FULL TRANSCRIPT
Slava Rubin (00:00)
In this episode of Smart Humans, we continue our pre-IPO series talking about CoreWeave. What is the right price to enter this GPU provider for the cloud specifically focused on AI? Are they the perfect pure play in this world of AI? Or are they going to be debt-ridden and are going to have a hard time to be able to pay off all their interest? Is this an opportunity to get in early? Or should you be scared about how the fundamentals are working out?
Will the growth rates support their entry price and when will it be going public?
Slava Rubin (01:00)
Hello and welcome to the latest in our pre IPO series. My name is Slava. I'm one of the founders here at Vincent and with me is Jan Erik from Sacra We're very excited to be talking today with about CoreWeave. So this is the latest in our pre IPO briefing series. We've covered all kinds of interesting companies already from SpaceX to Anthropic, ⁓ OpenAI and others. And today's conversation is going to be discussing
this powerhouse in the cloud computing infrastructure space specifically focused on AI. And they've already mentioned that they're looking to go public very soon at potentially a $35 billion valuation. Jan-Erik, welcome to the conversation. Awesome. Thanks for having me. Absolutely. So as always, we have our compliance department. ⁓ Always requires us to give you the disclaimer. So nothing in this presentation should be construed as an offer to sell securities.
or solicitation of an offer to buy securities. All investments involve risk and the possibility of loss, including loss of principal. And neither past performance nor forward looking information is a guarantee of future results. So with that said, think, Jan-Erik, how about you start by giving us the high level on what is CoreWeave? I mean, the name is interesting. Not everybody's really heard about it. And all of a sudden, it's the topic of the day. So can you give some context? What is and who is CoreWeave? Yeah, I CoreWeave is a cool story.
This company basically started off doing ⁓ crypto mining in 2017, know, in New Jersey, just off the river, kind of a bunch of ex-finance guys and crypto traders who started stocking up GPUs in order to mine crypto. And as the crypto market went down, they actually took advantage of that sort of, ⁓ you know, bear market, bought up a lot of cheap GPUs and in 2019 started building ⁓
basically for this emerging trend of AI. Three years before sort of chat GPT would make it super relevant. And yeah, obviously, know, lots of people have heard about their growth, which has been very rapid, 16 million revenue, 2022, 229 million, 2023, 1.9 billion last year projecting 8 billion for next year. So yeah, they've been one of the standout kind of AI companies. And one of the few kind of
really prominent startups at this infrastructure layer of AI. And then product wise, really the key ⁓ thing besides the fact they have a massive stockpile of GPUs that they acquire through a preferential relationship with NVIDIA is that they developed the software infrastructure to allow data centers to run with lots of these GPUs. And it was purpose built for
for AI specifically starting in 2019. So that really positioned them really well when companies started looking for huge sources of GPUs for training and inference. So it's not just your sort of typical AWS GCP server farm. It was sort of designed for this use case from the ground up. And fundraising wise, yeah, I mean, like you said, 35 billion is what they're targeting in their IPO.
