Montage image of tiles spelling ‘AI Bubble’, a bubble and circles and lines forming the shape of a brain
© FT montage

(noun) theory that tech stocks are priced far above their actual value

Investors and analysts generally use the B-word sparingly. Declaring an important asset class to be in the throes of an unstable speculative frenzy is an easy way to look foolish if you’re wrong — and to generate needless and painful market volatility if you’re right (and influential enough).

In 2025, though — the year ChatGPT hit its third birthday — the term broke confinement. After years of stellar performance in AI-related stocks, even the high priests of Big Tech and Big Finance could see the excesses. Jeff Bezos of Amazon said in October that sure, it’s a bubble, but a “good” one. Sam Altman of ChatGPT parent OpenAI breezily accepted that money could potentially be misallocated. JPMorgan chief Jamie Dimon said “a lot of assets out there . . . look like they’re entering bubble territory”.

That does not mean those assets are poised to collapse. The only way to spot a true bubble is when it bursts — that could be years away. It does, however, mean investors need to be on guard. A full-blown crash would hit not just the markets, but potentially the global economy.

Already, markets are eyeing AI with more suspicion. Shares in Meta have declined as investors balk at seemingly unconstrained AI spending. Chips behemoth Nvidia, which became the world’s first $5tn company this year, has declined in value as rivals appear to pose the first serious challenge to its dominance.

Whistling along apparently without care, however, is CEO Jensen Huang. He noted in November that while there has been “a lot of talk” of an AI bubble, “from our vantage point, we see something very different”.

Should that give you comfort? The picks and shovels in this gold rush are his, after all. But once the notion of a bubble sets in, it’s awfully hard to deflate gently.

katie.martin@ft.com

Copyright The Financial Times Limited 2025. All rights reserved.
Reuse this content (opens in new window) CommentsJump to comments section

Follow the topics in this article

Comments