we see that accelerating across all of our large customer cohorts, not only in number in terms of numbers of customers, but also in dollars spent on Cloudflare. Our largest and biggest customers are our fastest growing cohorts. The north of $5 million customer cohort is now up 31 times since 2018. We announced our largest deal ever in the fourth quarter last year.
Normally, it takes us about 12-15 months to jump to the next large customer cohort. I'm quite convinced that this will be the shortest lifetime of a large customer in our portfolio. Stay tuned. I think even the rate at which we jump to the larger cohorts is accelerating from what we see today in our pipeline. CJ did a really good job of the lessons he learned at ServiceNow, how important the land and expand is. Our platform is designed with this inherent easy use. Once you are on the platform, once you have your first product, every additional product is literally only a mouse click away. You see that revenue contribution from the number of products used, especially north of 10, is accelerating.
Our success in driving the platform adoption is evident by also the consistent increase in dollars spent with Cloudflare by customers and how many products they consume from us. This is, I think, the chart that literally shows how things are coming together. I remember when with some of you, we did the teach-in for the IPO, and we talked about five revenue-generating products, wave one, mostly small and medium-sized business. At that time, our enterprise share was less than 10%. Now, 60 revenue-generating products across three acts feeding off each other in terms of momentum. A massive increase in our TAM, six times over the last seven years. I think this describes the opportunity. Traffic has been exploding. We have architected a network and an architecture to take full advantage of it. You have workloads in all places.
You have workloads in all platforms that need to talk to each other, all under an umbrella of security and privacy. This gives us this massive TAM. And all of the goodness that Rita explained to us is not even reflected in those numbers. One of the places where I should have a smiley in my script. Finally, maybe a chart that you all have been waiting for. One of the big cases was also when you transition from one act to another, there will be pauses in terms of how you generate revenue. I think we can put this to rest now. We see accelerated traction in all the key areas that drive wave two and act two and act three, including Workers and SASE. In 2024, ACV bookings for Cloudflare One and developer grew by 43% year- over- year.
Accelerated growth rates coming compared to 2023. 76% up in developer, also significantly accelerating coming out of 2023. All good news. This is why we think a lot of things come together. That is why we were quite confident when we talked about how an acceleration of product, an enterprise-ready team to take full advantage of it, will lead to an acceleration on our top line moving forward. Let's switch now from growth to profitability on the last couple of slides. We have always taken a really disciplined approach and a very data-driven approach on how we scale Cloudflare, balancing what investments we need for future expansion and with the financial and operational efficiency.
The business model helps us because it allows us to follow the data, not only when we talk about GPU capacity, it is CPU capacity, it is how we expand legal entities, it is how we expanded our market and go-to-market footprint, how we ramped our enterprise sales force. This disciplined approach gives us flexibility to lean in when we see the opportunities. As we entered the year, the data we saw gave us the confidence to do what we talked about, to continue to reinvest and accelerate growth. We are accelerating growth and productivity when combined with the pipeline that I just showed you, which gives us this confidence. As I said many times in the past, we are focused on maximizing long-term value. We made a lot of changes in Cloudflare, including the transformation on Mark's side. It could have been disruptive. It was not.
It allowed us still to exit 2024 and maintain unit economic margin of more than 40% while we transformed and while we are re-accelerating R&D and innovation pipeline moving forward. As you can see on the slide, we perform in all metrics, how we serve, how we deliver our peer group on all KPIs. I remember discussions when we launched Act Two. People said, "It might be CapEx efficient, but it will drive OpEx." We did not. We continued to ramp down or drive down the efficiencies and lower cost of serve quite significantly. We will be focused on this moving forward. We are not satisfied. Michelle was here and said we are just getting started. We will remain focused to become even more efficient, more productive. Every dollar we want to put into Cloudflare needs to deliver more revenue and profits and show a healthier value.
I think we are a good path on it. The network, the processes are instrumented in a way that give us the levers to influence how our cost to book, our cost to serve, and where we probably have still the biggest room for improvement. While we are good compared to our peers, still the biggest room for improvement in getting better on expansion and driving DNR up. This brings it all together. Our North Star has been the rule of 40 plus. This will continue to stay in our North Star. The compelling unit economics that you saw allow us to support the business while we drive efficiencies forward. The gross margin, I think, is a good testament to it. Our range, our target range is 75%-77%.
