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

All right. We are going to go ahead and kick off the Cloudflare session at day 2, Goldman Sachs Communacopia and Technology Conference. Thanks, everyone, for joining us. And thank you to Thomas, CFO. It's very good to have you back.

Pleasure being here.

So Thomas, I know from the outside, we can see the metrics this year. And what's really interesting about this year is that you're actually poised to accelerate as we go through the year, probably consistent with what you said a year ago. And so putting some of the sales productivity conversation to the site just for now. What stands out to you about this year? What are your observations on what's driving or what's going better and has more momentum this year versus last year?

I think what has always made this business model at Cloudflare special was that growth was driven by many vectors, many independent vectors. And I think we see a lot of goodness coming together in the first half of this year that is driving that. So if you take aside the sales productivity gains, the contributions from adding net sales capacity, it fall in a couple of buckets. We see extraordinary strength in our largest customer cohort. So customers north of $1 million and north of $5 million of ACV are growing the fastest. We see strength across pretty much the complete product portfolio. Well, I'm sure we will talk about Act 3, Act 2, our Zero Trust, our Cloudflare One products are ramping very, very nicely. And we see a reemergence, so to speak, of Act 1, mainly driven by all the work and announcements we have been doing around what we call Act 4, pay per crawl, I'm sure we'll come to that. So it's a broad acceleration in -- across many different vectors that is making that happen. And we are, of course, pleased by the strength we see in the first half, and we guided for continued growth in the second half.

Yes, fantastic. Well, let's jump into Act 3 because there are so many products we can talk about. So Cloudflare talked about how your share within AI-native companies is higher than your share across the ecosystem as a whole. Maybe share with us your observations on that AI native cohort. How are you seeing your footprint trend within them? And how does that cohort compare to your classic cohort?

I think you have to spend some time talking about Workers first. So the idea was to give a product to developers that would be without lots of competition in terms of performance, in terms of latencies, in terms of ability to scale. And it is the basis of everything that Cloudflare is built on. Workers enables us to scale and have the pace of innovation we have. So having now a product platform in place that is rather complete as a platform. It offers unique interfaces. It has storage capability, it has database, it has CPU, it has GPU, enables now a world of -- a class of developers to develop products and us that is quite unique. Our developer count now is north of 3 million developers. It has been accelerating since the second half of last year based on all the AI work. More interestingly enough, we see we are now at a point where because of the latency and the performance differential in the platform, you see non-native -- non-digital native companies build on us. We just -- the other day within the company talked about an industrial customer that is building an agentic AI workflow on Cloudflare in a vertical that you would not necessarily associate with us first. But it's a highly latency-sensitive application, and we are built for that. So while the initial traction was very much digital native, we see now based on the completeness of the platform in terms of features and performance that it expands into our non-digital native customers. And we think there's a lot more to come if you think about agentic AI workflows in the future, that will be how seamlessly and how latency -- with low latency, you can interact with those agentic AI, so will be key, and we are built for that.

Maybe on this point on features and functionality in Workers, bring us up to speed on how you think about competition because Matthew has consistently been cognizant of the hyperscalers as a competitive cohort. But what about some of the newer companies like Vercel or some of what we're seeing in developer tools with prototyping? How do you think about your structural differentiation versus some of the newer players?

When investors start to work with us and then get the first time to take a closer look at us, I would say you have to start at the architecture of the network. This is the true competitive mode of Cloudflare, this idea of off-the-shelf hardware and the software stack that allows every product on every server and every location completely agnostic to the hardware it runs on. This is what makes this differentiation. And this is a global network. So every code you deploy as a developer, as a customer is instantly available all around the globe. So it's not a geographic architecture like some of the hyperscalers. It is because of how it consolidates so much demand from a bandwidth and traffic perspective, more distributed than anybody else. So in a world where latency and performance becomes key, where we all cannot change the speed of light, so how close you get to the eyeballs is going to be key, this network wins. And I think we see that differentiation play out across all verticals against the hyperscalers, but especially against -- it's hard to build a network like this and be competitive from a latency performance and the reach perspective.

