CoreWeave Inks $14b Meta Deal, Highlighting AI Demand

Bloomberg

Watch CoreWeave Inks $14b Meta Deal, Highlighting AI Demand - Bloomberg on Dailymotion

Transcript

00:00It does sound like another big deal with what feels like it's got a big window on visibility

00:05for CoreWeave. Lay it out for us. Yeah, CoreWeave has 33 data centers, mostly in the United States,

00:12some in Europe, and they are all humming and they're all being used up. And honestly,

00:17that answer in the context of the Meta News today was a bit of a cop-out because they are supply

00:22constrained. In other words, they have all these deals on paper that extend over a long period of

00:26time. In Meta's case, the deal that we broke the story on today goes through to 2031 with an option

00:32to extend to 2032. It's not yet real. There isn't a data center among those 33 that's going to,

00:39from today, run Meta inference or training workloads. But the thing to understand is that CoreWeave's

00:45magnificent run since its IPO, which Tim alluded to, if there was one piece of downward pressure

00:51on the stock that maybe was holding it back a bit, it was this idea that it only had one customer,

00:55which was Microsoft, accounting for 70% of revenues. And the simple answer to the move in

01:01the shares today is the market saying, good, we like to see diversity in your customer base,

01:06even if it'll be a few years that this stuff comes online.

01:09Why would a customer go with CoreWeave rather than another provider?

01:13Because compute capacity, whether you're talking about literally the building of a data center

01:17or changes in generations of chip, takes many years and billions of dollars to come to fruition.

01:24The way that Michael explains it is they're not going to do a deal where they don't have some

01:30idea of which data center or even some idea of like the energy supply for it, right?

01:34But the reality is this is probably two years away from Meta running these workloads. And at the same

01:39time, Meta is also trying to build its own data centers. It's just an acknowledgement that they

01:43will need more, more than anyone can do at any given time. And the thing you need to understand

01:49about CoreWeave that puts this all into one place is that largely what Microsoft does with CoreWeave

01:56is give it the workloads it can't handle itself. Microsoft has many of its own data centers.

02:02There's just too much.

02:03It's kind of an outsourcing.

02:04It's an outsourcing.

02:04Or it is an outsourcing.

02:05Exactly so.

02:06One of the things I want to ask you is this kind of starting to feel a little bit silly

02:10in terms of the spend the build out. It's got this also incestuous feel

02:15of the companies working together, supplying things to one another, benefiting by the partnerships.

02:19It's just starting to feel a little bit close. And some much smarter than me are suggesting that

02:26this doesn't quite feel right. Will the revenues and financial payoff be there after the spend?

02:31This is what Wall Street calls circular financing.

02:34Correct.

02:35And, you know, the greatest example of that is OpenAI and NVIDIA. But it could extend to CoreWeave.

02:41Well, explain the OpenAI and NVIDIA. Because if OpenAI gets money from NVIDIA to then go and buy

02:46from NVIDIA?

02:47Yes.

02:47OpenAI received a $100 billion investment from NVIDIA.

02:54Over a number of years.

02:54It's an equity investment over a number of years, a decade. So it's actually defined by gigawatts

03:00of data center capacity. For each one gigawatt, it gets $10 billion in the form of, you know,

03:05actually, we don't really know. That's the thing. We don't even know. Like, is NVIDIA just going to

03:09write a $10 billion check or is it going to give it chips in lieu of cash? So anyway,

03:13the argument is that OpenAI just takes that capital in whatever form it comes to then buy

03:19$10 billion worth of NVIDIA chips.

03:20Right.

03:21So the critics and the bears would say...

03:24Wait. Shell game? Oh, no. I'm not going to say that.

03:26As I said, some smarter than me have questioned this. And I'm just trying to be smart.

03:31You know, like, on X, people have used the phrase Ponzi scheme. So, like, I understand that.

03:38At the same time, Tim and I are playing around with ChatGPT...

03:42Don't bring me into this.

03:42...using these tools and are kind of blown away.

03:45The bare case is that the big picture forecast that the market or industry for AI accelerators

03:52or GPUs, before you take into account electricity supply, concrete labor to build data centers

03:57is $4 trillion. That number is being artificially inflated by the circular financing mechanism.

04:03The counterargument to that is, well, there are plenty of software companies making a ton

04:06of money, higher margin revenue because of the investments they made in the last two or

04:10three years.

04:10We see that in the earnings cycle.

04:11Yeah. Start looking at debt because a lot of this is underpinned by debt. And that's a

04:17little bit worrying to some in the market. Others are completely...

04:20These neocloud companies, right?

04:21There's a lot of debt involved.

04:22Yes. Correct. And also, you know, like collateralized loan offerings, they need issuance. M&A

04:31needs issuance. But I would start to start tracking it on your fingers or, if you're lucky,

04:35use your Bloomberg terminal.

04:37I was going to say, you might have started using your toes because it's getting so big

04:40at this point, too. The flip side of that, though, Ed, is that there are those out there

04:45who argue that the productivity increases that we're going to see from these investments

04:48and the innovation that we're going to see from these investments. That's going to be

04:53huge.

04:53Yeah. But like, oh, God, I've missed you guys so much. Do you remember like two years ago?

04:57And I know it's hard to remember last week. I would talk about the money machine.

04:59Yeah.

05:00So like productivity and stuff, just like customers that have spent some capital expenditure or

05:06they've made some kind of licensing deal with OpenAI or whatever it is, what is their top

05:12line growth and bottom line growth been in the last two years? Is that evidence that they

05:16have had revenues in new areas? Some cases, the answer is yes. Some cases, it's no.

05:22So I don't, you know, I'm trying to think of an example like Salesforce, like what does

05:27Salesforce even do? But, you know, like these are where you should look for evidence of that.

Category

Transcript

Be the first to comment

Recommended