U.S. artificial intelligence companies are expected to spend $650 billion on capital investments this year, mostly in the form of data center infrastructure. The companies clearly believe that the demand for AI services is at an inflection point. But the scale of their spending raises macroeconomic questions.

Does this level of AI investment amount to a fiscal stimulus for the United States? What ripple effects is AI already having on the U.S. economy? Is this what an economic bubble looks like?

Those are just a few of the questions that came up in my recent conversation with FP economics columnist Adam Tooze on the podcast that we co-host, Ones and Tooze. What follows is an excerpt, edited for length and clarity. For the full conversation, look for Ones and Tooze wherever you get your podcasts. And check out Adam’s Substack newsletter.

Cameron Abadi: Historically, what other businesses have invested on this level on this sort of time scale?

Adam Tooze: To place this in perspective, the U.S. economy last year was $30 trillion. So the planned investment of $650 billion for the coming year is about 2 percent of GDP. And that puts us in the ballpark of previous tech booms in the United States. This is a really large share of GDP.

We’re not quite at the level of the railway booms of the 19th century, let alone really large wartime spending or the really exorbitant private sector booms that we saw, for instance, in Japanese real estate in the 1980s. But a level of spending rising from half a percentage point of GDP to 1.5 percentage points to closer to 2 percent is clearly a major jolt to the U.S. economy in a positive sense.

CA: Does this level of investment amount to a fiscal stimulus for the United States?

AT: That’s the right way of thinking about it. In a Keynesian logic, investment and government spending are both what’s called exogenous shocks—exogenous impulses to economic growth. In this particular case, it is instructive because it’s essentially being funded through various types of private credit and the mobilization of cash by the businesses themselves.

Analysts point out that some of that, in macro terms, leaks to the outside because famously the chips are coming from Taiwan. So that goes in terms of greater imports. But locally in the U.S., this provokes the build-out of data centers, data center infrastructure. And then you have people clustering around those businesses, and the workforce has to eat, and they have to be housed.

So you get exactly the kind of standard multiplier effects that you would anticipate. It’s spread out across the year, it’s spread across the United States. It bundles in particular places where you really do get powerful effects. Parts of Virginia are experiencing a real shock, because that’s a major area for data center concentration. And if you look at, say, real estate in the Bay Area, it’s on a tear again, because there’s huge wealth being generated in the AI sector, at least right now.

So yes, in all of these respects, this is a driver of growth. It’s easy to exaggerate exactly how large this will be. But it is making up a significant share of American growth. People estimate that at the beginning of 2025, it was really the only thing that was driving the economy forward. That didn’t last, but the rest of the economy was contracting, and then AI did the offset. This isn’t untypical. In the green energy space that I think about a lot, it’s estimated that about 30 percent of Chinese growth right now is being driven by the new energy sector. So if it were the case, that 20 percent to 30 percent of American growth was being driven by AI, that would not be exceptional. But it would, of course, be a very powerful impetus.

CA: What are the ripple effects that AI is already having on the American economy? Has the market value, say, of existing companies taken a hit in ways that are already legible? Are unemployment rates already affected?

AT: Well, the positive effect is on the construction side, so we’re seeing a big surge in demand for high-tech construction and for data center energy supply, right? There are these huge waiting lists, basically, for, say, gas turbines, which will be used to power the data centers.

There is a vast amount of speculation right now about the impact on labor productivity. It’s not showing up in the macro data yet, so far, not to a significant extent. There are various correlations being put together by some of the Federal Reserve branches looking at reported AI use and labor productivity growth. It’s something more than vibes, but not a great deal more than vibes.

On the other hand, in the stock market, where people, of course, make bets on the future earnings of firms and being first in is more important than having the immediate macroeconomic estimate correct, we are seeing ripple effects. Notably, for instance, in the last week or so, Reuters, which has a major legal database service, saw a huge sell-off in its shares because people think that various types of AI agents will effectively be replacing those kind of database systems.

So we are seeing a lot of speculation about the impact. Many of us are using AI daily, and it’s difficult to imagine this isn’t going to change the world. We are seeing soft macro data, I would say. Nothing really that right now adds up to a clear read. And we are seeing markets, as they need to do, anticipate the effects on individual corporate names going forward.

And that’s not even talking about the big news of the last week or so, which is the sell-off in the AI firms themselves as a result of markets being shocked by the scale of their capital spending and the increasingly insistent question. I mean, sure, we get it, it’s an amazing technology, you’re going to roll out, you are going to build these data centers, but how are you actually going to earn any money doing this?

That question becomes more and more pressing. And that is weighing on the valuation of all of the tech stocks.

CA: Is this what it feels like to live through an economic bubble? Do the central players realize they’re participating in a bubble? Or does it appear to them differently than that?

AT: There are two levels of economics questions that we could focus on about this. If it is a bubble, what kind of bubble is it? And there’s at least three different types of bubble that we might think about.

One is a Metaverse-type bubble. In other words, it just doesn’t turn out to be that useful, and no one really wants to go there. And so, you spend a lot of money on it and then you don’t really do anything about it.

Or it could be a bubble like the railways were a bubble. Which, they’re not the Metaverse or tulips—they’re railways. And they interconnected the United States and Europe and totally transformed the world. They were a bubble in the sense that the revenue streams and the earnings that ultimately resulted came in a different time pattern from the buildup of financial obligations and resulted in huge financial meltdowns repeatedly, but not because the railways weren’t really vital, but simply because there was a mismatch between financial expectations and the resulting technological change. So we could be in that sort of bubble. It could be that all of these firms suffer major financial crises, but the technology is transformative over, say, a longer time horizon than they need to make this payback.

And then the third type of sort of bubble disjuncture would be not that it doesn’t pay back and it doesn’t transform the world, but that it just doesn’t turn out to be something you can easily build a motor around, in Warren Buffett’s terms, so as to generate profit.

And that might be the problem. So it’s more like the airline story or Chinese involution type stories, where you’re totally transforming the world with your technology—you’re just not making any money with it. And that, from the point of view of the market, would be terrible. From the point of view of the consumer, it would be potentially quite great.

And then there’s another level to all of this, which I’ve also been thinking about in terms of living in the bubble, which is the K-shaped economy narrative. Which is, whilst we’ve been fantasizing about the way in which AI might change our lives, and whilst the financial markets are trying to figure out whether it’s a Meta, whether it’s a railways, or whether it’s an airline story, all of that is being held in play by the fact the rest of U.S. economy in this K-shaped narrative is, in fact, not booming at all.

So it’s kind of like a bubble on a seesaw. There’s the tech bubble, and it’s being balanced, the boominess of the bubble is being balanced by the deflating of the rest of the U.S. economy. And that could tip. It could tip either so that the rest the economy deflates hard, in which case the bubble won’t hold the rest of the economy up and the U.S. economy might go into recession anyway. Or the rest of the U.S. economy could accelerate, in which case we could be back in a kind of inflation zone, and the bubble could get even bigger. We might even be lifted off the seesaw by the bubble because the rest of the economy is no longer holding us back.

So, I think there are these three dimensions. There’s a dimension of the psychological, the subjective—like, how do we experience this shock? There’s a dimension of, where does this fit in the history of technological change, the kind of general-purpose technology type question. And then there’s the, how is this particular moment—2024, 2025, 2026—fitting into the U.S. macroeconomy as a whole? Because it’s only part of a much bigger economy, a $30 trillion economy, which has a bunch of other dynamics going on. And all of this goes into configuring what it means to be in this moment.