Thanks, Nitin, for coming. Nitin's the CFO of CoreWeave. He's been at the company since March of 2024, helped lead a very successful IPO, and had spent a lot of time at amazing companies like Google, Amazon, among many others. Thank you for sharing your story and the wild ride we're on. It's been fun to watch.
Well, thank you so much for having me here. Really appreciate it. Thank you for your partnership since we've gone public.
Yeah. It's been a great story to cover. I think we're all witnessing this is an incredible shift that we're all seeing. I think the question we get is we're seeing it in the revenue now, I think the focus of how expensive is this going to be is front and center. I know there's been a lot of focus about the back half of the year. The revenue's coming through. You're blowing through your booking numbers. You're doing great on the top line. I think everyone's like, "Hey, we're all seeing what's going on with memory, we're seeing what's going on in the supply chain." How do you think about the management of this is probably the single biggest question that we get when we talk to investors, and how you're managing it.
Yeah, absolutely. There are a few components to it. When you think about our business, it is structured on contribution margin at a contract level. That is something that we monitor diligently and make sure that our economics across those customer contracts stays intact as we write those contracts through different generations of infrastructure. What you're seeing in our business right now is that we brought in a lot of power online in the late half of last year, first half of this year. That's happening through now, through the beginning of Q3. We are reaching an inflection point in terms of that power growth, and then deploying GPUs and delivering it to our end customers. All of this is very near term.
That, coupled with the disciplined contract-enabled contribution margins that we know we are going to accomplish because we know the cost structure for those contracts, we know the revenue structure, which is a fixed take of a long-term committed contract, gives us confidence that as we hit this inflection from Q2 to Q3, our margins are going to inflect, with them reaching low double digits as we exit the year. The dynamics of what we deployed last year are very different based on when power is coming online to where it is this year. We're going to enter 2027 in a much stronger position with this inflection point that we talked about achieving at the end of Q2, beginning Q3.
That's what gives us confidence that because of its near-term nature and what we are deploying today, as well as the strong fundamental underlying contribution margin economics for every single contract that we write, that we are going to hit those numbers.
The thing I think we're all learning is that things are changing every day. Everyone says, look, when you have thousands of components going into a data center build and the PVC piping's changing and the memory price is changing, how do you deal with all this change? When we see prices go higher, which we're seeing in memory, everyone says, "Well, how do you deal with that? Is this contracted up front? Do you put an escalation for memory in the contracts?
A bit of both. When we think about our contracts, we have visibility into the cost structure and which data center they're going to go to. What we do in our business, in a structured way, we place the POs for the infrastructure as we sign customer contracts, which basically locks in both our top line as well as a large fraction of our cost structure alongside with the data center leases. That gives us confidence that the contribution margins that these contracts will achieve once they are actually deployed and running at scale would be very, very static. The variable components are very small within the error bands of what we would consider tolerable.
In certain cases, to your point, as we write the next incremental contract, if the capacity prices are increasing at that time, you kind of have to consider them as pass-through cost to the end customer because that will get adjusted in the price that we write for the next contract based on the capacity availability and the cost at that point of time. Effectively, we're writing each one of our contracts to certain stable margin profiles across our business, depending upon the scale, size, strategic nature, term length of the customer, but that's the error bar that we look at.
That's great. The due diligence we've done is that you're among the elite of the elite. There's not many people that can do what you do.
We tend to agree with that.
Everyone says, "Well, how are they so good at this?" Maybe if you can describe what has made you the most elite builder of the new AI world.
Absolutely. When you think about the cloud that exists today, it was built in the early 2000s, and it was built beautifully. I was a part of building all the three clouds that exist today and from a hyperscale perspective, from Azure to AWS to Google Cloud. They were built for a very different purpose. They were built for hosting your websites, running your web applications, hosting your data lakes, which is what you call serialized processing. When you look at AI workloads, they are fundamentally different in how they operate. They are basically what you would call as parallelized workloads. The infrastructure stack, the networking layer, and the software that sits on top of it to support that is fundamentally different.
