What Happens If AI Hits An Energy Wall?

The technology driving the economy has some big problems

This is the way the bubble ends. This is the way the bubble ends. This is the way the bubble ends: Not with a pop, but with smog and brownouts.

What? I’ll explain in a minute. But first, a word from MechaHitler.

You’ve probably heard about the problems Elon Musk has been having with Grok, his entry in the artificial intelligence sweepstakes. He keeps trying to stop Grok from giving answers he doesn’t like, for example telling users (correctly) that right-wing violence is more frequent and deadly than left-wing violence. But his efforts to make Grok less woke keep making it racist and antisemitic instead.

But you may not have heard about the environmental crisis created by xAI, which operates Grok. From Politico in May:

Elon Musk’s artificial intelligence company is belching smog-forming pollution into an area of South Memphis that already leads the state in emergency department visits for asthma.

None of the 35 methane gas turbines that help power xAI’s massive supercomputer is equipped with pollution controls typically required by federal rules.

At the time these turbines were operating without permits. On July 2 Shelby County — overruling local protests — issued permits for 15 turbines, together with emission standards. But given the history, it’s hard to avoid being skeptical about whether those standards will be honored.

Did I mention that South Memphis is 99.5 percent Black?

There’s a lot to be said here about environmental protection and the corrupting influence of big money on politics. But for today let’s ask a different question: Why did Musk need to install his own, highly polluting generating capacity to power his data center? Why not just buy power from the grid, like everyone else?

The answer, as I explained yesterday, is that the grid doesn’t have power to spare; we don’t have nearly enough generating capacity to operate all the data centers being built to run AI. That’s why we’re hearing a lot of talk about requiring data centers to provide their own power. But what Musk is doing to Memphis shows that there are big problems with that solution too.

And that in turn suggests that AI may be a bigger short-term risk to the economy than many realize.

Notice that I said short-term, not long-term. This isn’t about AI causing unemployment by replacing humans. We’re talking instead about the risk of a recession if the current surge in AI-driven investment turns out to be unsustainable.

Like many observers, I see obvious parallels between today’s AI boom and the telecom boom of the late 1990s. (Telecom, not dot-com — overinvestment in fiberoptic networks etc. was a much bigger deal than overvalued websites.)

Then as now, huge amounts of business investment were driven by enthusiasm about an exciting new technology. And the enthusiasm back then was justified! I’ve been watching old Qwest ads about the wonders of fiberoptics, and pretty everything they promised has come to pass. You can in fact check into a grungy hotel with decent wifi and watch practically every movie ever made:

Side question: Why were tech industry ads so much better back in 1999?

But the economic payoff to telecom technology didn’t come fast enough to prevent a huge bust in telecom investment:

And this bust in turn led to a recession and a long period of elevated unemployment:

Fast forward to the present. Recent economic growth has been sluggish — only 1.2 percent at an annual rate during the first half of the year — and AI-related investment has probably been the only thing keeping us above “stall speed,” growth so slow that economic weakness becomes self-reinforcing.

The chart below shows the changes between the fourth quarter of 2024 and the second quarter of 2025 in real GDP and some of its components. Investment in information processing equipment — which at this point basically means data centers — accounted for more than half of overall growth, more than consumer spending, which makes up 70 percent of the economy:

Source: Bureau of Economic Analysis

So the AI boom is really the only thing keeping the economy’s head above water, and you have to be worried about what happens if and when it ends.

So far, so 1999. And as in 1999 you have to worry about whether the technology is really as great as the enthusiasts claim, and even if it is, whether it will generate large profits soon enough to justify today’s massive capital expenditures.

But we also have an additional worry: Will companies spending huge amounts on AI hit the brakes once they realize that they won’t be able to power their data centers?

The point is that the mismatch between the immense amounts of electricity data centers are expecting to use and the generating capacity we’re likely to have isn’t just a problem for the companies sinking hundreds of billions into AI. It’s a threat to the economy as a whole.

MUSICAL CODA

Talking about g-g-generation