Hi, everyone. I'm Jeff Hickey with UBS, and I'm here with Paddy Srinivasan and Matt Steinfort, the CEO and CFO of DigitalOcean. We're very thankful to have you both with us here today. I'm going to run through some questions, but just as a reminder, you should have some QR codes that, if you want, you can scan and type up any questions you'd like, and I'll get them on the iPad here. But firing away, I wanted to start with you, Paddy. You've been in the role CEO a little less than a year. Strategically, you've been building out the bench. I wanted to get a sense of what your just kind of key focus areas and priorities are, maybe over not just the next year, but then ideally, what's your medium-term vision for DigitalOcean?
Yeah, great. First of all, thank you, Jeff, for hosting us, and it's a wonderful conference. Very glad to be part of this. Answering your question, the first priority was to rebuild the executive team, and I think we have done a really nice job of bringing on real industry experts, so when I started embarking on this mission, my focus was to bolster our bench with cloud and AI expertise, number one. Number two is, from a go-to-market point of view, bring on leadership that has not only scaled the business at a billion-dollar-plus level, but also do that in a way that can augment our product-led growth motion with product-led sales and indirect distribution and things like that.
Talking about product-led growth, the third priority for us was to make sure that we are doubling down on what makes DigitalOcean the company it is today, which is the love of our ecosystem and developers, and really getting back to our roots from that point of view. From that perspective, I'm super happy and very proud of the team that has decided to join us on our journey. We have a really, really awesome technology team that has built and operated significant-scale cloud in various hyperScalers and other companies. On the go-to-market side, also, I'm very happy that we have the kinds of leaders that we've been able to attract that have built very large go-to-market teams before.
In terms of the priorities over the next couple of years, obviously, the number one priority for us is to ensure that we are maximizing the share of wallet with our existing customers. I'll get into a lot of details later, but that's our number one goal. The way it manifests itself in our financial statements is, of course, with NDR, starting with NDR, which eventually drives top-line growth for us. A very close second priority for us is AI. I think you can't flip a page on the Journal today without reading about the profound impact AI is going to have on the economy and things like that. Our job is to translate that potential energy to kinetic energy.
And the way we are doing that from an execution point of view is really focused on building a comprehensive AI stack that can scratch the itch of companies of all sizes with all different capabilities in terms of AI. So that's a major priority for us as well. So these are the three things. One, build a team and a culture that we are very proud of that can build momentum going forward. Number two is really double-click and deliver on our core cloud computing platform. And number three is Jumpstart AI.
Great. There's a lot of directions we can go from that. Maybe the first one is just thinking about just the go-to-market and sales and marketing investments. DigitalOcean relies especially historically on a direct funnel serving much smaller customers, developers, that cohort. How do you see that changing over the next couple of years? Is this really focused on things like customer success and enablement for your largest customers and trying to ensure they don't perhaps move workloads to bigger clouds? Or what's your vision for that?
Yeah. So if you think about it, it's an and, not an or, for us. So what I mean by that is a large portion of our 640,000 paying customers are what typically would be classified as SMBs. But our SMBs are built different in the sense that, yes, they are small in size, but they are typically cloud natives, and they are typically software application vendors, which means they make money selling software. And the reason why that is important is when we talk about SMBs, typical connotation is your local pizza chain or your dentist office or a local brick-and-mortar company that wants to adopt technology. But our typical customers are digital and cloud-native companies that are running workloads on DigitalOcean, like ad tech or online gaming, online multiplayer games or streaming, e-commerce websites. And these kinds of cloud-native applications are what these companies run.
So how many of them do we have? We have close to 17,000 of these customers that we call Scalers who, on average, spend about $25,000 with us. And collectively, these Scalers drive about 58% of our revenue. The reason why this is really important is this segment of the market is the fastest-growing segment we have. Their ARR is about 19% year- over- year. And these companies are increasingly moving more workloads to us. And you might think, OK, why? They have great options with the hyperscaler clouds. The reason why they are moving a lot of their workloads to us boils down to three specific reasons. One is these companies, just by the sheer size and the expertise they have, they don't have the luxury of having dedicated people focused on storage or databases or networking and things like that. Typically, these technologies wear multiple hats.
