Thank you for standing by, and welcome to the DigitalOcean Second Quarter 2024 Earnings Conference Call. All lines have been placed on mute to prevent any background noise. After the speaker's remarks, there will be a question-and-answer session. [Operator Instructions] Thank you. I'd now like to turn the call over to Melanie Strate, Head of Investor Relations. You may begin.

Thank you, and good afternoon. Thank you all for joining us today to review DigitalOcean's second quarter 2024 financial results. Joining me on the call today are Paddy Srinivasan, our Chief Executive Officer, and Matt Steinfort, our Chief Financial Officer. After our prepared remarks, we will open the call to a question-and-answer session. Before we begin, let me remind you that the statements made on the call today may be considered forward-looking statements, which reflect management's best judgment based on currently available information. I refer specifically to the discussion of our expectations and beliefs regarding our financial outlook for the third quarter and full year 2024, as well as our business goals and outlook. Our actual results may differ materially from those projected in these forward-looking statements. I direct your attention to the risk factors contained in our filings with the Securities and Exchange Commission and those referenced in today's press release that is posted on our website. DigitalOcean expressly disclaims any obligation or undertaking to release publicly any updates or revisions to any forward-looking statements made today. Additionally, known GAAP financial measures will be discussed on this conference call, and reconciliations to the most directly comparable GAAP financial measures are also available in today's press release, as well as in our investor presentation that outlines the financial discussion on today's call. A webcast of today's call is also available on the IRS section of our website. And with that, I'll turn the call over to Paddy.

Thank you, Melanie. Good afternoon, everyone, and thank you for joining us today as we review our second quarter results. DigitalOcean delivered a strong second quarter, building on the momentum from the first quarter and continuing to execute on all key metrics. In my remarks today, I will briefly highlight our second quarter results, provide an update on the leaders we hired recently, share tangible examples of our increasing product velocity, and discuss how we are capitalizing on our AI growth opportunity. First, I would like to briefly recap our second quarter 2024 financial results. Revenue growth has continued to reaccelerate on the second quarter to 13% year-over-year, reflecting the growing signs of success we're seeing from both a product and go-to-market standpoint and the continued acceleration of our AI and machine learning products, where ARR has grown over 200% year-over-year from the Paperspace ARR we acquired last year. In Q2, we also saw the largest step-up in incremental total company ARR in nearly two years, excluding the quarter in which we acquired the AI/ML business. We also delivered strong adjusted EBITDA margins at 42% and adjusted free cash flow margins at 19%, exemplifying our ability to demonstrate ongoing cost discipline and optimization while continuing to accelerate product innovation. Our second quarter financial results highlight the progress we are making and our ability to execute on the plans we laid out at the beginning of the year. We're also encouraged by the signs of improvement within both our growth profile and our key fundamentals. Net dollar retention was flat versus the previous quarter at 97%, as expansion within our customer base continues to be lower than historical levels, given we're still navigating a challenging macro environment, which is muting the positive impact of our increased product velocity and the stability…

Thanks, Paddy. Good afternoon, everyone, and thanks for joining us today. In Q2, we continued to execute on the plans we laid out at the beginning of the year. We made progress on key metrics. We continued to see revenue growth reaccelerate, and we delivered favorable adjusted EBITDA and adjusted free cash flow margins. Revenue in the second quarter was $192.5 million, up 13% year-over-year and up 4% quarter-over-quarter. We added $32 million of annual run rate revenue, or ARR, in the quarter, which was 158% higher than the incremental ARR we generated in Q2 of 2023 and was also the highest step-up in nearly two years, excluding the quarter in which we acquired our AI/ML business. Contributing to this growth was healthy incremental revenue from new customers, increased momentum from our AI/ML platform, which saw significant growth quarter-over-quarter, and contributions from our managed hosting platform, which continues to be one of our faster growing platforms, all of these together offsetting a flat quarter-over-quarter net dollar retention rate from our existing installed base. Our Q2 net dollar retention rate was 97%. As we saw last quarter, we continued to see stable performance in net expansion, which is defined as expansion net of contraction on our core DigitalOcean platform. Contributing to this stability was our increased product velocity that drove an increase in ARPU, helping to offset the broader macro pressures on net expansion in our customer base. Our churn levels have also remained very stable for over a year across the business. We are encouraged by the stability in NDR and the modest sequential improvements we are seeing, despite the challenging macro environment, which is muting the pace of improvement in NDR, and despite a positive but lower contribution to NDR from our managed hosting platform, now that we have…

Thank you. We will now begin the question-and-answer session. [Operator Instructions] Your first question comes from the line of Gabriela Borges from Goldman Sachs. Your line is open.

