Will the AI Bubble Take the Economy Down With It?

Whenever AI stocks fall hard, as they did this week, the same sentence shows up in my feeds within the hour: this is a bubble, it is about to pop, and it will take the whole economy with it. It is a tidy story, and tidy stories about markets are usually wrong in interesting ways. So it is worth pulling the claim apart rather than nodding along or waving it off.

First, the part that is not a myth. There really is a lot of money piled into a very small number of companies, and a bad day for those companies is now a bad day for everyone's pension. This week's rout, which TradingKey described as the worst for the AI trade since 2020, happened in part because so much was riding on so few names, with borrowed money amplifying the move. The financing has also started to loop back on itself, with chipmakers, cloud providers and model labs increasingly invested in one another, as the $40 billion SpaceX chip deal showed. Concentration and leverage are the two ingredients every real crash has needed. Those worries are legitimate.

Where the comparison strains

The favourite analogy is the dot-com crash of 2000, and it is a revealing one, because that era contained both things at once. There were companies with no revenue and a clever domain name, and there was Amazon. The crash cleared out the first kind and the second kind went on to remake retail. The lesson is not that it was all air. The lesson is that a market can be overpriced and the underlying technology can still be real.

That is the distinction the doom version skips. A falling share price tells you what investors will pay today, not whether the thing works. And on the question of whether the thing works, the evidence does not look like 2000. The chips are selling. The data-centre operators are full. Firms outside the technology sector are reporting that AI has become a genuine line of business rather than a demo, as Palantir did when its commercial revenue overtook its government work. You can believe the stocks are too expensive and still believe the revenue is real. Both can be true.

So what is the honest answer?

A correction, even a severe one, is entirely plausible, and anyone who tells you they know the timing is guessing. A broader slowdown is possible if enough of the circular financing unwinds at once. But the leap from there to the economy collapsing is a leap, not a forecast, and it rests on treating a stock-price swing and a technology failure as the same event. They are not. The related fear that the technology itself has stopped improving is a separate claim, and we looked at the evidence for it in our piece on whether AI progress has hit a wall.

My own reading, and it is a reading rather than a prediction, is that the people calling this a bubble and the people calling it a new era are often describing different things and talking past each other. One is talking about prices. The other is talking about capabilities. Prices can fall by half while capabilities keep climbing. If you hold only one idea about AI and the economy, let it be that those two lines do not have to move together.