raised 1.5 billion in equity, more in debt, almost 8 billion. Yeah, secondary market right now, they're right around that 35 billion, a little bit under, so like 32, 33, which is a big premium on the last primary of 23 billion. So yeah, that's sort of the quick overview. Super great context. So you mentioned it's not exactly AWS. So for those that are familiar with cloud computing and such, there's
There's AWS with Microsoft, there's Azure, sorry, AWS with Amazon, there's Azure with Microsoft, and there's other places to go as well, ⁓ even Oracle, et cetera. What makes the CoreWeave experience and product unique? Why is it able to compete with those and stand out to have those crazy growth rates? Yeah, so like I said, they built for AI, so where AWS, Azure, and Google,
were basically built for ⁓ web hosting primarily, which was the big use case in the early 2000s and databases. I was just mentioning like what makes it different for the cloud compute versus CoreWeave against the other players that were there before. Yeah. So like I was saying, they sort of built for AI. they didn't, a lot of the sort of performance like
efficiency driven technical aspects of building this kind of business that AWS and Azure did early on, where we didn't do, this sort of design for AI. So there is this like, you know, they say 20 % better performance on AI models in terms of like serving the speed of inference. So that's key. They've built software around making it easier for companies to serve the AI models. That's all important. I think the other really big component is that
they have had ⁓ the best access to NVIDIA GPUs of any ⁓ company in the world, probably, ⁓ because NVIDIA has made them sort of a preferred, ⁓ most favored nation provider of GPU cloud services. So while AWS, Google and Azure have been sort of, on the one hand begging NVIDIA for more GPUs and on the ⁓ other hand trying to sort of develop their own
⁓ alternative to GPUs, CoreWeave has had this ⁓ very mutually beneficial relationship with NVIDIA. So is that the main difference? The actual chips, meaning that CoreWeave has these chips from NVIDIA and the other players don't? Said another way, if the other players get the same chips as CoreWeave,
does CoreWeave's product get commoditized or is there something about the way CoreWeave is set up with their software and such that they have a differentiated offering? Like if I'm Walmart or Boeing as a customer, how do I think about that? I think for, until today, the big advantage has been, you know, the sort of access to GPUs and the just supply constraints, know, AWS Azure haven't been able to get the same, ⁓ the quantity, they've been able to get some, but it's been highly kind of throttled. ⁓
But I think over time they've really built out the software side of the business. So there is genuine appeal to that of having a sort of AI native data center with GPUs, but a world where the latest Nvidia chips are available, sort of commoditized, anyone can buy whatever quantity they want would definitely weaken ⁓ Coreweave's value prop. Although,
That doesn't seem likely anytime soon, given that a lot of the constraints are sort of upstream with semiconductors. And how exactly does CoreWeave make money? mean, their revenue has been skyrocketing. ⁓ So what is their business model? What's their unit economics? Yeah, so the vast majority of their revenue comes from these sort of long-term committed contracts for preserved ⁓ GPUs.
a company like Microsoft, OpenAI, another big AI lab will basically go to CoreWeave and say, we need this many GPUs. And generally, it's like on the larger end, like a five-year deal or longer. customers pay a certain amount whether or not they use all of the capacity.
They also have an on-demand product for more like a startups that want to pay on an hourly basis to use CoreWeave, but that's a very small part of the business from what we can tell. So four or 5%. But yeah, currently they have like 15 billion in sort of unfulfilled ⁓ contracts or remaining performance obligations. So that's the lion's share of the business. So it's interesting. Microsoft is their top customer to date over 60%.
actually have the revenue to talk about client concentration. Why does Microsoft use them so much? Why does Microsoft not have their own ⁓ cloud computing capability that they're using, you know, like Azure, or is it they literally just haven't been able to build it up fast enough? And is that then a competitive threat that they'll do it themselves over time and not go to CoreWeave? Definitely. So a lot of the Microsoft use of CoreWeave comes back to their ⁓ deal with OpenAI.
And so as OpenAI really took off in terms of usage, Microsoft didn't have the compute necessary and they couldn't build out enough compute fast enough, like you said, to fill that demand. So they were using CoreWeave as a stopgap. The CEO is on record a few months ago saying, we rented out, we do a lot of leasing at Microsoft and we leased out some compute from CoreWeave.
⁓ But they are ⁓ working on building their own infrastructure. There's been some conflicting news about Microsoft canceling contracts or scaling back its infrastructure investments. ⁓ But yeah, mean, 60 % of revenue is pretty significant. I think the only sort of like ⁓ devil's advocate view there is that a lot of these industries that are in takeoff mode exhibit these kinds of concentration dynamics. Like if you look at Twilio's IPO,
you know, a while ago, ⁓ Uber was a huge percentage of their revenue, which was cited as a risk at the time. But, you know, the fact is that Uber was just growing so fast, so much faster than everyone else, ⁓ that it inevitably was sort of high concentration, which is similar in ways to Microsoft, OpenAI, ⁓ and CoreWeave. So in the last few weeks, there's been a lot of breaking news about
know, Nvidia's connection to CoreWeave, then there's Microsoft revenue concentration. Are they pulling out of their contract? OpenAI all of a sudden having a bigger commitment, over $11 billion of committed contracts with CoreWeave. Supposedly they're getting 1 % of company. ⁓ Nvidia has, I believe, 6 % of the company already as part of loaning them the GPUs. So there seems to be a lot of circular movement of money here.
