Payback on AI Assets Proving to Be Solid: Crawford

Bloomberg

Watch Payback on AI Assets Proving to Be Solid: Crawford - Bloomberg on Dailymotion

Transcript

00:00I want to start off with some of the comments that the Corey Weave CEO made on Bloomberg

00:03television earlier today. You know, whatever was going on with that one single contract,

00:07he's talked about this idea, this gap between the computing demand and at least for what right now

00:12is not enough supply. Are you seeing something similar? Yeah, I think you see it all over.

00:17There's another company called Nebius that reported this morning. And what's surprising

00:21is that they're actually seeing an increase in pricing for their GPUs because there's simply

00:26not enough GPUs. So people are willing to increase the pricing per hour that they're willing to

00:32contract for that GPU. That is incredibly unusual if we were in a supply glut of compute. You heard

00:39it from Lisa Sue today. She's having her analyst day today and she said there's insatiable demand

00:44for compute. She reiterated a trillion dollar TAM for AMD. You heard it through earnings season with

00:52all the hyperscalers that, you know, they don't have enough capacity. And in fact, their numbers

00:56on their cloud businesses would have been higher had they had the capacity. Microsoft said that they

01:01will be short through 26. So, you know, I think there's a lot of different signals that the compute

01:08capacity, the compute capacity is not meeting the demand signals. So I am curious. I mean, it's fair as to

01:15why people would talk about a bubble or at least would question it. But I'm sort of drawn to some comments

01:19that Chuck Cloud, the famed CIO over at Merrill Lynch back in the 80s and 90s, was kind of talking

01:24about the parallels between now and what we saw in the dot-com bubble. And his question, his question

01:31that he posed was basically that everyone is asking the wrong question about AI and more importantly,

01:37AI valuations. And I am curious as to what is the right question to be asking right now?

01:41Right. I definitively don't think that the right question is, is there enough demand? I think the

01:48right question is probably, can we finance that demand? And are we going to, like, try and build

01:55something that's too big, too fast? But, you know. Do you think we're doing that? Not yet. And that's

02:03the thing. Like, right now, we've just started this build. I mean, this year we're spending $380

02:08billion on the hyperscale with the hyperscalers spending $380 billion. Next year will be $550

02:12billion. You know, I think as enterprises and sovereigns get involved with this, we can actually

02:19start to approach these, like, trillion-dollar numbers because the build that's happening is

02:24not just an AI build. You're taking all of this old infrastructure and converting it into accelerated

02:30compute, which is far more economic to run compute that way. So, you know, I think

02:38people misinterpret the stocks going up as a bubble. However, without realizing that the demand

02:47signals are, in fact, much more powerful than the demand signals were in the 2000 bubble.

02:52I want to go to something you just said. Can we finance that demand? That sounds like the debt

02:56markets have a really big role to play. And there have been some concerns brewing in the debt markets,

03:01particularly with the collapses of first brands and Tricolor. And the credit market supposedly always

03:07leads. So what does that tell you about how things are shaping up? Yeah, I mean, even Jamie Dimon said

03:12something, well, there's not only one cockroach, right? So I think I think when it comes to the debt

03:19markets are complex. And what what I have been worried about is that is the impact on the debt markets

03:26for from the private credit side, which, you know, they may have underwritten assets that were not

03:33necessarily the best assets. Does it actually constrain the liquidity that they have to fund,

03:38you know, some of what we're seeing today on the AI side that is independent of AI. Now, I will tell you

03:44the bull case on that. If they can't fund it and we actually can't do this kind of financing, you actually

03:49are going to be in supply demand constraint for longer. So it sounds like the 20 billion, 30 billion dollar

03:55corporate bond sales that we saw from the likes of Meta and Alphabet are going to be a more common occurrence.

03:59That's something that we should get used to. Yeah, I think so. I think that they are going to be

04:04flexing their balance sheets. They haven't done this before. And what we have to watch for is to

04:10make sure that the ROI that they're seeing from these investments is actually a positive one. So,

04:16you know, you think about Microsoft, I think if you look at Microsoft's ROI on the investments in

04:23their own cloud. So into Azure, I mean, we think that they get paid back within 12 to 18 months.

04:31It is highly, highly positive and a creative investment for them today. And I think that is

04:37the little appreciated fact that the payback periods on on some of these assets is actually quite,

04:45quite good. So as you're trying to select, I guess, either new stocks to get into or you're evaluating

04:52your current portfolio of sort of AI slash tech plays. What is kind of one of the main criterias

04:58that you look for right now to sort of make that decision as to whether it's going to be a good

05:02buyer? That's a great question. I think, you know, we had actually, you know, we've been in this space

05:09and investing in this space for two years. You know, I would say, let me let me go back in time a

05:14little bit. So, you know, a company like Nebius, it was the opportunity to go from having almost no

05:22revenue to a very large amount of revenue and being the technological leader. That's what attracted

05:29us to a stock like Nebius and somewhat under underappreciated when we first bought it. And so

05:36there's aspects of this market that people are still not investing in where, you know, where the where

05:45the opportunity set is still still large. Yeah. And maybe, you know, it's it's a fraction of their

05:52business. One final question. And this kind of has to do, again, trying to draw the parallels, if any,

05:56between then and now. We're going to get Nvidia reporting next week. But tomorrow we get Cisco, which, of course,

06:01was kind of the Nvidia of its day to a certain extent. And I know there have been a lot of people

06:06talk about all the big infrastructure build out we saw with regards to telecom back in those days

06:10and Internet infrastructure. Is that unfair to sort of look at what's happening with Nebius and CoreWeave

06:16and Nvidia trying to sell all the create all these GPUs? Is that just not one in the same? It is absolutely

06:22not one of the same. And everyone wants to point to the valuations. I look at it completely differently.

06:27If you look in 2000, we had dreamed of the world that we have today in terms of

06:31having a ubiquitous network and that ubiquitous network enabled, you know, Uber and streaming

06:37video and the cloud. However, if you remember using the Internet in 2000, if you could just

06:43take yourself back to that moment in time, we were on dial up in the late 90s. We were on

06:48dial up and the experience of using the Internet was wholeheartedly unproductive. It was it was

06:55yeah, it was it was just awful. Yeah. And so we had dreamed about our world today without

07:01a ubiquitous network. We put into into place the fiber, the Cisco, you know, networking equipment

07:07to enable our world today. But we didn't have last mile. We didn't really have mobile devices that

07:15that we could use the network as much as we do today. So the infrastructure wasn't in place today.

07:21All of that is in place. The only thing that we've been missing to deploy artificial intelligence

07:28has been compute and NVIDIA democratized that compute by introducing accelerated computation.

07:36And in doing so, it's almost like you've been waiting for gas and now you've got the gas

07:41gas and the car can go.

Category

Transcript

Be the first to comment

Recommended