Good afternoon. Thank you, and great to see the step up in ARR. Paddy, I wanted to follow up on your comments on AI strategy. Help us understand, are you saying that other GPU providers don't offer the same level of fractional access that you do with droplets? And then, broadly speaking, comment on the competitive environment. There are a number of well-funded GPU providers that are spending orders of magnitude more CapEx than you are. How do you think about the sustainability of your differentiation when you're up against competitors that have that kind of CapEx fund? Thank you.

Thank you, Gabriela. Great question. So as I started the prepared remarks by talking about our AI strategy and how it is differentiated, I want to remind everyone we are also a very strong player in the GPU infrastructure as a service today. So we have very competitive offerings all the way from bare metal to virtualized environments. And what we announced with the GPU droplets is, it's a strategy which enables access to fractional GPU access and also with the ability to orchestrate the lifecycle through the virtualized environment that we provide with GPU droplets. So it just takes the overhead of having to manage the whole infrastructure, makes it super, super easy for those workloads which we believe are going to be more important for customers that are digital natives but also the types of customers that typically reside on our platform. So they are looking to build applications that consume AI and leverage different parts of the AI infrastructure in a way that extends existing AI models. They're not model builders per se in the sense that they're not trying to build another foundational model. So we feel our AI strategy, which includes the GPU infrastructure, is tailor made for customers that are looking to consume AI, not necessarily build foundational models. So when I talked about the GPU droplets, that's an abstracted version of the core GPU as a service. And then we also in the deploy conference, we also announced and showed an early preview of what we call the next generation of our platform as a service offerings. So these are endpoint APIs for well-known open source models like Llama 3.1 and so forth, which we will be releasing in early Q4, which enables another layer of abstraction to build applications on top of…

Thank you for the detail. Matt, the follow-up is for you on the updated outlook. How did you think about the pace of net new ARR into 3Q and 4Q? Back at the end of it, Matt, I think it implies a moderation from the result that you just put out this quarter. So help us understand how you thought about handicapping that, call it, $7 million, $8 million that you were able to deliver in 2Q into 3Q and 4Q as well.

Yeah. Part of it -- as we think about the growth, particularly around the AI side of the business, it's somewhat lumpy and we talked about earlier in the year that we had some risk baked in, and as part of the reason why we gave a wide revenue range in our initial guidance was, we had some supply chain risk, and it was a new service for us, so we wanted to make sure that we were comfortable with the implementation. And we managed past that risk, and that enabled us to bring on a fair bit of incremental capacity that we had ordered last year in the second quarter, which drove up the ARR. And so as we look over the balance of the year and we're continuing to add capacity, it won't be as lumpy, so it's going to be a little bit lesser in terms of the guide than what we had in the second quarter. The other thing that you're getting when people look at the implied third quarter and fourth quarter growth rates is, from a year-over-year perspective, we've got two lapping that we're doing that are making the growth look like it's decelerating, even though we're accelerating clearly in the core and we're showing growth in the other parts of the business. And that's because this is the first quarter or third quarter will be the first quarter where the Paperspace AI revenue is in the prior period. And so that's a 200 basis point reduction in the growth rate. And then, we're lapping a price increase and customer growth, kind of nominal customer growth in the Cloudways business that's also contributing about 100 basis points to the year-over-year growth apparent deceleration from second to third quarter.

Got it. Thank you for the call.

Your next question comes from a line of Raimo Lenschow from Barclays. Your line is open.

Thank you. Can I go back to the first question that was asked? Just given that we are kind of early on that Gen AI journey, so we don't really fully understand how this is going to play out. So is the message from you guys that as I train a model, I need like hundreds, thousands, ten thousands GPUs, but as I run the model, then if I have an application, I can do that with a much, much smaller GPU number, like the one to eight that you were kind of suggesting today? Like, it just feels a little bit like, because you're early in the journey, it feels a little bit too good to be true. Can you speak to that a little bit more?