Can you unpack that a little bit and what's going on? Yeah, definitely. I mean, Nvidia has been a big sort of, ⁓ you know, promoter of CoreWeave. ⁓ I think, you know, you can look at it as somewhat sketchy, you know, that Nvidia is sort of the prime, one of the prime customers and suppliers to CoreWeave and investor. ⁓ I think, you know, that's just sort of depends on how you feel about that. I think the sort of
underlying reality for me is that ⁓ NVIDIA really wanted to have a player in the sort of GPU rental market or compute market that was not AWS, Google or Microsoft because AWS, Google, Microsoft all buy NVIDIA chips but they're also all developing their own chips. Google has TPUs, Amazon has Trinium chips. And so they don't want to be fully reliant on
these cloud giants that are trying to make them obsolete. ⁓ yeah, it seems to me like they promoted CoreWeave's ⁓ existence, helped make it a major player ⁓ in the space. And I think it was self-interested and it helped CoreWeave succeed and get a moat. ⁓ And then with Microsoft OpenAI, yeah, it's somewhat complicated because it seemed like ⁓ Microsoft deal with CoreWeave originally was built around this
demand from OpenAI. ⁓ Now the news is that OpenAI is doing a deal with CoreWeave. So it makes sense. It sort of seems like the same demand is being served by CoreWeave, whether or not, ⁓ let's say the name on the contract sort of changes to OpenAI. ⁓ I think it's still kind of representative of the incredible demand there's been for OpenAI, ChatGPT in particular. You didn't say this, but
is CoreWeave just a business unit of NVIDIA? I mean, they're definitely like a cousin or a nephew of NVIDIA. And that could change in the future, but I do think today it's still quite firmly within the NVIDIA ecosystem. I'm not sure it would exist in the form it exists, if not for NVIDIA's help and partnership there.
So I think, yeah, I think that's fair to say. What do you think the near-term future holds? I mean, we're talking about having it go public very soon. The talk is $35 billion. Obviously, you have to think about what the next six months, year, two years, three years looks like. What are the growth opportunities here? Yeah. I mean, obviously, the revenue growth has been huge. know, they're seeing their top customers increasing their spending.
they're expanding geographically. So they have 20 or 30 data centers and they're expanding to the UK, Spain, Sweden. There's plans for Canada and Norway to sort of get big into sort of Europe. ⁓ And there's also no strong indication that compute needs are going down. ⁓ I think a variety of things sort of prove this out.
One big one is the sort of emergence of ⁓ these reasoning models, which have become ⁓ really the primary form of LLM that's coming out now. ⁓ Over the last six months, virtually every big AI model release has been a ⁓ reasoning model, which are ⁓ more compute intensive ⁓ to train and serve. So I think the secular trend is sort of in the direction of
of more compute being needed, ⁓ even if you take into account on the other hand DeepSeq ⁓ and other sort of open source models that are doing what ChatGPT or AnthropicSclod are doing for cheaper. I think there's the angle on that that actually is sort of ⁓ indicative of more compute being needed rather than less. I mean, I the market itself is really, you know, ⁓
There's no evidence that that's slowing down, which is really good for core weave. And do you think that they'll stand alone as kind of having this AI focused capability, or do you think the others will catch up? And what's that timeline look like? Yeah, so there are other players, ⁓ you know, they sort of were the first in this market, but this sort of Neo cloud market has grown quite large. There's companies, ⁓ you know, there's like sovereigns in China and elsewhere that are doing this.
