AI Needs Trillions In Infrastructure Spending

During their recent episode, Taylor, Carlisle, and John Love discussed AI Needs Trillions In Infrastructure Spending. Here’s an excerpt from the episode:

And Chris put some numbers around it. So, the hyperscaler CapEx was over $400 billion last year. And it’s headed for something like $4 trillion over the next decade.

That’s the projection. Here’s the part that starts to rhyme. Bloomstran lined this up with every capital cycle boom before it.

Canals, railroads, fibre, etc. Everyone’s a frenzy of building. But of course, most of those builders went bankrupt.

Only 4% of the fibre was actually lit during the dot-com boom. We just had crazy overcapacity. But the internet still got built.

They’ll use the fibre later. 20 years after the wreckage for all the people who paid that original tab, unfortunately, they didn’t survive that. So, let’s run this bamboo question on the AI stock.

Can the engine feed it? So, we’re spending $4 trillion on chips and data centres. If you depreciate that over 10 years, that’s $400 billion a year baked into the cake every year.

Whether revenue shows up or not, by the way. And 10 years is quite a generous assumption. It’s probably more like three to five years depreciation schedule for these chips.

But if we cut that 10 years in half to five, the bill doubles to $800 billion a year in depreciation. Then you have to put a return on that. Just to break even at the cost of capital, you need maybe $400 billion a year in profit to earn what these companies are actually used to earning at 20% returns on invested capital.

It’s more like $800 billion that you need. And right now, the entire S&P earns about $2.5 trillion. So, in this one buildout, to throw off something like a third of all the profit in corporate America today, we need it fresh on top of all of that existing that they’re making in order to justify the spend.

And of course, today’s revenue numbers, if we look at Anthropic is running at $41 billion or something, Chachupic at $25 billion. Real money, big numbers, but nothing compared to $4 trillion. And then here’s the concerns when you get to financials of big tech.

Right now, in two years ago, CapEx ate about 40% of the cash that these companies generated. Today, it’s effectively all of it. And these cash machines are spinning everything that they make, basically.

Buybacks are shrinking, and the rest that’s coming now is debt. And a lot of that is parked off balance sheet when you see some of these structured finance things. And of course, fresh equity raises now, which you saw with Google and Berkshire recently.

So plus today, a lot of that revenue that’s showing up is a lot of these same dollars that are circulating around between a handful of companies. How much profit you can make out of that, who knows? So we’re all in this poker game where everyone basically keeps buying chips from the house, which the house is Nvidia today.

The pun was intended there. So anyway, no doubt, the tech is real. I think we’re all using it.

We’re all experimenting. We’re all benefiting from it. It’s awesome.

But the internet was real too. It’s pretty awesome. Still did not create high returns for the companies that built it.

In fact, it killed many of them. So whether you’re in a bamboo garden, in a startup, or in a data centre that’s the size of a city, there has to be a real engine down there in the rhizome that’s throwing off more than you feed it. Or you’re kind of just looking at maybe a ravenous hole in the ground eventually.

So I honestly don’t know how this all ends or over what timing, but it’s going to be exciting to watch it play out. So where do you guys see this thing going?