The AI stock bubble, much debated through the back half of 2025, has already burst. That’s the conclusion of John Higgins, chief markets economist at Capital Economics. He’s more worried about what’s still brewing.

A bubble typically refers to when assets have valuations that far exceed their intrinsic worth, usually seen when share prices soar despite solid evidence of strong financial results. “If you’re judging whether a bubble exists or not in relation to how stretched or otherwise its valuation is, then there’s an argument that the bubble has burst,” Higgins told Fortune.

In a note to clients published this week, Higgins found that for information technology and the rest of Big Tech, the ratio of the current share price to earnings per share has risen over the last few years, showing inflated valuations. But as of around October 2025, that price-earnings ratio fell and is now the smallest since the pandemic. The dotcom bubble at the turn of the century largely followed the same pattern, though the price-earnings ratio was much greater, exceeding 150% for the IT sector in the early 2000s, compared to a peak of nearly 75% in late 2024, Higgins noted.

AI valuations have indeed soared. As of fall 2025, there were 498 AI unicorns with a combined valuation of $2.7 trillion, according to data from tech market intelligence platform CB Insights, 100 of which were founded in 2023 or after. More than 1,300 AI startups have valuations over $100 million. OpenAI’s valuation reached $730 billion last month, according to CFO Sarah Friar, up from $500 billion in October, less than six months prior.

However, the tech sector has come back down to earth, a result, in part, of the “SaaSpocalypse,” a rapid selloff of software-as-a-service (SaaS) stocks as investors fear agentic AI being able to easily replace traditional software business models. Both Salesforce and ServiceNow have lost about 30% of their respective values since the beginning of the year.

“Investors had sort of honed in on that software services industry group as being one of those sectors that was relatively vulnerable to that rollout of AI,” Higgins said. “And therefore we had a big paring back in the valuation of that sector in particular.”

It’s not just the SaaS industry taking a hit, Higgins argued. The semiconductor industry has also seen a recent slowdown, with high demand fuelling a chip shortage, and recent geopolitical tensions, such as tariffs and the war in Iran triggering supply chain challenges.

AI’s next bubble is a rare one

Another bubble may be hiding within the story of these industries’ obstacles, according to Higgins. Tech companies’ earnings have rocketed upward in the last few years, raising the question of how sustainable this amount of growth can be. Bloomberg Intelligence estimates earnings growth for the Magnificent Seven to be around 18%, compared to 11% growth from the remaining 493 companies in the S&P 500. Last month Nvidia reported a revenue of $68.1 billion for its fourth quarter, a 73% year-over-year increase.

“There may be one [bubble] actually in the fundamental side of things, which is quite rare,” Higgins said. “Normally we think of a bubble as being something where the price has gotten out of whack with the fundamentals themselves…In this case, the bubble actually may be in the earnings themselves.” By this, Higgins was referring to the main argument that tech boosters in the anti-bubble camp have turned to: the enormous profits being produced by the biggest public tech firms that dominate the Magnificent Seven. In other words, he’s asking, what if these profits go down?

There’s a couple of reasons why AI earnings may soon reach a cliff and end up in a market correction. For one, Higgins said, demand for AI may be much lower than initially anticipated, leaving tech companies to reckon with the estimated $539 billion in AI capex for 2026, per Goldman Sachs. While 88% of companies report regular AI use, according to McKinsey, adoption may be stalling as a result of employees’ anxiety around the technology displacing them from their jobs.

The greater risk to AI earnings will be if the economy remains in a precarious position, Higgins suggested. The ongoing Iran war has halted the helium output in Qatar, responsible for about one-third of the world’s supply of the odorless gas used to manufacture computer chips. Not only have data centers become a target of attacks during the conflict, but energy prices could also drive up input costs of these facilities.

“If the economy, more generally, were to weaken, that could also weigh on the stock market and weigh on the earnings of companies who are making money from the rollout of AI,” Higgins said, “even if demand for AI itself isn’t really weakening much.”