Is the AI Bubble About to Pop?

If you read enough financial commentary you will see two confident predictions about AI markets, often in the same week. The first: an enormous bubble is about to burst, taking the broader market with it. The second: this time really is different, the leaders generate real revenue and a correction would be a healthy pruning rather than a 2000-style collapse. Both camps are loud. Neither is fully right.

The bubble case is not absurd. As of late 2025, five companies held roughly 30 percent of the S&P 500, the most concentrated the index has been in fifty years. The Shiller cyclically-adjusted price-to-earnings ratio crossed 40 for the first time since the dot-com peak. The Bank of England has formally warned of a possible global market correction tied to overvalued AI firms. OpenAI tripled its private valuation between October 2024 and the following year, hitting roughly $500 billion before this year's funding pushed it well past that.

What the bubble argument leaves out

The trouble with the bubble narrative is that the underlying revenue is real and growing fast. Morgan Stanley analysts have called bubble fears "premature," noting that today's market leaders generate substantial revenue and positive operating margins. Neither was true of the 1999 darlings. JPMorgan published research concluding that AI does not meet the classical criteria for a financial bubble: there is no obvious excess of speculative new entrants, and the largest beneficiaries are profitable incumbents redirecting earned cash flow.

What is actually happening, on a closer reading, is two different stories overlaying each other. Inside the listed market you have a small number of mega-cap names trading on plausible cash flows but stretched multiples. Outside the listed market you have a huge cluster of pre-revenue AI startups trading on narrative alone. One of those is bubble-shaped. The other one might just be pricey.

The boring truth

The honest answer to "is it a bubble" is: parts of it are, parts of it are not, and the parts that are will probably correct without the broader index collapsing. Software stocks have already had what some analysts are calling the "SaaSpocalypse," a sharp re-rating as investors price in AI replacing traditional subscription business models. That deflation is happening in real time and is not catastrophic.

The harder question, which the bubble debate tends to skirt, is when the data centre buildout currently absorbing tens of billions of dollars in capex starts showing up as profit at non-tech companies. If the answer is "in 2026 or 2027," the bull case holds. If the answer is "we still are not sure, ask again in five years," some of the higher-end valuations will have to come down.

None of which is the same as a 2000-style crash. The companies driving the AI cycle have actual customers and actual margins. They are not WebVan. The risk is more mundane: a slow grind lower as the market recalibrates expectations to match the real pace of enterprise adoption. That is uncomfortable but ordinary, and it makes for worse headlines than "the bubble is about to pop."