So we take a lot of pride in making the complex simple on our platform. And it is not only simple, but it's also a complete cloud. So that's number one. Number two is we provide immense scalability, both in terms of performance as well as in terms of ROI and the elasticity of ROI. And number three is, by the virtue of who we are and our size and our friendliness towards our ecosystem, we are very, very approachable. And I like to say that we are large enough to scale with our customers and small enough to care about their success.
So these are the reasons why our customers absolutely love doing business with us. And that's our core focus area in terms of the customer segmentation. If you look at the next 12 to 18 months, our big focus is tending to the needs of the top end of these customers because they're growing and their needs are growing. Their needs are typically very global. A lot of the product innovation that you're seeing from us are anchored towards meeting the needs of these large-scale customers.
Got it. With the focus on meeting those needs, are there any specific areas that your Builders and Scalers customers are asking for? You've been working on just hardening the broader DigitalOcean platform, maybe more so from a functionality perspective as opposed to a discrete product you'd sell. But what areas maybe do you still have to work on?
There are a lot of things that we have already done over the last couple of quarters that you can see in our earnings. Let me rattle off some big areas of focus. Networking is a big area of focus, global load balancing, expanding our geographic footprint, improving our storage performance and our storage footprint, enhancing and enriching our database capabilities. Even our core Droplets, which is our flagship capability, we have released half a dozen new flavors of Droplets: bigger Droplets, smaller Droplets, Droplets with more memory, Droplets with more compute,
Droplets with attached high-performance networking, all these kinds of flavors to cater to the exact workload of what our customers are running on the DigitalOcean platform. These are some of the main themes, if you will. And then, of course, there's always better management, better visibility into the operations, better security, better access control, all of these things that you need as you run more sophisticated workloads on the DigitalOcean platform. That's what we're focused on.
Got it. That's very helpful. Maybe moving to some of the numbers. I know, Matt, you gave the guidance for sort of a soft baseline next year of low to mid-teens growth. Maybe you could walk us through what the bridge is to that. You called out NDR being a key priority for you, Paddy. So you have a goal to get that from 97 up to 100. But what are the other key drivers? And maybe how sensitive is each driver?
Yeah. So if you think about how we characterize the baseline growth for 2024, it was the number of building blocks. It was, as Paddy said, we have a very durable self-serve funnel. We get new customers every day. They come and sign up on our website and put in their credit card. And we're getting some right now. And it just generates a good baseline of growth. This year, it was around 8% that we had said we would get in overall growth from the self-serve funnel. And we'll do probably a little bit better than that this year. That's a good foundation to start with. This year, we had NDR as a headwind. So as you said, it's been 97% for a number of quarters, although that masks the fact that the core NDR, core DO NDR has been a little bit higher than that.
We've had some headwinds from the managed hosting side of our business, but so one of the first key levers is getting that from 97 to at least not a headwind. So getting that closer to 100 for the year next year, that'll be a big driver. The other two big levers are the managed hosting business, which is the acquisition that we had made of Cloudways several years ago. We had said that would generate two to three points of growth overall for the business this year, and we'll be in that range. So that's a good baseline to build upon, and then lastly, and probably the other of the two big levers, the first being NDR, the second is AI.
We had said that we would get 3% growth overall from our AI business in 2024, and we're a little bit ahead of that. That gives you a good solid foundation for low to mid-teens revenue growth next year. The sensitivity is, hey, we've got to get NDR up, and we've got to get it back so that it's not a drag. A lot of the product innovation that Paddy's been talking about is aimed entirely at that. We feel pretty good about the green shoots we're seeing with traction we're getting with our larger customers and the impact they're having on NDR.
If we think about getting NDR back to 100, is a lot of that just predicated on seeing expansion come back a bit? Or is there still, so to speak, a bit more meat on the bone of just limiting contraction?