When all of these clouds kind of looked at the AI wave, they looked at it in an evolutionary manner, and they were tasked with taking the hundreds of billions of dollar of infrastructure and almost two decades of engineering to fit that into that construct. We built an AI-first cloud from the ground up first principles in terms of what was needed for these workloads and these developers. That is what fundamentally differentiates us. Which is why when you look at our platform today, our platform is expanding beyond GPUs. Two quarters ago, we talked about the storage product in our platform exceeding $100 million ARR, and that's still growing as weeks. We also talked about three other lines of businesses in our portfolio. Networking, CPU, and software, all three crossing $100 million of ARR by the end of this year.
What differentiates us is the software stack, along with the operational ability for us to execute, which again, is aided by our software stack, to deploy and execute this performant infrastructure at scale. We were the first ones to deploy H100 at scale. We were the first ones to deploy H200 at scale. We were the first ones to deploy the Blackwells at scale. Vera Rubins are coming just across the corner. We tend to achieve similar fates there. For the last two years in a row, we've been the single platinum cloud provider in the AI infrastructure space in the SemiAnalysis, which is a very detailed, in-depth technical analysis of the platform. What differentiates us is from a technology standpoint, what we've built, our operational capabilities, coupled with our ability to execute at this scale in the capital markets.
Michael Intrator, our CEO co-founder, always says that when you're building a business, when you're starting a business, access to capital is most important, and when you're scaling a business, cost of capital is most important. We've been very diligent in lowering our cost of capital over the last couple of years, with more than 600 basis points taken out on our weighted average cost of capital. Now we've come to a point where, for investment-grade customers, we have rated facilities, which are A-minus rated by 3 rating agencies, which allows us to basically narrow the cost of capital differential between us and the hyperscalers in a material manner. All of these 3 pillars, technology, operations, and our financing ability, makes us a unique one of one.
When you think about just the next steps, I sat with one of your close partners, and they showed me what the data center's going to look like and how complex the racks are going to look like, and my head was spinning. I'm like, "There's no one that can keep up." There are pitches that you guys are the only ones that can do this. The complexity we're about to see in the next year to two years is going to blow everyone away. How are you staying in front of this leading-edge curve, and how are you distancing yourself from others that are doing this?
Yeah. I think one of the key things is the customers we work with are the most leading-edge customers in this space. They're the frontier labs who are basically redefining the face of AI. They are the first adopters of AI, and they're the hyperscalers. We get a lot of signal from these folks in terms of understanding where the market is going and what are the next requirements are, which allows us to then build our product roadmap, our engineering roadmap, as well as our operational capabilities to what's coming next. That has been the success mantra for CoreWeave for a very long period of time, where we are not going where the puck is going. We are basically leading our way to where things are heading.
About two years ago, when the world was still in air-cooled, we had practically determined that liquid-cooled is the future of the world, pretty much every single data center that we signed at that time, two years ago, was all effectively liquid-cooled. We are able to identify those trends and effectively act upon them much ahead in advance in terms of our ability to execute against them and shape those in the industry versus following them in the industry. That's been a great success parameter for us.
The signals that you don't think this is a bubble, what are the signals that you get? Someone sent something around the last time the Knicks were in the championship, the market peaked and sold off. Everyone's asking, maybe it's gone a little crazy inside some of the infrastructure stories, but what's giving you signals that this is more sustainable than bubble?
The most sustainable signal that we're getting from the market is how the demand is now shaping towards inference. A majority of the demand that's coming to us now is for serving inference workloads, which is giving us immense confidence that this is a sustained economic change that's happening. From an end customer perspective, customers are productionalizing AI now, and they're seeing massive returns on those, and that's what is leading to the belief that the infrastructure that we are building has longevity. Few for facts that this kind of dovetails into a lot of questions I get around the life of the GPU and how we think about the life of the GPU. We talked about this in our last earnings. The average selling price for an A100, H100, H200, and an L40 all increased quarter-over-quarter for us last quarter.
We are selling H100s at a price higher than what we were selling them two to three years ago when they first came out. That is sustained demand that you're seeing for serving workloads. Another important factor in this inference growth that we've seen, particularly over the last couple of quarters, is that it's not one size fits all. Customers have models that have different usage. Different infrastructure fits their needs. They're very happy to run those on the infrastructure where it is best supported, which oftentimes includes A100s, H100s, L40S, all of them. This is a sustained demand signal that we continue to get from the market, and it's only accelerating at this moment.
