00:00Every time we talk, I mean, I'm always curious about these kind of, you know, these neoclouds
00:04and whether kind of these sort of ancillary players in the hyperscaler space are actually
00:09legit. When you look at these numbers today, does it give you a little bit more sense of
00:13proof of life or what? It's incrementally positive. There is still a really big question
00:19about whether we even need neoclouds, whether they're a good business. The base of it is,
00:25are they getting a good enough return on investment to pay for the very expensive capital
00:30that they have to borrow? And I think that what investors are seeing today, they took a small
00:34step in the right direction. The operating margins, which were only 1% last quarter, are now 5%. So
00:41that's not a lot. That still means a 1% return on invested capital, but it's a step in the
00:47right
00:48direction. So a step in the right direction, is it a durable step, a long-term step, Gil? Or is
00:54this
00:54just, you know, they're taking advantage of what right now is just kind of a crazy market?
00:58Let's put it in perspective. So if I'm saying they're getting about a 1% return on assets,
01:04their borrowing cost has actually just gone up. It was closer to 8% and now it's close to 10%.
01:12So you're borrowing at 10% to get a 1% return. That's not good news. So again, they have
01:18to improve
01:19the returns very significantly. And they have to reduce the cost of capital, which recently has
01:24actually gone the wrong way. That's why we're questioning the whole category is that this is
01:29like borrowing on margin to buy treasuries, borrowing a 10% to buy a 5% return isn't a good
01:36idea. And for these guys, again, they're still getting a 1% return. So until, and they're at a 10
01:42billion revenue run rate scale. So it's not like there's a tiny upstart. And they're still generating
01:48pretty low returns on that slightly better than before, but still not good enough to justify their
01:54existence.
01:55Well, what do you make of kind of the splashy announcement that Jensen Wong made yesterday,
01:59the idea of trying to raise what is effectively a $500 billion financing pool, if you will, for
02:06presumably those folks who are going to be buying its chips. But the idea is that some of these
02:10companies that are dealing with a higher cost of capital or maybe no access at all would
02:14potentially have access. Does that not help CoreWeave? Or is this just basically you're still
02:19paying the same cost of capital, just maybe to a different set of a separate set of lenders?
02:25That is exactly what Jensen's trying to do. He's seeing CoreWeave have to borrow a 10%. He said,
02:31I need to do something about this. And so if I guarantee some of these loans, I can reduce the
02:37cost
02:37of capital for companies like CoreWeave. So he is trying to help them. The price is, of course,
02:43that you can only buy Nvidia chips. So he's not doing it out of the generosity of his heart.
02:48No.
02:48He's doing it so he can see the whole ecosystem with capital that's tied to buying Nvidia chips,
02:55not Broadcom chips, not AMD chips, not ones made by Intel, but Nvidia chips. And so he's getting
03:01something for that backstop. That backstop, though, is a financial commitment. If these data centers
03:07fail, the loan goes back to Nvidia. So there is a cost, at least down the line, for them making
03:14this
03:14type of guarantee. But for now, they have $60 billion of net cash. They're going to generate
03:20$200 billion in the next 12 months. He needs to do something with that cash and guaranteeing
03:25his customer's loan seems to be a good way for him to make sure that they keep buying Nvidia
03:29and don't buy anything else. So with CoreWeave, I mean, one of the big questions and that you and
03:35a lot of other analysts raised was also this idea of who their customers are and whether it will be
03:40able to diversify. Obviously, at least in the most recent quarter, still heavily dependent on
03:45Microsoft. And it's honestly a couple other hyperscalers. And I guess my question to you is,
03:51why does Microsoft need a CoreWeave? Why does a Meta need a CoreWeave? Why does Alphabet need a
03:57core weave or any of those other neocloud competitors?
04:01So the reason Microsoft has a three-part strategy is that they don't want to build it all
04:07themselves. If they build it all themselves, they're taking a risk that they overbuild.
04:11So they're building some infrastructure, they're leasing other infrastructure, and then they're
04:16leasing compute from companies like CoreWeave and reselling it to their customers. They're the ones
04:22that have the customers. And that's why they want to make sure they can meet demand without
04:28overinvesting. And CoreWeave is a tool to do that. The problem with being a tool is that once you use
04:33the tool for, let's say, the next three to five years, by then, Microsoft believes they will have
04:38enough capacity to serve all their customers. Are they going to need CoreWeave? Is Meta going to need
04:43CoreWeave when they're done building out their own infrastructure? So this is great for Microsoft and
04:48Meta. For CoreWeave shareholders, there may be a ticking time clock.
Comments