We've been way better despite all the re-architecting, all the isolates, the investment in CPUs and new server generations, the amount of GPUs we have put in place. I think Matthew has done a really, really good job trying to help you understand why we are highly convinced that the metrics will hold up from a CapEx efficiency moving forward. We've made significant progress on the sales and marketing side in terms of leverage, despite shifting the footprint to enterprise, despite all the transformation we've done. We are light today compared to our business model in terms of R&D spend. We want bigger development capacity, not necessarily at ramping up dollars only. We are in the process of setting up legal entities in places that allow us to increase R&D capacity significantly while we live in the spend envelope we have.
On the G&A side, as we said, we're getting really, really close to our target range. Our long-term operating model is holding up nicely. We are very much convinced that 20%+ is very achievable in this setup. Despite everything that is in front of us, we've been improving our free cash flow margin significantly. We think 25% as a long-term target is, after all, a really good target to have. With that being said, and before I hand back to Phil, I have to put on one more big smile. Despite the fact that there's a pause in my scripts, thank you for listening. I hand it back to Phil. Thank you.
All right. That was fantastic. We'll take questions from the audience now. If you'd raise your hand, then Heather and Alexei will bring over a mic.
Why don't we start with Joel? Joel first. Thank you. Joel also had nice things to say about Colorado when he walked in. He had said first question. He asked, "How are you?" Exactly.
Joel Fishbein from Truist. Mark, I'm going to pick on you. It's a multi-part question. Since Cash isn't here, I'll make it complicated. When you talked about consolidation, vendor consolidation, and go-to-market there, are we past the education phase that Cloudflare is a vendor that can take out the network stack and some of the other DNS layers? Are companies still looking at incumbent vendors to do that as well, number one? The second thing is, how do you balance between that part of the go-to-market sales and the greenfield opportunity around AI? How do you have your sales force going after those two opportunities at the same time?
Yeah. Yeah. Great question, Joel. Thanks. A couple of things. I think there's two pretty massive colliding forcing functions that are facing every IT and infosec leader out there. I talked about this need to consolidate and manage expense. The second one is, what is AI going to do to my business, and how will that profoundly change everything? Those are really top of mind for everyone. Are we there yet? No, I don't think we're there yet. I think we're still building out capacity. We've hired a lot of really good people. We've really upticked our enablement activities to make sure we're teaching our people how to sell business outcomes to customers, and especially to some of these traditional companies that maybe haven't naturally come to Cloudflare in the past. I think it's a work in progress. I think we're well underway.
I think we're a lot farther ahead today than we were a year ago. I think this takes time. I think every customer I've met with, every head in the room nods when I talk about these forcing functions. Oftentimes, they're learning about things that we do that they didn't know. I think that's another reason why partners are really important to us. It gives us a lot more feet on the street with existing relationships. This is globally. Definitely work in progress, but I'm delighted about where we are today. I mean, I've gone fast. I talked about the 250 AEs that I hired. There's a multiple of that of people, especially after the sale, that are going to deliver the white glove service that a Truist would require to migrate from legacy technology to more modern architecture.
That ramp has been substantial. I think on the AI front, that was one of the primary reasons I wanted to build a dedicated, focused speedboat. The chance that I could get Ali Cabral to come back and run that for us was far too good to pass up. Like I said, she's been shot out of a cannon since she joined here 60 days ago and really driving up the attention. It was a big part of our training at sales kickoffs over the last five, six weeks. The inbound interest from the field into her team is off the charts. I feel really good about where we are.
Great. Thanks.
Yeah. Great. Andy Nowinski. Joel. Hey, Joel, can you pass the mic this way? Oh, can I say one more thing? Sorry, Joel.
It's interesting.
In these meetings that I have, especially if I know the customers, I've been doing this for a long time, right? I will almost virtually poke them and say, "Who are you going to pick for your partner for this coming wave of AI? Who are you going to pick? You're going to have to pick somebody that looks a lot like Cloudflare." Especially if I know the people well, they'll chuckle and they'll go, "Yeah, you're right."
All right. Thank you. Andy Nowinski, Wells Fargo. Thanks for hosting it. Great event so far and a great location. The theme of the day was accelerating. A few years ago, you put out a $5 billion target for the end of 2027. Do you have any updated thoughts on that goal?