⁓ And then there's kind of a differentiation on like customer segment. So ⁓ for example, we have ⁓ Lambda Labs, Together AI, and a few others that are really focused on landing more of startup ⁓ kind of mid-market companies. they generally pair, you know, they have far less sort of GPU capacity, ⁓ but what they offer in return is kind of,
more ⁓ software around it, more of a developer experience to try and pull in ⁓ software engineers. The other sort of big player in the US to look at is Caruso Energy, who also was a Bitcoin miner about a decade ago, pivoted into GPUs. Their approach is much more based around clean energy. So they tap into ⁓
know, oil fields that have trapped natural gas that can't be used for anything. So it generally just gets burned off and they sort of use that to power data centers. And they have some big deals as well with some big enterprise companies, but still not the scale of CoreWeave right now, but like OpenAI has deals with them. So we are seeing in short, ⁓ other companies getting in on this market. ⁓ But CoreWeave has a lead and there's no,
indication that the lead is becoming smaller, right? Because they keep building out more more data centers, ⁓ both building out their own and renting. So they appear to have a lead, you know, that I don't see why it would, you know, they would fall behind any other startup. You know, we have the revenue chart up right now. And as you mentioned, the revenue went up skyrocket growth to practically 2 billion in 24 and the projection is nearly 8 billion in 25.
What is there any visibility that we have on the 8 billion and 25 now that we're coming towards the end of Q1? I think it's mostly just these like big deals like with like with Microsoft slash open AI. You know, lot of that revenue is yet to be recognized. So that's why it's on the that's why it's sort of part of the 15 billion of of remaining obligations. And so I believe we're expecting a lot of that revenue to hit next year. What's the there's a lot of positives.
around CoreWeave, what's the biggest concerns you have right now in terms of risks? Yeah. Well, one is sort of the financial engineering piece, which we can talk about more, but running at huge losses right now. And those losses have been cut from last year. ⁓ Sorry, typically we're talking about making $2 billion last year, but basically losing a billion. So basically, we're at 50 % almost.
Yeah, exactly. And that's better than the year before they were losing like 250 % of revenue. ⁓ So it's definitely an improvement. And a lot of that is these massive interest payments on the $8 billion in debt they've taken out, and then also sort of the building out of data centers. So if you look at it very fundamentally, there's a big question around that. And can they become profitable? ⁓
mean, granted gross margin is like 75%. So the business should be fundamentally profitable, but there's a sort of big gray area that hinges on these ⁓ debt agreements and the GPUs and whether they can sort of make the math on this work long-term, ⁓ especially with GPUs being effective life of three to five years or so. ⁓
So that's my, yeah, for me, that's a concern because I'm not sure if they sort of ⁓ built out the models ⁓ exactly right. And then, I do think ⁓ there's a risk here with Microsoft OpenAI. Obviously ⁓ I said before that Twilio had customer concentration risk. ⁓ That doesn't mean CoreWeave is necessarily a great sign. ⁓ So there is a risk there with both OpenAI
or Microsoft, whoever it may be as their biggest customer, finding compute elsewhere. Just to double click on the debt. So they owe like almost $8 billion of debt, getting something like over $300 million of interest a year. And they owe that by 2029, which is a long way away, but not that long way away when you start thinking about it. Because all of sudden we're in 25. Next thing you know, we're in 26.
they lost a billion last year, are they gonna make money this year? I'm guessing not. Or are they projecting they're gonna make money this year? I don't think so. I don't think they've spoken on whether not they think they're gonna make money. But I would be surprised given, you know, there's still infrastructure to build out here. Yeah, then kind of gets us to, you know, add in the interest, maybe $9 billion worth of money to pay in three years. Yeah.
It's a hard one and I think people have compared it to WeWork ⁓ signing these really long-term leases ⁓ and how that ended for them. ⁓ I think obviously different market, but it does sort of make the business structurally dependent on the demand for their compute continuing to grow. Just to counter that argument myself though.
If they do grow at this level and all of sudden go from 2 billion to 8 billion of revenue and then 8 billion to who knows what, let's call it 15 or something, you've got to believe that at some point there's maybe some free cash flow there. And if the market keeps growing, they could start printing, printing, printing money, right? Yeah, exactly. Exactly. Especially because they'll have all the sort of data centers, you know, infrastructure built out already. So I think, yeah, in that case, it's a win. What does the projections look like for you, Bull or Bear?
taking into account kind of the pros and cons we discussed? Yeah, ⁓ you know, Bullcase, I think kind of goes along with what you just said, which is, you know, we get a lot of demand for compute. ⁓ know, CoreWeave is like essential infrastructure for the next, you know, GBT-5, GBT-6, so on. ⁓ And they also diversify sort of the customer base. So, you know, they've got Microsoft, have OpenAI, Anthropic, ⁓ every big...