The stickiness of these applications are, I would assume, they're going to stay inside your platform for a long time. I don't think they're portable as easily as maybe some have made out to be. The question we get is the durability long term. Once we build this infrastructure, then how do we ensure that they stay at CoreWeave? How do we ensure, what's the next layer of value creation for you?
Unlike any other vendors, we don't believe in customer lock-ins or anything like that. What we believe in is the quality of the platform and the quality of the service we deliver. What we are seeing repeatedly from our customers is customers who have the first contracts kind of expired with us, are coming back to us for not just renewing those contracts, but buying incremental extra capacity from us. That is what gives us confidence in our ability to serve these customers and these workloads to stay on our platform for a longer period of time. It's the quality, performance of the infrastructure, and the software platform that we deliver that actually keeps the customers on our platform over an extended period of time.
That's the foundation of the business that we've built, and we continue to believe as we add more services incrementally on our portfolio, customers are adopting those services. As I talked about, like storage, CPU, networking, software stack, all of those are growing rapidly within our portfolio because customers come here for the GPU infrastructure, but are finding other services that we now offer to be increasingly of value and are adopting CoreWeave as a full-stack cloud platform for their AI workloads.
I know that the software business is still tiny, this could be pretty exciting.
There are two elements of it
two to three years out. Maybe it's not this year or next year, but when you look out longer term, or should I temper my enthusiasm? I look at this and I'm like, Could this become the next AWS?
Absolutely. Which is where we are heading. If you look at from our growth statistics and the revenue backlog, which approached $100 billion last quarter, is where we are approaching.
I think you had Wayne Gretzky 99. Is that right, 99?
99.4.
Yeah. Wayne Gretzky.
So-
We have some AI images for you on that, but we'll show those later.
From a software perspective, there are two elements of it. One is what you would call as managed software within our portfolio, from Weave to our inference platform and so on and so forth. The second element of this is what we call CoreWeave Omni, which is monetizing the CoreWeave AI infrastructure stack in GPUs and in data centers which are not owned by CoreWeave. Effectively, for sovereigns or for other international entities which want to own the infrastructure, or for any enterprise entities who want to own the infrastructure, they can own the infrastructure, but can be supported by the CoreWeave stack, which we call CoreWeave Omni. Both of them are very exciting, incredible opportunities for us to increase our software footprint, which to your point, we do expect to be a meaningful size of our portfolio over the coming years.
They're not near-term opportunities because they're going to take time to scale relative to the existing business we have. We talked about exiting this year at $18 billion-$19 billion ARR, next year north of $30 billion ARR. This is already a sizable business. For the software piece to get to a sizable percentage of this business will take a few years, but we're incredibly excited about the margin accretive opportunity to build our software portfolio.
I think investors push back on me and say, "Well, I feel like they keep kicking the can down the road on margin and profitability," but there's nothing that's changing in your view or this.
On a structural basis
I don't know how you think about your long-term margin targets.
We feel incredibly comfortable with the combination of the factors that we described on the scaling element, as well as the contract economics underlying them, as well as the incremental margin accretive portfolio that we are building, that all of them contributing, we feel incredibly comfortable in the long-term margin trajectory of our business. Where it's interesting is the growth profile. Because we are growing so rapidly and we do incur certain cost as an upfront cost before the infrastructure comes online and we start revenue generating, what we call is a denominator problem. Our base is so small relative to the incremental capacity that we are bringing on, aka growth is so high, that it actually has an impact on near-term margins.
As we talked about, we are going to inflect in the second half of this year based on our projections in terms of how we can see the power coming online and the underlying unit economics. We feel incredibly comfortable in the near-term trajectory that we've described, as well as the long-term trajectory of the business that we hope to and wish to accomplish, and we have a game plan around that.
We've asked other execs today, you get the magic wand to take one or two constraints away. For you, is it power? Is it something in the supply chain? Is it the people? What are the constraints for you right now?