It's interesting that you brought up an industrial company as being a new customer in Workers. I know that one of your areas of focus with Aly Cabral has been how do you expand the developer motion top down and bottom up in some of the enterprise customers that you have today who may already have existing tools in their stack. Maybe give us a little bit of an update on how that effort is going and how you feel about your traction displacing or augmenting some of the existing developer stack?

It's going really well. And if there is -- if you want to nag on us internally, why didn't we come up with the speedboat and the idea earlier, but it's not only about the idea, the product needs to be ready for it and the market needs to be ready for it. So Aly runs the speedboat. She was -- she switched site. She was -- she developed workers on the engineering product side before and now runs what we call a speedboat that is literally looking -- they have a list of ideal customers and applications that can be targeted with Workers and Workers AI applications. And we learned from our large deal that we announced in the first quarter that the best way to win a $100 million customer is to help the customer save $100 million. And because of the architecture of the network, we can do this with significant gross margin. And there are a lot of applications out there of this size where you can help customers save significant spend or significant spend reduction on their hyperscaler bills and still do great things. So they are a dedicated effort. It's a self-contained team, independent of the rest of the go-to-market motion, and they are just working on their target list of customers. And they're making good progress. It's super interesting discussions because we learn a lot also about features that are need to be built out in order to enable that. Lifting an application with this spend is not easy. So how do you enable the customer, how do you take risk out of the migration is key, but good progress. It doesn't mean we will announce 9-figure deal every quarter, but there are a lot of these deals out there.

Well, I won't hold you to announcing a 9-figure deal every quarter, but maybe give us some qualitative commentary on the pipeline. I think there was a comment earlier...

It's a longer list of targets that you would think, and they're chipping away on it. I don't want to give guidance here, but it's a huge significant opportunity set.

There is a nuance here where we've spent time at the Analyst Day, for example, understanding why Cloudflare Workers is better for inference workloads in particular. But it sounds like you're also describing lift and shift for other types of workloads. So what are the types of workloads where you're able to realize this sort of order of magnitude...

There -- the $100 million deal is a good example. It wasn't just Workers AI. It was plain vanilla Workers, where there are many apps out there that can be targeted. It was surprising to us to see how big that universe is. And your persona changes a bit. If you enable a customer or your business partner to go to his or her CFO to save $100 million, that is a lot easier discussion getting big deals through a pipeline. So quite encouraging what we learned. And as I said, really excited about the opportunity set that is there for us.

So part and parcel with the demand signals that you're talking about on the Workers' side comes the CapEx build? So we're seeing a step-up in CapEx this year in the order of magnitude of about 50%. And you've talked about how you wouldn't make that kind of investment if you didn't have the demand signals. What are the demand signals telling you about the durability of that CapEx increase? Should we be thinking about a similar size step-up for next year or help guide our expectations?

So it's a step-up in terms of dollars. It's -- we are still within the CapEx ratio that we guided to. So we talk about 12% to 14% this year, 12% to 13% of CapEx in relation to revenue. We feel quite confident about this. The levers we have to model what we need are quite significant. It's -- Cloudflare has always been a business where we invest behind the demand signal. We never invest ahead of it. It's a playbook that served us really well on the CPU side. We, on average, run slightly north of 80% CPU utilization. And when we started Cloudflare, it was pretty much an Intel-only CPU environment, and then it became flavors of Intel CPUs and then became AMD CPUs and they become ARM CPUs. And today, that software stack that runs on it is completely agnostic to the hardware that it is running on. We deploy the same playbook on the GPU side. We started off with one flavor and one company, and now it's diversifying into many flavors from one company, and now it's diversifying into many flavors for many GPU suppliers. Our big levers are getting the hardware stack right. Inference task is not like inference tasks. The need for what GPU compute capability you need differs. So you have to optimize your hardware stack that is allowing us to be efficient. Getting utilization up is really important. We are today better than anybody in the world when it comes to GPU utilization because the architecture, how we run our workloads on isolates, no wait time, no spin-up time is super-efficient. But we still have tens of percentage points of room to improve to our CPU performance. So that is impacting it. And then there is how much capacity do we deploy? And what helps us on the CPU side is what we do on the GPU side. We have a baseline of capacity that we feel comfortable with what we understand in terms of customer commitment. And then we have certain arrangements in place that allow us to outsource spikes in GPU capacity to third-party vendors. That might -- it's CapEx efficient. It's a bit more expensive. But as soon as we get more comfortable with our baseline, we increase that and we in-source. But that means always we are, to a certain degree, under provisioned in terms of the capacity we have visibility to. And that allows us to run our CapEx investment highly efficient. We think there's plenty of room to stay in this range. The other thing that helps us is you really have to understand how the network works. So we -- Cloudflare is built on Act 1 products, product like firewall routing, DDoS mitigation, where you move traffic to the eyeballs. Our second generation of products, what we call Act 2 is all the Zero Trust and Cloudflare One products. They pretty much transport data back from the eyeballs to wherever the data center is. And they take advantage of an infrastructure that is already paid for by the other products. So every dollar -- revenue dollar in Zero Trust is literally at margins that are north of 90% for us, and it literally comes at 0 CapEx. So every dollar we have opens up allocation degrees of freedom for us to move CapEx spend to GPUs. So many vectors that you can influence many that you can work on, and that makes us highly comfortable that we are going to stay CapEx efficient for a while.