Are there any new contributors to that $5 billion that maybe you did not think about when you put that out originally?
Yeah. We put that out in 2022. A lot of things have changed. 2023, 2024, IT spending came down. Also, Cloudflare changed a lot. We made a lot of changes. You saw that clearly today. To be honest, I think the net impact is probably less than we all thought. Maybe a year later is probably where we stand. Just to frame this, maybe if you annualize consensus for the fourth quarter of this year and say $5 billion in 2028, this would be a CAGR of 28-29%. I think where we are today is this is a very reasonable assumption.
I think Mark and CJ, and Mark can put pressure on them, are significantly more ambitious and hopeful that we can get there faster. I think that CAGR would be a very reasonable assumption at this point in time. Okay. Thank you. Jonathan? Andy, do you have the mic still? I have it. Yeah. Jonathan.
This question is for CJ Desai. This is Jonathan Ho from William Blair. With 60-plus products, how do you balance between focus and broad coverage by your platform?
Thank you. First of all, innovation and laying that out sometimes in front of the customer is a very positive thing. When you say, "Okay, these are the specific use cases that this particular product can solve the problem," it's always a good thing.
However, to make it easier from a go-to-market perspective for Mark and the team, we are working on what is the right way to not only package the products, say, for security for external users or for employees and contractors and so on. One is having the right packaging and the right pricing is something work that is going on right now where we are making progress so that we simplify both the message that we give to our customers and partners on how they should position Cloudflare and make it easier for a customer to say, "At the end of the day, this is security for all your external users, as in your end customers or end partners, and these are the products that does it." Thank you.
Shaul?
Thank you so much. Shaul with TD Cowan. Actually, question for Matthew.
Over the course of the past two years, in some quarters, you have become maybe unofficially the harbinger of economic direction. You recall 2022, the crypto, 2023. Actually, 2024 seems to be getting a little better. My question to you, what's your current thought? I think also in terms of federal spending, DOGE, etc. Thank you.
Oftentimes, we see what macro trends are because so much of the internet uses us. I guess at some level, the Trump campaign and properties have been customers, but that I do not think gives us a very unique insight into what's going to happen with tariff policy tomorrow, which seems like what is driving the macro right now. I feel like my crystal ball, which really is some insight today probably into what's going on in the Trump White House more than anything else, is pretty cloudy.
I don't know that I can tell you what's going to happen next because I don't think we have a particularly unique insight into it. What I will say is a couple of things. First, regardless of what happens, we're going to be great. We know that we can execute. We know that our products are must-haves. We know that we're able to deliver on that. In particular, with DOGE, there may be some unique advantages that we have. If you look at DOGE in particular, all of the public-facing bits are built on Cloudflare. A lot of what are largely young kids that are running it have been Cloudflare fans for quite some time.
In fact, there was a story in New York Magazine about one of the DOGE members who happens to go by the nickname Big Balls who, it turns out, four years ago wrote to me saying, "Huge fan of Cloudflare, but I think I can disrupt this. Here's my business plan." I wrote back something like, "That's the dumbest idea I've ever heard. Why would you waste your time?" He, strangely, over the next four years, continued to write emails, including, as recently as about two months ago, wrote an incredibly thoughtful set of suggestions for our Workers platform, which I forwarded to CJ. He's a big Cloudflare fan. I think that there are a lot of federal contracts that legacy vendors we would think are overcharging the government for. I think that there's an opportunity for us to potentially win business there.
I also think this is an example of how having a product that a 15-year-old kid can play with and experiment with usually, it's not the 15-year-old kid who ends up running the federal government. In this case, it seems that way. Oftentimes, we find ways that people who start experimenting with us early bring us to work, grow their careers around us. That has been just a very natural way that we go to market and that we build fans. I think that historically, the practitioners have been huge Cloudflare fans, and we haven't had as much of the relationship with the C-level executives. Over time, those practitioners become the C-level executives, and we've brought in folks like Mark and CJ and Stephanie who have the relationships with the C-level executives. We are going to meet in the middle.
I think that creates an enormous opportunity regardless of what happens in the macro.