This industry's rapidly evolving. If you would have asked anybody two years ago, two and a half, three years ago, they would have said chips is the biggest constraint. Today, if you look at that, power and shell capacity is the biggest constraint, which involves a whole component of supply chain, including human supply chain. There aren't enough electricians, plumbers to do this kind of stuff that is needed to build the infrastructure. Tomorrow, you're already seeing some component kind of shortage come across. Memory is a great example that's happening in the market right now. I do believe that these supply constraints are going to continue to shift across the industry because we are in a structured supply constraint environment, which is not designed to support the customer demand that's happening right now. This is where it's interesting from a CoreWeave perspective.
This is the only environment we know of. CoreWeave was born out in an environment where supply constraints were always present, and we had to navigate our way through each one of those. We feel incredibly comfortable with our operating abilities as well as the relationships that we've formed across the stack and our own capabilities that we've built in the company that allow us to manage these supply constraints.
You have three and a half gigs under contracted power, and your goal is to get to eight of active power by 2030. The question is the shift towards self-built sites and what's going to happen to get to those targets.
Absolutely. Look, initially, we leased a lot of our capacity because it allowed us for two things. Number one is speed of execution, because we wanted to get power online faster, quicker, because power today is more valuable than power in 2030. It allowed us to kind of get to a certain amount of scale rapidly and quickly. At the same point of time, balance CapEx versus OpEx for us as we grew from a small base of a company. As we look into the future, two things are going to be incredibly important for us. Number one is operational control about certain of these sites so that it allows us to innovate faster. You talked about how do you keep ahead in the innovation curve, is having larger operational control over some of these variables that exist in our industry.
Second is through vertical integration, recapturing some of the margin in that layer. As we think about building our portfolio, it'll still be a healthy mix of lease versus self-built, but you will see an increasingly larger percentage of our self-built coming across over the next several years. We're excited to bring our first self-built site online in the next few quarters this year, and we're excited about the next following site coming online in 2027. You'll see an increasingly larger fraction of our portfolio become self-built, but we will continue to manage a healthy balance across both leased as well as self-built sites.
A lot of private players emerging, lookalikes. How would you describe the environment?
Look, I think the key three differentiating factors that I talked about were, you need scale in this business. The way you have to approach scale is you need to have those three variables working for you. You need to have your technology stack working for you. You need to have an operational ability to execute against the revenue backlog and the contracts and the power that you generate. This is not simply plugging the GPUs and making them work. It doesn't work that way. Third, you need to have a financing engine that actually supports that growth. Given the demand constraints, supply constraints, I'm not surprised that these new entrants are coming in because if you look at it from a customer perspective, if they can't get what they require as capacity from us because we are pretty much sold out, what are their options?
Their options are either to bet on one of these players and see if it works or to accept defeat today. There's a huge difference between signing a deal and delivering that in a performant infrastructure perspective to an end customer. I'm sure some of these new players would be reasonably successful, and I'm sure some of them would not be. It doesn't alter anything from us in terms of our execution map.
You get to see the inside. I think Nick's here, too. Anytime we talk to Nick, the energy level's off the charts. What are you seeing that we can't see? If you describe the industry versus just us looking from the outside, what do we miss that we can't see?
I think that three somewhat interrelated things. Number one is a megawatt of power today and delivery today is way more valuable than that in 2030, 2031. People are not factoring in the nearness of what the delivery looks like for a lot of other players. Similar to that is when you think about signing a contract is one thing and delivering against that contract is another. There's a lot of, to your point, new players who've signed customer contracts because from a customer perspective, it's a bet that they're taking that I'm going to sign with five of them, maybe one or two of them deliver. There's a huge difference between signing a customer contract and delivering.
The third is having a discipline of a financing structure that allows you to continuously scale this business for the rapid growth that it requires is not something that has been replicated. We've been using GPU-backed financing for almost four years, or pretty much since our existence in the company, and nobody's been able to replicate that in the market because the combination of the quality of the customer contracts, our ability to execute, and the longevity of the ability for that ability to execute across the board just hasn't been there in the market. I think these three factors put together are not necessarily well understood by the market today. Delivering GPUs and scaling them is super hard.
We could go for another hour, but thanks for the time. We got to jump to our next presenter, but thanks for coming.
Thank you so much for having me.