Yes. There's a question buried in here on unit economics a little bit because I think there are some question marks for the industry as a whole as to what the unit economics look like on GPU heavy investments. So I guess 2 parts for you. It sounds like you're already comfortable that you're benchmarking better than some of your peers on GPU utilization. Maybe just share with us how you think about that benchmarking. And then the second part is, how would you compare and contrast the unit economics of the GPU-based workloads versus the CPU-based workloads?

So on the infrastructure layer, margins are lower than on the application layer for sure. But as we have seen in the Googles and Amazons of this world, there are huge benefits from an economies of scale perspective when it comes to leveraging OpEx, when it comes to leveraging R&D spend and others. So we think there's plenty of room for us to stay in the band of unit economics we need. And the second is we are not a point solution provider. We play across a wide variety of products. Today, 64, I think, is the latest count of revenue-generating products with very different unit economics and especially the unit economics on our strongest growing product is providing us all the degrees of freedom we need in order to lean in when it comes to Act 3 and Act 4. That's why despite all the growth you have seen over the last years, all the product launches, the different apps, the explosion in traffic there, the gross margin has been rather stable within 2 or 3 percentage points over the last 9 years.

Absolutely. Well, let's talk about Act 4. The industry moves so quickly when it comes to working through some of these new developments. So tell us a little bit about your observations on how the conversations are going between the independent publishing sites and the LLMs.

Yes. Maybe just for those who are not that familiar with all the Acts we have. Act 4 is a product that we launched in June of this year. It's paid to crawl. It's pretty much a tool that allows the content generators to decide who is allowed to call their websites and under what conditions, for free, paid, paid for how long. And we think this is in a world where the economic structure of the Internet is changing from a click and link-based model to where pages are crawled today for free to a large extent, allows us to help level the playing field. So we announced this in June, all the major publishers around the globe are signing up for it or have signed up as customers. Everybody deals with the product in a different way, but it gives them as an individual company or as a group of companies, a higher bargaining power when it comes to how you monetize content. So that part of the equation is going really well. In parallel to that, we, of course, have discussions with the other side of that equation, that is the companies, especially the large language models. And I think they also understand that their ability to train to a significant extent depends on high-value content. And I think they're ready to monetize for that. And we've seen announcements just in the last couple of days of payments that are made and compensation that is made. They are just nervous about having a level playing field on their side, too, and not one LLM or one company having an unfair or having an advantage at all. So I think conversations on both sides are progressing. They are progressing really well on the content creator side. And we are making good progress. There is a lot of work that still needs to be done to figure that out, lots of interest that need to get aligned. But we think after the first couple of months, the interest it has generated, the momentum we see customers coming to us, and we should talk about that in a minute, shows us that this business model has legs. There are a lot of things that need to get figured out in terms of economics and transaction models, but this business model will have legs.