Hey, guys. Adam Borg with Stifel. Thanks so much for the time today and taking the question. Maybe for both Mark and CJ, just as we think about Cloudflare One and Zero Trust more broadly, talk a little bit about how you feel where we are in the go-to-market motion, specifically where we are on the R&D side, what still needs to happen to maybe win at the uppermost enterprise level on both sides. Thanks so much.
Stephanie, you want to start with the product?
Sure. First of all, Zero Trust, massive space. TAM is massive and still growing. Fundamentally, it's a good place to have great products. That's number one.
From an R&D perspective, we have a phenomenal team, great leaders, Anika, who is here, and a few others who run that team. We are very focused on making sure that where Cloudflare's strengths are, which is around access, gateway, we go really, really deep and become world-class to get hundreds of millions of users on it. That is where the focus is. The team is already delivering great innovations this quarter, next quarter, and so on as we penetrate the enterprise and the large enterprise space.
What I hear from customers when I ask them, "Okay, what do you think about this portfolio specifically for Zero Trust and where we stand?" They always say, "Your performance for the end users is so much better than the other players that are out there that you have a chance to win, and you provide a great pricing and value." CJ, we will always have this conversation. Literally, I had a conversation yesterday with a Fortune 35, who said they are going to kick off a project to evaluate us as well, and also a large bank in Japan. Very, very optimistic on where the product direction is and specifically going deep where it's our strength, which is around network and, of course, access and gateway.
Yeah.
On the go-to-market side, we spent a lot of time in the last year making sure that all of our sellers, all of our solution architects really become very competent in being able to articulate why security controls belong over a highly performant, highly secure network like ours, right? You have heard Matthew in the last couple of earnings calls talk about some pretty large Zero Trust or SASE wins on the earnings calls. It continues to be a work in progress. I think the products, from my perspective, are terrific. Our sellers are getting more and more capable of articulating that. I think just my earlier point about this drive for consolidation, I spent almost eight years acquiring companies at Palo Alto trying to chase the security controls that were moving off of people's data centers and real estate.
That is why I really believe the market is coming to Cloudflare because they belong over an edge network like ours.
Yep. Gabriella.
We'll sneak through here. Thank you so much. Two quick ones. One is a clarification. You mentioned doubling the CapEx, doubling the compute capacity in the network this year. You also said you normally invest CapEx behind demand. Inherently, does that mean that demand signals are doubling demand this year? Maybe just clarify that for us. The follow-up is on isolates. Maybe just a little bit more about how that technology has evolved over the last four years. I think it was 2021 when you first talked about it coming out of the remote browser isolation category. Would love to hear a little bit about what is unique versus serverless. Thank you.
Yeah.
I mean, I think it is fair to say that we see traffic and demand growing, and we're investing behind that. That is roughly what we see today. If that grows faster, then we'll invest faster. Because we can invest behind the demand, because we can invest behind the revenue, that means that CapEx as a percentage of revenue should hold fairly consistent. In terms of isolates, again, I think it's the natural evolution. We used to have bare metal servers, then we had VMs, then we had containers. Isolates are the next stage. I think there were, much like when VMs and containers came out, there were concerns around, "Is this actually secure?
Is this actually something that works?" We have done the hard work to be able to validate that, work with some of the top research labs in the world to be able to validate that. Being able to have what is just the lightest weight underlying infrastructure is what gives us the ability to put code all around the world. I was actually with the person who for a long time ran AWS Lambda, which is kind of the equivalent product, although it is actually much more of a container-based product at AWS. He is at a different employer now, and we were seated next to each other at a dinner.
He said, "One, you guys just beat us in this space, and you've run so much further ahead that we sort of decided that we were going to sort of double down on our previous strategy, not try and chase you because they couldn't catch up." Two, which was both exciting but also somewhat concerning, he said, "Inside of AWS, we think of Cloudflare as our top competitor." Madeline?
Oh, Matt?
Hi, Madeline Brooks with Bank of America. Thanks for taking the question. It is really for Matt and for Mark. Just to follow up earlier on macro and outlook, just talking about enterprise, I mean, I think the last couple of weeks have been really volatile outside of just government, but for everyone here. When you talk with enterprise customers, are they starting to feel the uncertainty that's ahead of us?
Is that giving them any pause or hesitation in moving forward with any type of purchasing decisions? Thank you.
