Transcription
Okay, let's talk about this concept of circular funding to start with. Why has it created such a fuss amongst market observers, and what is it?
Oh, well, I guess the easiest way to frame it is a company like OpenAI doesn't have the balance sheet of Meta or a Microsoft or Google in terms of the huge air infrastructure buildout that they want to go ahead with. And so far, they relied on Microsoft to consume all the data center capacity they needed for its output and the enterprise business they have. And now they feel the numbers and the demand is getting so big that they are planning ahead. They are looking to three years ahead where they would need probably two or three times more capacity than they are consuming right now.
And for that, first, they did a deal with Oracle, a $300 billion deal that spans almost five years. Then they did this deal with an area, actually to build ten gigawatts of data center capacity. Now, the interesting part to answer your question on circular funding is NVIDIA is spending, you know, $10 billion and giving it to OpenAI for them to add up to ten gigawatts of capacity, and that investment could go up to $100 billion. So every gigawatt involves almost $40 billion of spend. That's a rough calculation. So you can imagine for ten gigawatts, we are talking about $400 billion in spend. Out of that, NVIDIA is going to put $100 billion. So one fourth of that is coming from Nvidia. The rest, OpenAI still has to raise it through a private vehicle or some other type of instrument, but that's the challenge OpenAI has to grapple with and figure out the rest of the funding. And they feel they can do that in conjunction with, I mean, I'm not even talking about the AMD partnership, but that's essentially what they're trying to do, is to build out data centers on their own as opposed to relying on hyperscalers.
And so we're going to get to the whole idea of an AI bubble here shortly. But do these companies have all the money to back it up? Do they have the financing to support these efforts? We're throwing around a lot of big numbers here. And it feels like with the tech space, it doesn't matter what number you throw out. You could find some company that's saying that that's what they're going to spend when we think about CapEx. So you're right, the traditional business used to be asset-light. You needed CapEx spend of maybe $30 to $35 billion if you're a Google or a Meta. And that would suffice for the whole year, and it would stay, you know, around 5 to 10%, give or take. But now it's already, you know, two times larger than that $35 billion. And next year, it's going to cross $100 billion. It's going to be three times larger than existing companies that run data center assets. And the reason for that is the compute has gotten so expensive. Plus, you need more compute for generative AI. So think of traditional search. It needed compute that equated to maybe $0.02 cents per query. Now, with generative AI, that chatbot query is almost ten times more expensive than your traditional search query. So that's where the compute demand is coming from. And these companies have no choice but to invest in infrastructure because users are spending more time. I mean, Google's token usage has grown 50X, so people are consuming a lot of these products. And if the pace continues at this level, then you have no choice but to add more capacity. And that's why the numbers are so big here.
Well, you know, you talking about it seems like it really is just an insatiable appetite for investment into the space right now. We're talking about not just kind of the chips and and the servers, but also the physical space. But when you compare that with the flow of money and investment into the stock, whether it's from investors on the street or companies looking to get in, and it seems like there are a lot of willing players still in the market. Do you think there really is such a mismatch between the flow of money coming out and also the demand for funding to create and further fuel this AI infrastructure situation?
Yeah, everyone, I think now realizes that this is a very long runway in terms of both the build-out of the infrastructure and how it will eventually monetize. I mean, look, there's a lot of upfront investment, and that's where, you know, the hyperscalers CapEx guys, they were willing to invest money with their operating cash flow. They had the balance sheet, the Google, Meta, Microsoft, they had the balance sheet to put their operating cash flow in the form of. But now the numbers are getting even bigger. And so if Google generates $100 billion in operating cash flow, now they're talking about spending more than that $100 billion. And that's where the private guys come in. Everyone feels it's an attractive asset because you can rent it. You can generate a return over three or four years. And that's where I do think there will be a lot more of these deals. I mean, the whole cloud space has come about literally because of this demand and, you know, willingness to rent compute from someone else.
Right. Lots of activity happening in the AI space. Can't stop, won't stop. A lot of bubble allegations, too. Anything that you can point to that would dispute that?
Well, right now, we are still in that phase where the demand far outstrips supply. And to me, the biggest indicator of that is what NVIDIA tells us on their earnings call in terms of their margins, their pricing. I mean, right now, their Hopper chip prices have started to go up because the demand is far outstripping supply. So when you see a trend like that, yes, there will be probably some misallocation of capital that will find out in retrospect that this capital was misallocated in some way. But right now, it's very hard to question the pace of this build-out because there's still that big gap between demand and supply. And until that narrows, it's hard to question why capital is going in this domain, because, I mean, it should.
Well, thank you for mentioning the earnings season, because that's exactly what we want to be asking you about. Obviously, our CapEx was such a big theme last season, especially for NVIDIA. But really, just in general, the Magnificent Seven companies and anyone involved in the space, I think that's still the big question for investors is, is what companies are spending on AI? Does that justify the returns that they're getting now and in the future? What would you say to those investors kind of weighing that to that balance as we head into the next season's earnings?
I mean, look at the year-to-date performance of Mag Seven. The stocks that have outperformed are the ones that have spent their CapEx, that have increased their spending on AI, not the ones that have been conservative. Your Apple and Amazon haven't outperformed, right? Nvidia, the best performing. And I mean, and really is not the one who is putting CapEx dollars, but look at, you know, Google. They, I think so that's where I do think the market is still rewarding companies that are spending on AI. And if you have been conservative and just focused on margins, your multiple has shrunk. And that's the sort of mood we are in. So my feeling is you will see this earnings season also reward companies that show higher revenue growth, even if they have large businesses. If that AI component is accelerating, you will see multiples expanding, and that's where there is scope to be positively surprised.
Are valuations too high, though? I mean, you could argue that, you know, in certain pockets, for maybe the index as a whole. But when you look at individual companies, I go back to Mag Seven. These are wonderful businesses that are investing in AI, which everyone deems, you know, will have high usage. The question is, how will they monetize? And if they can answer that question in terms of, you know, how it monetizes over time, not in the next 12 to 24 months, but over time, I think investors will be happy.
Now, of course, part of the jitters around this, whether AI is in a bubble and in all that, is really this idea of concentration risk. Right. And the fact that more and more gains in the S&P 500 in the U.S. equity market in general are being driven by such a small number of companies. Is that something that investors really need to worry so much about when, as you say, at the moment, demand for these companies' products and services is still far outstripping supply?
Yeah. So, I mean, again, the picks and shovels trade has carried on. Now we are in that phase where the last frontier models are, you know, clear which ones are those and which ones have the lead. And now it's about, you know, how the other software companies adapt in terms of using these models as a distribution layer and making sure they can thrive in the world where AI is also part of the tech stack. So we'll start to see that pan out over time. Not every company will be as dominant as they used to be. But then the businesses that end up using them as an intelligence layer and continue to show positive surprises, I think you will see a rerating in their multiple.