Maybe let's stay on the business model for a moment. So I fully appreciate that. We're very early in this discussion. Matthew made a comment, I think it was about a month ago on $10 billion being the opportunity associated with, I believe it was ad revenue tied to independent publishing websites. Maybe give us your perspective on how you think about that, and what...

Yes, I'm not sure -- so I didn't hear the $10 billion from him. I mean, if you look at their digital ad volume in the U.S. alone, that is $300 billion a year, right? So the free web is anywhere $30 billion to $50 billion just in the U.S. on that part. So that somehow needs to be replaced or moved from one bucket to the other. So it's a big opportunity. If you just look at the amount of fees that Spotify distributes for music, I think there's a lot of dollars on the table, so to speak, and we'll figure out a business model that takes that into account. It's not an easy feat, but it has legs. That is what we have seen. We have seen so much momentum on publishers normally and content creators was not normally a vertical that was strong on us, right? We never considered us a CDN that moved just data -- a byte of data from point A to point B. This allows us to generate interest in a vertical from a completely different perspective on the fat tail of that distribution, big publishers and content providers. But the interesting part for me as a CFO is Cloudflare came into existence consolidating the long tail of a market where people just didn't have $500,000 to spend on a firewall, right? This is how we consolidated significant traffic on our network. We have an opportunity to do this again in the long tail of the market with content providers and diversified content, high-quality content is an important part of us training and developing LLMs further. So we think we have a very unique opportunity that is -- that comes again to us repeating what we've done in Act 1.

It's actually really interesting because when we started this conversation, I think you alluded to Act 4 impacting Act 1. So are you already seeing a change in your win rate for classic CDN, WAF, DDoS?

We have significant shifts from very large legacy installations that move to us in non-digital native companies. There are large shifts underway, especially in Europe, some big wins that we also talked about in the earnings call. That is one. We talked -- if you listen carefully to our earnings call, we talked about gross margin being impacted by shifts in free traffic to paid traffic. That is one topic we see where customers that are free customers on our website that might have a super interesting, I don't know, travel blog post on Africa, all of a sudden wanting to become a paid customer, so they have an opportunity to monetize content. So you see already shifts in free-to-paid traffic. That reminded us again how important free customers are for our started initiative going into the year with a dedicated team focusing on free customers was a big part of our Act 1 a long time ago. So yes, Act 4 is impacting Act 1 and not only in one part of the market, but across the whole range from free to established Fortune 10 companies.

Yes, really good color there. Okay. So I wanted to ask you one question about SASE. And my favorite data point from 2Q was actually about the customer that essentially said no to Cloudflare 18 months ago and then came back, reevaluated the portfolio and you won that customer. So maybe level set us, how do you feel about being best-in-breed in SASE today? And are you seeing more of those types of deals that perhaps wouldn't have been in your addressable market 18 months ago?

Yes, very much. So you have to keep in mind how we started to build what we call our Cloudflare One Zero Trust product from the ground up, targeting the small and medium-sized customers and then trying to feature complete and go up into the right on a Gartner Quadrant over time. We are at a point now where we have feature parity across most of the SASE components. There are 1 or 2 where there is still room for improvement, especially when it comes to DLP, but in general, you are there. So you can address also large legacy installations. So that is very helpful. We have seen significant traction in the financial service and financial vertical over the last 2 quarters because of that. You know this vertical as good as I do. There are some marquee banks. And if you win them, you get approval across the whole stack, and we've been really successful. This is also to a certain extent, the success of CJ joining us. There are customers who have been successful betting on him in the past. And even if a feature is missing, they will understand if he commits to it, that it will come. So we have made significant progress. What is helping us also in a world of agentic AI that even on the Zero Trust side, latency is becoming a significant decision driver. That is where we excel. It's performance, latency and ease of installment. So the customer you mentioned that came back after 18 months, what we saw was that the competitor was only 30% rolled out. And it was a result of hard to install, hard to get to performance and really bad latency compared to our product. So this coming together now sees momentum in really difficult verticals like critical infrastructure and financial services. So we are quite pleased with the progress we've made.