⁓ AI company that needs more compute to serve the most recent models and putting some percentage of their needs ⁓ on CoreWeave. yeah, that's sort of bull case model. We had them at like 17 billion revenue ⁓ and let's say nine X multiple or like a $150 billion valuation. ⁓ I think the
The flip side is kind of just, well, what if compute needs for CoreWeave don't rise that high? What if Microsoft reduces reliance? ⁓ What if OpenAI, which is working on its own chips as well, ⁓ reduces its reliance? AWS, Google, Azure, ⁓ which have very strong cash flows, respective to CoreWeave, and sort of take up more of this market.
I think that's where, yeah, you get to a world where revenue doesn't grow ⁓ fast enough to satisfy ⁓ what they need. Where do you come up with the, like if they're doing really well, why the 9X multiple? Just a big, at that point, really big company. I think looking at some of the comps, ⁓ like right now we have, ⁓ Nvidia right now obviously have a huge. ⁓
revenue year, 130 billion value that 22 X. But if you look at something like IBM, know, IBM at 16 billion, so roughly that like 17 billion, you know, their value much lower. yeah, nine X could be, could be conservative, but I'm also kind of pricing in the, you know, you could argue that the current pricing of the company is a little bit conservative. If you, if you look at $35 billion valuation on 2 billion in revenue, it's a little bit low relative to like other.
sort of AI companies and Neo clouds that are gone public. And then obviously if you come in at 35 billion today and then it's worth 20 billion in 2029, that doesn't sound very good. No, that yeah, I would agree. I it's not necessarily one of these, it's not a company like OpenAI where if you invest today, you're looking at a thousand X potentially in the bull case in the next few years.
Corweave's a little further along. It's a little bit more of a commodity space, just in terms of what they're doing. So I don't think it's quite as big of a potential win as other companies. And then there is sort of a downside case. Yeah, I think the base for me is kind of like a small win, like 70 billion. Got it. So let's talk about price, because typically we talk about this and we're not sure if SpaceX is going to IPO.
or when OpenAI is going to IPO, all the messages are that CoreWeave is going to open up for trading soon. We don't know when exactly. There's no official date. But let me start with a provocative first question. Are they going to IPO in the next month? I think so. I keep hearing that they might be delayed, but I haven't seen that officially reported. So I'm going to stick by what I've read before that it's going to happen by the end of March. But I could be wrong.
For the sake of having a conspiracy theory conversation, ⁓ it seems like the IPO was thought about while ⁓ the jump up in the early part of this quarter, where things topped off into Nvidia being worth the time in the market doing great. It seems like the drop in the last month might be somewhat concerning for CoreWeave. So I'm not saying I know anything.
that they're not an IPO, but I wouldn't be surprised if they held off for a minute. Any thoughts on that? Yeah, I mean, there's been a bunch of these kind of, you know, IPOs having trouble getting lift off this year. I mean, you had Cerebris, which obviously had its own problems, different problems. And obviously, I think with everything that's going on with the administration, I think that's called a lot of consternation as well around going public. So then you add in this
you know, apparently a week before their IPO, the news comes out that their biggest customer is canceling contracts, which they've come out and said that's not true. But I mean, if you had to take three events to make you delay your IPO, mean, that it's been kind of rough sledding for sure. So, yeah. Right, they tried to get some more positive support there with OpenAI saying that they're coming in with big contracts and the open.
getting 1 % of the company signaling that they're getting it for $350 million investment, which equals a $35 billion IPO. So that seems like some, I don't know, some PR there to send some positive messages towards the pricing. So what do you think about the $35 billion IPO price? Let's assume that's true and let's assume that's happening next week. What do think about that? Yeah, like I said, I think it's actually kind of a, it could be considered a little conservative. If you have belief in this business and you think the economics will work out.
and that compute demands going up, then I think it's somewhat cheap. Especially look at Nvidia, like 22X, this company Nebius, another Neo Cloud went public. They're at like a 57X revenue multiple right now. So I I like 35. It's also pretty close to where they're at on the secondary market, which if you go buy ⁓ core weave shares right now on secondary market, they're like a 32.