I'd say that expansion and contraction are correlated a bit in that as you knock off some of these big product blockers, you'll keep more workloads in addition to winning more workloads, but in terms of the bigger lever, it's definitely expansion. Expansion is down the most relative to kind of historical levels. Churn has been very, very stable. It's at historical levels. Maybe there's a little bit of juice there, not a lot. Contraction has been a little bit elevated, but it's been steadily improving, and we project to get that back to kind of historical levels in the next handful of quarters, and it's not that elevated.
It's really the opportunities in expansion. Our customers that are growing just hadn't been growing at the same rate they had been historically. We think we can unlock that through a combination of the product innovation that Paddy's described, but also the go-to-market motions that we're layering on top to enable our customers, so we can identify the customers with growth potential, engage them, and try to accelerate the expansion that they can drive on our platform.
Got it. That's great. Paddy, it was helpful earlier, I think, sometimes just addressing the misperception of SMBs and you're working with digital companies, people that are creating software. When you talk about you brought up streamers and ad tech, et cetera, do you have a sense of maybe how tied the customer base is to macro? If we see an improving macro, if we get rate cuts or something in December and we move into 2025, could that layer on to additional upside, perhaps, next year?
Possibly. But we try not to think or plan around it. I think we have enough headroom to improve our operations, our product posture, our platform performance. That's what we are focused on, our customer engagement. So the list is fairly deep in terms of what we can do to control our own destiny. Hopefully, the market and if we do the right things, I think the market will take care of itself. That's the philosophy with which Matt and I are operating the company. And specifically on any kind of projections around planning, we are not factoring in any secular lift in the mood or performance characteristics of the macro.
Got it. Let's talk about AI because we have to, the UBS Tech and AI Conference. It's interesting. A question I sometimes get from investors is if you look at these neo clouds, GPU clouds that have crazy growth rates, and they ask about GPU supply and everything, you are a bit different in your strategy and vision. You talk much more about inferencing as being a source of perhaps longer-term durable demand. Maybe you could just talk a bit about your vision of DigitalOcean Paperspace as more of a platform and an AI platform as opposed to maybe just purely renting out GPU compute for training.
Yeah, absolutely, Jeff. I think this is a seminal question in terms of the future direction. And you're absolutely right. The name of the game for us is durable growth. We are here to build growth that we can sustain over the next two or three or four years. And for that, we start from the customer and work backwards. So our strategy is deeply rooted in knowing who our customers are and what their challenges are and how can we solve them for AI. And I already talked about who our core customers are. We are not changing that in any dramatic fashion just because there's a new technology on the horizon. So if that's the customer base we're going after, what are they trying to accomplish? Most of these customers are not trying to build a model. Building a model is not an end in itself.
Building a model is a means to an end. And the end is deploying AI in service of their business operations to improve a workflow, to do something which was not possible before, or to save time, to improve productivity, and so forth. So there needs to be a business outcome at the end of all of this. That's why our strategy, as we have articulated many times, is yes, of course, there's an underlying infrastructure aspect of the AI strategy. We must do that. That's the foundation. And on top of it, we have a platform. And then on the topmost layer is the AI agents or application layer. The infrastructure side, we have four different offerings right now. One is the bare metal, which gives you the maximum power and the least amount of abstraction for companies.
The next one is GPU Droplets, which also provides a tremendous amount of raw GPU power, but it abstracts out so few things, makes lives easier for AI/ML developers. We have a Kubernetes version of the same thing, and we have a Hugging Face version of the same thing. They're all pretty much the same thing, different front doors in order to cater to slightly different needs of the same class of customers that are trying to do pretraining or fine-tuning of models. So that customer base is typically fairly sophisticated. They have deep AI/ML backgrounds. They have a very specific goal of building and tuning a model to solve a very specific use case. Moving to the next layer is the platform. The platform, we are still in early availability. We have not even named the platform. Let's just call it GenAI Platform.
The GenAI Platform is a set of tools to enable rapid consumption of GenAI into everyday software applications. So one example that I talked about in the last earnings call is an IoT company that has a field service management software. Now, what happens in a highly industrialized plant maintenance kind of scenario? You have field service people calling on a customer to troubleshoot a machine. They often have to consume hundreds of pages of technical documentation to be able to troubleshoot a device. Now, can GenAI help in short-circuiting this and increasing the productivity of this field service person by digesting all this knowledge base or these PDFs and answer questions in a plain English manner? That's exactly what this company did. And they built a POC in less than a day, and they went to production in a matter of a few days.
