The Fear That an AI Bubble Will Take Down the Economy

Open a financial paper this spring and you will meet the same anxiety in a dozen forms: the artificial intelligence boom is a bubble, it is about to burst, and when it does it will take the rest of the economy with it. The fear is loud and widely held, and it deserves to be taken seriously. It also deserves to be split into its two parts, because they are not equally well supported.

The case for worry

Start with the half that has real evidence behind it. Spending on AI infrastructure has reached numbers that are hard to reconcile with current income. Fortune has cited Goldman Sachs estimates of roughly 539 billion dollars in AI capital expenditure for 2026 alone. By some projections, US AI capex tops 500 billion dollars in both 2026 and 2027, while American consumers spend only about 12 billion a year on AI services. That distance between what is being built and what is being bought is the core of the bubble argument.

The physical build-out tells a similar story. By one widely repeated figure, markets are pricing in around 114 gigawatts of new data-center power, while only about 15 gigawatts is actually under construction. The stock market has already flashed a warning of its own. Software companies including Salesforce and ServiceNow have shed close to a third of their value this year in a sell-off some traders nicknamed the SaaSpocalypse, on fears that AI will erode traditional software margins.

Where the doomsday version overreaches

Now the shakier half. A bubble in AI assets is not the same thing as a bubble that wrecks the wider economy, and the people doing the most careful work tend to keep that line in view. The World Economic Forum has walked through how a reckoning might unfold and stops well short of forecasting collapse. Morgan Stanley analysts, for their part, have called the bubble fears premature, noting that the median cash reserves of the largest US firms are roughly three times what they were in past bubble periods, and that today's market leaders earn real revenue at healthy margins.

That matters, because the damage a bubble does depends on how it was financed. The dot-com crash and the 2008 crisis spread so far because the losses sat on top of heavy borrowing and tangled balance sheets. Much of today's AI spending is coming from a small number of cash-rich companies that can absorb a bad bet without defaulting on anyone. A painful correction in AI valuations is entirely plausible. A 2008-style cascade through the whole financial system is a different and much bigger claim, and the evidence for it is thin.

The honest reading

So where does that leave a worried reader? Roughly here. Parts of the AI trade do look frothy, a sharp repricing would surprise no one, and some firms now raising money on the promise of AI will not be around in three years. Derek Thompson and others have sketched believable ways the air comes out. But frothy is not the same as fatal. "The AI bubble will burst" and "the bubble will take down the economy" are two separate forecasts that often get bolted together, and only the first is a reasonable bet today. The second is a leap, and it deserves the same scrutiny you would give any doomsday prediction before you let it change what you do.