Yeah, Raimo. Thank you for the question. So there are three types of AI users today. One is the true foundational model builders, like, an Anthropic Mistral, folks that are building the foundation models. And the second one is, what I call the AI extenders. So they take these models, say, a Llama 3.1, an open source model, and they inject or they enhance it using their custom data. It could be a company that has, say, for example, a geospatial data, and they enhance an existing LLM with their own custom data source and create a slightly modified version of an existing foundation model. So for this class of customers, they definitely don't need hundreds of thousands of GPUs. And the third type of user is an AI consumer. So for example, let's say you're creating a new AI native CRM application or a supply chain application for which you are relying on a very robust foundational model, again, Llama 3 or Mistral or something. But bulk of your application is to deliver that supply chain forecasting algorithm, but you're leveraging a heavy dose of AI. But you don't need the same kind of physical raw compute power that the Category 1 or Category 2 company needed. So these are different types of use cases. So the AI model builders, AI model extenders, and the AI model consumers all have different requirements and need GPU capacity at different scales. And what we are seeing is, we are definitely addressing the needs of the second and the third categories that I just talked about. And another way to think about this, Remo is, as with any technology wave, you have infrastructure providers. In this case, this is NVIDIA and all the foundational model builders. They are laying the infrastructure, but the true business value is going to be when this infrastructure is leveraged to build platforms like simple example would be operating systems based on x86 architecture. And then you have applications, which are the ones that truly deliver business value for everyone. So as this AI wave goes up stack from one layer to the other, we feel there's a tremendous amount of need to democratize the access to these GPUs and also provide other software frameworks LNB on the platform layer and infrastructure layer, which is what we are building now.

Okay. Perfect. That makes total sense. Thanks for the clarification. And then one for Matt, if I think about the improvement in gross margins you mentioned from just the optimization work you do around the data centers, like, what sort of magnitude are you thinking? I don't want guidance, but is this like a few basis points? Is this kind of more meaningful? How should we think about that? Thank you and congrats from me, as well.

Thanks, Raimo. Yeah. It's a great question. And I think I've touched upon this on prior calls. When I first joined the company, one of the first things that I observed was that the data center footprint was, I'd say, interesting in that we were in tier 1 markets and tier 1 buildings. And that was pretty much the way the company had architected its data center network, which is a relatively expensive way of building your architecture. Over time, you'd want to move to some tier two locations and you'd want to be more of a wholesale than retail kind of data center kind of footprint. And so the Atlanta data center that we just announced is a great example of that, where we're going to be able to take down an incremental capacity. It'll give us room for the expansion of our AI capabilities, but we'll also be able to take workloads out of our more expensive locations and kind of reduce or consolidate those footprints and shift some of the, maybe the less latency sensitive workloads into different locations. And that's going to be a multi-year effort. It's not going to be something that we can do over time, but I think there's a meaningful improvement that we can drive in gross margin. And the objective is to continue to drive gross margins up, and that will offset the core business, and that will offset some of the lower gross margin mix that we'll get as our AI business continues to grow. So I'm not going to give guidance on it, but I think it's a meaningful opportunity, and Bratin and I are very focused on driving that savings and driving the gross margin up.

Your next question comes from a line of Kingsley Crane from Canaccord Genuity (ph). Your line is open.

Great. So first question would be, you cited customer success efforts as helping to create stability in churn and contraction. Especially now with Larry on board, how much could customer success be used not to just prevent churn, but also drive usage in these high potential accounts?

Hello, Kingsley. Good to hear from you. It's a great question. So we currently, our customer success efforts are fairly nascent, but ramping up. And they already are tasked at not only managing a relationship with our top customers, but also drive product usage and show them the breadth of our platform, which after I joined six months ago, we have put some considerable wood behind the arrow in doing two things for our core platform. One is, we are rapidly closing the gap in terms of what our top end of the scalers need from a core product point of view, but also adding a lot of other new capabilities on our app platform, as I mentioned in the prepared remarks. So we are doing both, closing the gaps in the product functionality that our top end scalers need, but also expanding our platform footprint. So our customer success reps, they get trained in all the latest and greatest, and they take that message to our customers and help expand the footprint of these large customers in how they're leveraging our platform. So that's already there, but I think it is fair to say that it's in its nascency. And with Larry coming on board, there's going to be a big push for us. So the purpose of getting someone like Larry is to ensure that he can leverage all his skills in building a high velocity motion to augment our phenomenal product-led growth and make sure that these customers that are getting onboarded to our platform, first of all, know the breadth of our platform, but also start leveraging the capabilities that we have so that we can drive up expansion.

That makes a lot of sense. And then I want to return to the GPU topic. A few comments on the availability. It's great to see that out in the market. And the fractional demand consumption is relatively unique. We've spoken in the past about software will differentiate you from the other providers. So I just want to check in on gradient specifically. How important is that to the vision and how is that acting as an on-ramp for customers?