So that, to us, is the perfect use case and the enablement that we want to do with our GenAI Platform. Now, it is not just wrapping an OpenAI API, but it's providing the building blocks of how do you build a knowledge base, how do you quickly build a chatbot, how do you put guardrails, how do you monitor the total cost of your consumption. There are many other building blocks that are essential to consume GenAI quickly in an existing application. So that's the second layer. And the third layer is the AI agents. And I'll make it super short. So we released a new agent called AI SRE Agent, which is Site Reliability Engineering. It basically takes out all the manual tasks in going through logs and coming to root cause analysis of web performance management from hours to seconds.
This agent is great at just crawling different logs and coming to the root cause of any kind of performance degradation in a matter of a few seconds. So that, again, has found good resonance with our Cloudways customers because they are managing large-scale websites on our platform. So these are some concrete examples of how we are enabling our strategy through our software. But again, it all starts with starting with the customer and working backwards to solve a problem because I feel infrastructure is not an end in itself. GPUs are great. They should be building blocks in a strategy. And that's exactly what we're doing.
We got a question from the audience that kind of works off this platform play. Somebody's asking, earlier, we heard from Lambda about building a GPU developer platform. So how do you think about the emerging set of competition, maybe from these other players that are also starting to encroach upon actually providing a developer platform as opposed to just raw GPU compute?
Yeah, I think that is super welcome. I think we all have to be on the same mission of making it super easy for developers to build and deploy applications. As I said, we need to reduce it from the quarters it is taking today to months and days and hours. And that should be our collective ambition to make AI mainstream. Right now, AI is a barbell where you have very, very large companies using AI, and you have very, very well-funded startups on the other end of the barbell building models and infrastructure. The middle of that barbell is just waiting there to be enabled. And that, I think, is a very welcome development where the more companies that innovate to making it easy for any software to consume GenAI to deliver on a business need, I think, will be awesome.
With your AI exposure right now, it's roughly tripling, I think, year- over- year. Just curious, do you have a sense of how much mix of that is still maybe some of those customers doing a lot of training? And then, Matt, you called out on the last earnings call that those AI customers, that's not included in NDR right now, maybe because you get some spikes and pauses. It's just very spiky at a customer level. Could you see that changing over time? Is that mostly, as you assume, more mixed turns to inferencing versus training that you could actually put that into NDR?
Yeah, we certainly are evaluating it. I mean, the challenge we have is if you look at our customer base, we've got a lot of customers that have been with us for quite a while relative to the capabilities that we provide. But even if you track them on a monthly basis, their utilization is spiky. So they may have a big training exercise that they do, and then they go back and they do some fine-tuning, and they refine the application somewhere else. And then they come back months later and do something similar. So the patterns just don't kind of fit what you would think of with a traditional kind of SaaS application, and so we're just keeping that out for now.
But as you think about the strategy that Paddy articulated, as you get more of our customers that are using the GenAI application that Paddy described or using some of the other capabilities that we offer, we expect those will be more inferencing-like and that would be more of a steady, recurring kind of a revenue stream. And then you would see growth with those customers, much like you do with our existing customers, where as their businesses grow and their utilization grows, you see expanded usage on the platform. Right now, it would just cause too much noise. And as we said, the two big levers for us are separate. Driving NDR in the core is the first big growth lever. And AI revenue is the second. So we'll just continue to report those separately rather than muddying them together.
Got it. Maybe shifting a bit just to margins, sometimes I get asked from investors, how should I think about players like DigitalOcean and just your margin structure competing against the hyperScalers and just the immense amount of capital they have and the scale they have? And if you have lower pricing with simpler offerings targeting smaller, less sophisticated customers, how can you even maintain that margin you do have today? So maybe you could talk a bit about how you differentiate either at the software layer or you could also bring up the Atlanta data center investments that you announced very recently.