I'll start, and then Mark can add. I think we're not seeing it show up in the conversations that we're having with salespeople, with our customers right now. They're continuing to invest. The balance is, again, there is a fundamental shift where, yeah, we're still working on going to the cloud. We hear from every enterprise. Security is still top of mind, but now I've got to work AI into that as well. The fact that we can engage very much in all three of those interests and concerns means that we haven't seen any slowdown in taking meetings. We haven't seen any slowdown in pipeline. All of those things continue to be positive.
That said, I mean, obviously, as markets are volatile, that can have an impact, but it's not something that's showing up at this point. Maybe, yeah,
I think the longer-term trends around consolidation and embracing more modern architecture to exploit the benefits of AI and all that that means, I think those are the things that people are more focused on than sort of one, two-month kind of horizons. I don't know, CJ, you talked to a lot of customers as well.
Same. I would just echo it has not shown up, and customers are very focused on whether there is a modernization initiative on the network side or the cyber side, but it has not come up.
Maybe one follow-up, if I could, in terms of putting any of the risk or uncertainty in the future into that five-lane guide that's now 2028.
Kind of what were the puts and takes for moving it to 2028? Was it really more just previously things that Cloudflare has experienced, or is there a level of uncertainty going forward that you also took into account?
To a large extent, it's just math, right? If you put it into 2028 and we get to the consensus run rate in the fourth quarter, 28%-29% CAGR to get to $5 billion in 2028 seems like a very reasonable assumption at this point in time. With ambition to do better, of course. I think this was clearly obvious in the presentations you saw, but we are where we are today. We were in an earnings call where even a slight acceleration were, "Are you sure?" From where we are today, that is a very reasonable assumption. Mark.
JP Morgan. Is this on?
Fantastic presentation. Wondering if you can speak to the GPU fleet and the nature of it, the direction it's going to take. Just specifically on two topics. One is, do you want to be able to handle super complex, multi-step reasoning, the type of thing we see when we do deep research, which is pretty amazing? It'll work on something for 15 to 20 minutes. It'll write Python scripts for you to answer a question. Do you want to architect to that, or do you want to architect to sort of smaller language model types of use cases? The other concept that's coming up is the intertwining of the training and the inference, right, where you have runtime training.
If that is a trend that you think is emerging, do you want to architect to that where then you might want to take on some of the training workloads?
Yeah. First of all, I think we do want to continue to innovate to be able to support as many inference tasks as possible. As we look out at where demand is right now in the market, it's actually, again, there are the OpenAIs of the world. There are a handful of large companies that are building kind of these frontier models, but that isn't what the average company is necessarily building for. We have relationships to those companies. I think you'll see them use our inference platform more and more. I think we will have to invent tech in order to make that work.
I think we have the right team to be able to do that. I think we want to support everything over time. At the same time, there will be just the nature of the beast will be that it might not be at every moment we can support every possible model. I am amazed at the work that our team has done to support the breadth of models, including very large models across our platform. We are inventing the technology to be able to do that. Behind the scenes, we've been an NVIDIA partner for quite some time. They make great silicon. We have deployed that. You will see us have much more diversity of silicon across the board.
I think it's kind of like the old Steve Jobs line, focusing on speeds and feeds, like exactly what chips we're using, exactly what the ratios are, we think is a mistake. We want to abstract that all away from our customers. What we're seeing as we experiment with both the silicon, which is coming out of companies like AMD, Qualcomm, Intel in the GPU space, as well as a number of the startups that are in this space, is that the right answer might be different depending on exactly what the use case and what the model is. It might be that an AMD GPU is better for one model and an NVIDIA GPU is better for another. Behind the scenes in our CPUs, we have Intel CPUs, we have AMD CPUs, we have ARM-based CPUs that are running.
Customers have no earthly idea which of those is actually running behind the scenes. I think that that is what we want to do in the GPU space as well. I think we will become more heterogeneous over time. That will allow us also to drive down what the underlying hardware costs are. What we are seeing is that there are very big advantages that you can get in terms of performance, even on NVIDIA GPUs, if you do not necessarily use CUDA and some of the platforms that are there, if you start to program a layer lower than that. Those are the engineers that we have, the ones that love taking those challenges. I will give you just a really specific example. To load and unload a model using standard kind of APIs on an NVIDIA chip takes about two seconds.