So I think throughout this conversation, we've been pairing the product excellence with some of the changes you've been making in go-to-market. And what I found really interesting coming into this year was you had already succeeded in making a lot of the block and tackle changes that you wanted to make under Mark Anderson, such that you had a higher number of ramped sales executives at the enterprise level coming into this year. You've also talked about increasing hiring this year now that you're where you want to be from a productivity per rep standpoint. That should set you up really nicely to have more ramped sales reps next year, which then increases the amount of opportunity you can go after. Maybe just level set us how should we think about ramped sales reps impacting reacceleration and the pace of Cloudflare.

They are an important part of that equation. When we went into the sales transformation more than 2 years ago now, we set ourselves certain goals that we wanted to see. This idea at Cloudflare that you never invest ahead of demand is true for everything we do. So we wanted to see certain progress on productivity levers for reps before we get comfortable to open the hiring struggle. And we talked about it on our earnings calls on the third and fourth quarter, we've got comfortable that the productivity was where it needed to be. There's still room for improvement, but they're diminishing returns now in terms of the levels we have achieved. That allowed us to get comfortable to increase hiring. And we said that last year, much of the sales growth was productivity driven. This year, it will switch from productivity-driven sales growth to net sales at capacity. And this is what we see, what we've seen a bit in the first half. We'll see a lot more in the second half. And then every sales rep you add has a ramp-up period, right? So until they come to full productivity. So that impact will spill over into next year, too.

And I'm curious to what extent, I know how thoughtfully you think about AI internal tools for your organization. To what extent are you seeing AI accelerate the productivity ramp for your sales reps?

So we were quite careful in the beginning, how we allow people to interact, what LLMs we use, where the data needs to reside. The first lessons learned were that every time we opened for a productivity tool, the results were significantly more impactful than we thought. So we have a broad initiative across the company. We have a dedicated VP who does even tutorials for every employee on how to use AI to automate his or her workflow. So it's not only limited to go-to-market. We have significant tools now deployed that help us from a productivity perspective in terms of first contact, reducing the numbers of PDRs you need to support an AE. One of my biggest challenges is building a budget for next year that incorporates it all, right? So where can we -- where can -- and I'm sure I'm not alone, I must be -- every other CFO must think about that, that the benchmarks of yesterday are going to be disrupted. What used to be, I don't know, 8% of revenue for G&A might be half of it moving forward, and we want to be ahead of that game. So we have a vast deployment. And it's not so much that we want to save the money, we want to reallocate it. Our biggest engine is our innovation engine. So every dollar we can put into more products and more innovation is going to be helpful. And our -- we are way ahead in terms of experimenting with it. It's a broad-based effort, and it's not only limited to go-to-market at this point.

Perfect. I want to close here on pool of funds. I also want to -- share with us your observations on how the consumption patterns within the pool of funds deals are tracking relative to your originals.

Yes. So maybe for the audience, pool of funds was a concept that we started to deploy last year because the amount of products we have now 64, as I said, is harder to sell. So how do you get customers comfortable and how do you find a way to make this process of consuming as much as possible from the platform and as fast as possible as frictionless as possible. So pool of funds for us is we negotiate with the customer contract, and it has a rate cut literally for every product we have. And there's a commitment to consume, I make this up now, $100 million over 5 years on a rate card that could mean any product. So when we started this, we -- it was quite successful. But that also means that all of a sudden, every salesperson in the company wants to a pool of funds, and it's a complicated tool. It's not good for every situation. So we -- there was some learning we had to do, but we've become better at it. And the second part was once you have a customer sign up for it, you have some conversations, they tell you where they want to start, but there's a lot of variability in this process. So we had no idea how fast they would ramp, how fast they would consume. And there was an uncertainty in the forecasting process because of that. Today, I think that noise is lifting and is disappearing. If I were to click on my phone here, I could see across all pool of funds where we are at this moment in time in terms of consumption, who is ahead, who is behind. You would see that on average, across all of pool funds deals, we are pretty much on target, yes, and where we need. That's hard work. But it helps us, I think, to be more precise in how we look at the business moving forward, and it's helpful that the noise is disappearing. So good progress.

Very good. Well, thank you so much for all your comments. Please join me in thanking Thomas for his time.

Thank you.