Yeah, I think from a margin standpoint, I mean, it's all about price to value. And we provide a service that's simple, that's easy for them to consume, that is scalable on their platform. We provide a higher level of service than they would get at the hyperScalers. And there's value in that. And customers are willing to pay that. Our customers are also, even our top customers, are at the small end of the size of a hyperscaler customer. And so they likely wouldn't get super aggressive discounts at the hyperScalers. And so there's room for us to price in a way that is good value for the customer and yet still maintains good margin for us. And we also have the benefit of probably the most efficient customer acquisition model in the industry.
I mean, if you look across all software companies, I mean, I don't think anybody has the same kind of sales and marketing expenses as a percentage of revenue that we do. It's very, very efficient. We're looking to bolster that with additional go-to-market functions. But we've been able to maintain, I think, attractive margins. On the COGS level, like you said, with the Atlanta data center, we think there's an opportunity for us to improve our gross margins by optimizing our data center footprint. We're in a lot of really expensive colo facilities today and really expensive markets. And we could go to places like Atlanta and other places that are maybe the next neighborhood to the expensive places and save quite a bit. We still have a lot of leverage, I think, we can drive on the operating expense side.
We're hiring quite a bit of R&D engineering right now in India, and we've demonstrated the ability to operate globally, much like our customer base, so we think there's continued scale that we can achieve there, and so I think with our guidance for our kind of soft indication for next year of revenue, we also provided a soft indication that we think will be about the same free cash flow margins next year as we are this year, around 15%-17%, and that's on the back of driving a lot of operating efficiency into the business while at the same time absorbing what's a lower kind of margin AI business in the near term.
Got it. Would you consider investing a little more just incrementally into the AI piece if you saw just even stronger demand signals than you do today? Or sometimes I get asked, like, why aren't you just going more for bare metal GPU compute? You could drive more revenue in the near term, but.
So the short answer is we did that this year. And let me take a step back. There are two angles in which we have to look at this. One is the financial constraints, and the other one is what does our strategy require us to do? And again, I want to go back to the word durable. We are in this to drive durable growth, not spiky growth. And for us, durable growth means democratizing access to AI to customers that we typically serve. And that requires a careful orchestration of these different parts of the puzzle to make sure that we are building a very comprehensive AI platform that enables our customers to take advantage of AI and deliver a business outcome for their customers. So from that point of view, we think we are investing appropriately.
And if you think about relative to our size, our CapEx, and specifically talking about our AI CapEx, is pound for pound very comparable, if not more aggressive. Because unlike some of the hyperScalers, cloud is our only business. And any capacity we are deploying is all in service of our customers. Nothing gets diverted to running an optimization algorithm for a different part of our business or anything like that. So from that point of view, we feel fairly good about our plan.
And also, we have to get the order of operations right in terms of the generational shifts that are happening with the GPU chipsets. And as we keep moving upstack to platform and applications, we want to make sure that we have a proper distribution of heterogeneous composition of inferencing platforms and things like that. So it's a very thoughtful, calibrated strategy. If we see opportunities to accelerate our durable growth like we did this year, we will absolutely take advantage of that.
Got it. And then I just want to check to see if anyone has one in the audience before I wrap up, maybe just with a final quick question. I'll just hit it. This is more financial just for you, Matt, just with the convertible note, a couple of years away still. But how are you assessing how you will approach that as we get near and near over the next couple of years? Could you look at just term loans? Or are there any other financial mechanisms you are considering right now?
Yeah, we're in a great position with respect to our balance sheet. We've got a lot of cash on the balance sheet. We've got that zero-coupon note that doesn't mature until the end of 2026. We have a lot of different options, and we're in active conversations with bankers and talking about the various levers. We could do a combination of a smaller convert and then a traditional straight debt term loan or some other instrument. And that's something that we'll pay, I think, more aggressive attention to as we get into 2025, still have a year before it goes current, so we've got a lot of option value.
OK, perfect. Well, I think we're at time. So thank you both for joining us today. And let's give a round of applause.
Thank you, Jeff.