Is the AI Bubble About to Burst?

You have probably seen the headline in some form this month: the AI boom is a bubble, and it is about to burst. It is one of the loudest fears in the market right now, and unlike a lot of AI panic, it is not coming only from cranks. Serious analysts are making the case. So is it true? The honest answer is that the warning signs are real, the counter-signs are also real, and anyone claiming certainty in either direction is selling something.

The case for the bubble

Start with the uncomfortable numbers, because they are the strongest part of the argument. There is a large gap between what is being spent and what is being earned. Writing in Forbes on August 19, Harvard senior fellow Paulo Carvão pointed to trillions committed to infrastructure against revenues still measured in billions, to Alphabet posting its first quarter of negative free cash flow since going public, and to off-balance-sheet commitments that dwarf the headline capital spending.

Then there is concentration. A handful of large tech companies now make up more than a third of the S&P 500's value, a heavier tilt than tech carried at the dot-com peak. When that few names hold that much of the index, a stumble in one becomes everyone's problem.

The part that worries people most is circular financing: one company invests in a second, and the second uses the money to buy the first's products. It flatters everyone's numbers on the way up and unwinds fast on the way down. We have written about the scale of these arrangements before, from Nvidia's move to recruit Wall Street for a $500 billion buildout to the fact that two AI labs soaked up nearly half of all startup funding in the first half of the year.

The case against

Here is where the tidy doom story runs into trouble. Bubbles are usually built on demand that turns out to be imaginary. This demand is not imaginary. Roughly half of American adults now use AI chatbots, a large share of workers use them weekly, and adoption across US federal agencies has doubled year over year. By several measures the technology is being taken up faster than personal computers or the early internet were.

That matters, because a company burning cash while millions of people use its product every day is a very different animal from a company burning cash while nobody shows up. Skeptics of the crash narrative, including investors who lived through past manias, argue that valuations are stretched but have not reached the irrational, everyone-quits-their-job-to-day-trade mania that usually marks a top. Elevated is not the same as delusional.

So which is it?

Probably neither clean answer. The most careful version of the argument, and the one Carvão lands on, is not that the bubble bursts overnight but that it may deflate, letting the air out gradually and sorting winners from losers over 2026 and 2027 rather than in a single crash. The pressure points are specific: a wave of giant IPOs is coming, and public markets are far less forgiving of unprofitable companies than private backers have been. Anthropic, for one, is lining up investors for a fall listing, and those debuts will be an early read on how much of this the public actually wants to fund.

The claim to treat with suspicion is the confident one, in either direction. "It is all going to collapse" and "this time is different" are both predictions dressed up as facts. What we can say is narrower and more useful: real usage is climbing, real money is being spent faster than it comes back, and a lot of that money is wired between the same handful of companies. That is a genuinely fragile setup. Whether it cracks or simply cools is not something anyone knows yet, no matter how certain the headline sounds.