$33 billion valuation. So I don't think 35 is outrageous at all. And if you had to predict what the valuation would be three years from now, what would you say it would be? Yeah, think probably my sort of personal base case is like that 9 billion, 9 billion number probably 9 to 10. And we could say 6X, but maybe maybe, you know, we'll be
be a little more aggressive and say, and say maybe a 10 X. And so I would say 90 billion ⁓ probably would be my estimation. So, yeah, 2.7 X on where they are now. So you're probably a buyer. Is that right? Yeah. Yeah. Like I was saying before the show, I think, I think I'm a buyer, but I'm a, I'm a, I'm a cautiously optimistic buyer versus, you know, buying a space X or, or open AI personally. Great.
Do you think they're, what are the chances they're not a standalone company in three years? Ooh, nice. I it's, I think it's possible. I hadn't actually thought about it before. Maybe Nvidia. I mean, yeah. I mean, Nvidia is kind of the obvious one, right? Yeah. It could happen. It could happen. Maybe the IPO doesn't happen. And we see that happen instead, but yeah. I think it's a good idea.
It'll be interesting to see what happens. Either this IPO happens and it goes well. And if that happens, it just shows that a cloud providers as a standalone company could be an interesting investment, which then all of a sudden, I think there's going be talk about AWS and Azure, especially AWS start to spin out, right? Is it going to be like their own assets? And that becomes interesting from a competitive perspective. Now on the flip side, if the IPO doesn't do well, and there have some challenges over the next couple of years,
it would not surprise me at all if this is not an independent company. I mean, we really don't know many cloud compute companies as independent, you know, strong public companies. Usually they're now inside of other revenue lines. ⁓ This has been growing fast and furious, you know, with some financial folks ⁓ engineering it this way. Hopefully the growth continues and hopefully the math all works out.
But I do think it's a risky play, ⁓ but the risk could be fun. I mean, it could be a vol traders delight ⁓ with CoreWeave. I kind of see it having similar similarities to the way Bitcoin and Coinbase trade. And what I mean by that is I almost see Coinbase just extra vol on Bitcoin, right?
So typically you can get even more movement because it's a smaller priced asset overall. I kind of see where AI goes and where Nvidia goes is kind of where Horweave is going to go, but at a greater volatility because it's just such a lower price. Any thoughts on that? Yeah, I think if you are excited about trading, know, I think this is one for the retail investors would definitely love, which I've seen a lot on the
On Twitter, on Reddit, people are kind of going crazy about CoreWeave. And this is exactly what you're describing is sort of the thesis. So yeah, I think it could be fun. And that's why I think I'm also ⁓ a cautious, optimistic buyer and ⁓ excited to see what the ride looks like. If the AI market is massive, CoreWeave has a chance to really ride that. I do think it's gonna be hard to just hold your nose and just go ride along the whole way.
⁓ But yeah, people need to decide what they want to do about it. And, ⁓ you know, this is a very interesting player, a little confusing at first, but you really shed a lot of light. Thank you for all the color. Any closing thoughts, Jan-Erik, on ⁓ CoreWeave? Yeah, no, I think it's cool. ⁓ You know, like you said, it's really cool business. And what they've done is really impressive. A lot of financial wizardry combined with sort of being early to a big trend of GPUs. ⁓
So think it's cool business and I agree that it's very much indexed on AI. So if you're optimistic about AI and about it being a very important part of our lives in the future, then I think Core Weave is sort of an obvious play index on it. Perfect. Well, thank you very much for your time and thank you everybody that joined and listening online. You know, we're all waiting to find out when the IPO is going to happen. Will it price at what price? Obviously there's the connections to
OpenAI to Microsoft to Nvidia. And if you're bullish on AI, this is a very interesting player to hold. But of course, make sure you do your diligence. Understand the math, understand the debt, understand the growth expectations. With that, thank you very much for joining.