Yeah. So we are, so yes, I completely agree that software as a differentiator is going to be the one that we are going to bank on. And so right now, so let me again refresh everyone's memory in terms of the offerings we currently have. So at the bottom most layer, we have a bare metal GPU platform. And we just announced the GPU Droplets, which provides virtualized and fractional and on-demand access, which is also a pretty big deal. Virtualized, on-demand fractional access to GPUs is what the GPU Droplets provide. And then a layer up above that is our platform as a service offering, which is Gradient. And we are doing a lot of work on that platform as a service layer. And we'll be announcing a slew of enhancements over the next 90 days to add a lot of capabilities on the platform as a service layer. We believe it is going to be an important front door for customers to start adopting these AI capabilities. And the abstractions that we are providing right now, like, for example, when customers come into the Gradient platform, yes, they know that they are leveraging GPUs, but that's not why they are coming to our platform as a service offering. They're coming because it is super compelling to organize their AI/ML workspaces and share it amongst their colleagues and publish it to different repos and things like that. So there are many different reasons why they come there. And by the way, when they have to push something to testing or production, they leverage the GPU infrastructure we have, in which we have a variety of different hardware, not just H100s. We also have a variety of different GPU hardware that helps customers for different use cases. So I think software is extraordinarily important as we go up the stack. And Gradient is a very differentiated offering for us. And we are working quite a bit on enhancing Gradient over the next 90 days.

Really encouraging. Thank you.

Our next question comes from a line of Patrick Walravens from JMP Securities. Patrick. Your line is open.

Great. Thank you. And first of all, I want to commend you on your recruiting. I mean, the background, like Bratin in particular, is just fabulous. So someday I'd like to hear how you did that. But my more specific question is, what are the bottlenecks to growing your AI footprint in terms of the data centers? So data centers that were built for CPUs don't work very well for GPUs, right? So how are you thinking through that?

Yeah, Pat. Thank you so much for the kind words. Yes, you're right. There is surely some constraints, but all of, so just to be clear, the new data center that I just talked about, we are not live yet on that. So all of the AI workloads that we are seeing now, we have crammed into our existing data centers, whether it is New York or Amsterdam and other places. Is it easy? No. But we have figured out a way to do that. See, there are multiple constraints. One is the stuff that you read in the press is power and cooling. Yes, those are really important. The other one is network stack. It's very different from our CPU offering. So we have to build parallel network stacks to put in these GPUs. And then real estate. The density of our ability to rack and stack GPUs is very different from our traditional density of our racks for CPUs because of the power and the heat that these GPUs throw out. So there are some physical constraints, networking constraints, and of course, it is just super expensive and super hard to find more data center space in our existing facilities. So that's why we decided to go to Atlanta. And by the way, it also helps us consolidate and load balance some of our very expensive data center footprint in the short term.

16 data centers across nine regions. Did you have your slide deck? Does that include the Paperspace, whatever Paperspace has?

Yes, it does. And Paperspace coincidentally happened to be in our New York City data center. So, yeah, it includes that.

Okay. Great. Thank you.

Your next question comes from the line of Mike Cikos from Needham & Company. Your line is open.

Hey, guys. Thanks for taking the questions here. And I just wanted to continue the thought on the data center optimization strategy that you guys are working through. With the Atlanta data center that you guys are announcing here, should we think about that as really taking on most of the AI workloads or is it freeing up capacity for you guys to further build out those GPUs in the existing footprint that you have? And the reason I'm asking is I know when you had gone through those three types of categories, the AI foundational model builders, the extenders, the consumers, I would think that at the very least the extenders would probably be more concerned about security. You'd need to be in like a tier three, maybe even a tier four data center given security restrictions. And then, can you help us think about the ability to re-architect the capacity that you have for these GPUs? I know we're announcing Atlanta today, but should we expect a more significant build out even beyond Atlanta?

Yes. Thanks, Mike. I'll start with the last question. Should we expect a more significant build out? I don't think so. I think we have enough capacity now for the foreseeable future. But I think the way I look at it, Mike, is slightly differently, which is as we go up the stack, the workload goes from training to what we in the industry call as inferencing. And inferencing, it is important to minimize the latency. So if you ask me whether all of Atlanta is going to be consumed by AI, no, because that's why Matt and I have been talking about we need to consolidate and load balance across all of our footprint, because a lot of the AI capacity will also be in the West Coast or in Canada or in Europe or Asia, because as our inferencing demand grows, some of our customers are already asking us to put GPUs in inferencing mode closest to where their customer density is, whether it is West Coast or Asia. So luckily, we have a fairly distributed data center footprint across the globe. So, we are going to take this opportunity to optimize and load balance and consolidate some of these data centers so that we can do a couple of things. One, in our core business, we can continue to look at improving our growth margins, get out of these expensive locations and leverage larger scale data centers like the one that we just talked about. But B, as the AI world moves from training to a more inferencing mode, we want to also have the ability to deploy the inferencing GPU capacity closest to where our customer demands are so that we can reduce latency and give them the horsepower they need for inferencing. So that's how I look at it. Matt, is there something you want to add?