In most cases, who cares? It's two seconds. In our case, though, we want to be able to load and unload models incredibly quickly. Our engineering team said there's no physics reason. The bus supports that. The memory is fast enough. There's no physics reason that that has to be the case. It's just no one is optimized for it. Whereas we've now optimized for that and can load and unload models in orders of magnitude less amounts of time. You can imagine that that then allows us to have much higher utilization rates because I don't have to dedicate a particular chip to a particular model. I can literally load and unload models as they come on board. That's the type of innovation which I think actually creates a real moat for us as we go into the space.
It allows us to get more practical use out of every dollar of CapEx.
In the middle here? A friend from UBS.
Awesome. Thank you for the day. Roger Boyd with UBS. Matthew, you were talking earlier about giving content owners more controls over how their content's accessed by AI. That seems like a massive opportunity. Just how do you think about that playing out, and how do you see that kind of trickle into kind of Act One, Bot Manager? Yeah.
I mean, the first thing is there are a number of media companies that had used other providers that kind of were like, "It's not worth switching." We'd say, "We can do it cheaper or something." This is the first catalyzing event where I think you're going to see a huge amount of media companies switch to Cloudflare from some other providers because no one else is thinking about this and helping them. This is what they think of as the next central problem for them. How it plays out, I think we'll see. I think in the short term, the challenge is the AI companies haven't figured out what their business model is.
Exactly what that looks like, I actually think it's a mistake for content owners to say, "$20 million, but you get access to our entire catalog." I think you want to do something that is more like, "As the AI companies grow, we get to grow with you," and figuring out what that model looks like. I think some of the work that folks like Bill Gross have done is really interesting. I think you'll watch this space because I think there's a lot of things that we can potentially do with them. The steps feel like we're kind of it's a three-step process. Step one is give people the analytics to understand what AI companies are using.
Every publisher that puts content behind Cloudflare and looks at the data, they're like, "Why is ByteDance crawling you that much?" They have never had that visibility. Step two is give people the controls. I think we're kind of in the middle of step two now, and you'll see us in the next few months. We think we've figured out and patented some really interesting technology that can help differentiate between content that publishers would want in AI. For example, Cloudflare, we have a whole bunch of knowledge-based entries on how to write Cloudflare Workers. We want that to be in AI. We want to be able to provide that. On the other hand, if it's a New York Times article, New York Times probably doesn't want that.
We want to make that just auto magic to be able to differentiate between those things. We think we've actually figured out a pretty creative way to do that. In the next little bit, we will default block all AI crawlers, unless a publisher permits them to, on content that we believe that publishers don't want them to have access to. That then creates scarcity. Once you have scarcity, now you can create a market, right? That's step three. I think exactly what that market looks like, I'm not sure. There's probably some fee for crawling, and then there's some fee for using the information to crawl the information in a response. We're going to figure that out.
What I am encouraged by is that the leading AI companies that are the most thoughtful in this space are saying, "This is something that has to happen." If you're a top AI company, one of the challenges is right now there's not a ton of barrier to, I mean, we're seeing this right now where just everybody has, it's just everyone's leapfrogging it every day. Access to content might actually be the barrier to the AI companies that are out there. In my kind of optimistic view, we could actually see a reemergence in really high-quality content that is super valuable. If you're Condé Nast and you're supplying data to an AI chatbot that talks to teenage women about fashion, the fact that you have exclusive access to what happens at the Met Gala is really interesting content.
You can imagine that someone sitting in Indianapolis who has a fleet of drones that reports on what traffic patterns are on I-70, all of a sudden, that becomes incredibly valuable. That content creator could make millions of dollars a year off of actually creating that content in a way that, I mean, that content has to be in an AI system if you've got that. I think there's actually a, there's going to be a new business model for the next version of the web. I don't think there's any company that's better positioned to help figure that out than Cloudflare is today. It's like Google figured it out for the last version of the web. I think we hope that Cloudflare is the one that helps figure it out for the next.
All right. That'll actually be the last question of the day. The day went fast. Got it.
Very fast.
Thank you, everyone. Thank you, everyone on the live webcast for joining us. For those in the room, stick around for just a moment while the webcast ends. Thank you.