I think you answered it, Patty. I think to answer the first part of your question, Mike, it's not going to be filled with the Atlanta data center. It's not going to be filled with GPU, but it will take a big chunk of our capacity. And part of that is just getting that GPU capacity that, as Paddy said, we put in some of our more expensive locations. There's a more cost effective way for us to do that and we'll migrate a lot of that. That will free up capacity in those more expensive locations. For workloads that are more relevant for that, where they require that latency or it's justified to have them in the more expensive locations. So, this is just a great opportunity for us to combine two very important initiatives. One is extending our GPU runway and the capacity we have. And the other is to just load balance and cost optimize in our existing footprint.

And just to add one more thing, to make it clear, we are not going to rush to fill out the capacity to the brim. So, this gives us land and space and power, but we are going to fill it up as part of our consolidation strategy and as our AI demand grows over the next few years.

Got it. And then just a quick follow-up for Matt. Just coming back to the prepared remarks on the net dollar retention. So the takeaway here, we're stable at 97% and based on the guide that we have, the team's assuming that that 97% remains where we are for the rest of the year. And then the second clarification here, but I know last quarter you guys had cited 19 million in AI-related ARR. Can you give us an update on what that is this quarter?

We're not going to guide, we're not going to provide the details at the platform level at this point going forward. But I think with the 200% increase in growth, the 200% growth rate, not just increase, but 200% growth rate on what we acquired, you should be able to figure it out in terms of what the ARR was for the AI business. And I'm sorry, the first part of the question was around the 97%. No, what we're trying to be is, I mean, we're doing a lot to drive NDR up. Clearly, our goal is to try to get that above 100 and have that be a tailwind. And we're making good progress on a lot of the initiatives as Paddy talked about and the new products and the increased product velocity is contributing. But what we're seeing is primarily the expansion is just stubborn. It's not increasing at the pace at which we'd like to see it. And a lot of that is just our customers aren't growing as fast as they were previously and their recovery is not as fast as we would like. So I don't know, if we think it's going to be flat. It's just, I mean, we only show it on a rounded basis. And so it looks very flat, but we are seeing kind of green shoots and steady improvement in that. So, I think the plan and the guidance that we have doesn't require it to get any better than it is today. And so that people shouldn't worry about, but we're clearly working aggressively to try to get it up above 100. And I'm optimistic that we'll see movement in it. It won't stay completely flat, but -- because it's such minor movements, we're just kind of saying it's going to be stable.

Terrific. Thank you.

Your next question comes from a line of Pinjalim Bora from JP Morgan. Your line is open.

Hi, great. Thanks for taking the questions and congrats on the results. Paddy, you have really strengthened the leadership team, obviously with the number of hires that you highlighted. As these leaders kind of settle into their seats, how should we think about some of the changes that you might be trying to drive? And are these changes more focused on for the next year or the second half of this year and any expenses related to those changes that we should be thinking about in the model?

Yes. Thanks, Pinjalim, for the question. I'll answer it in the reverse order. Should we expect to see a different expense profile? No. I think we are going to be super disciplined. And I don't think there's anything fundamentally different that we are going to be doing that will call for a foundationally different looking P&L or expense profile. In terms of what we should expect in terms of their impact, I would say let's take the three key functions, right, on product and technology. Bratton obviously brings a wealth of AI/ML background. So, as we go through the rest of the year, you should hear a lot from us in terms of both core product innovation as well as our AI strategy. So, my objective is between now and end of the year, we have to pick up even more pace. As I said, we have doubled the product velocity, but we are looking forward to pushing that even higher. And we absolutely want to make our customers super successful on our platform and give them no reason to think about leaving us or leveraging any other infrastructure. So we want to really ramp up the pace of innovation, and you will see that this year. The second thing is AI strategy. I want to make sure that we continue to refine our AI strategy in response to our customers. As I've been saying for the last couple of calls, we want the AI strategy to be our strategy to serve our customers. So, you're seeing a lot of different voices in the market in terms of companies that are chasing the big foundational model builders. That's not us. We want to be very disciplined to ensure that we know exactly which customers we're going after and what they…

And that concludes our question-and-answer session. I will now turn the call back over to CEO Paddy Srinivasan for closing remarks.

Again, thank you, everyone, for spending your hour with us. As I mentioned, we are super excited with the progress. We have a lot of work to do, and we are looking forward to staying engaged. And you all have a great rest of the day.

This concludes today's conference call. Thank you for your participation. You may now disconnect.