The AI Bubble, Explained Without the Panic

You have heard the phrase everywhere this year: the AI bubble. Usually said with either glee or dread, rarely with a definition. So here is the definition, because everything else follows from it.

A bubble is when the price of something rises faster than the money it actually earns, because buyers expect someone else to pay more later. The thing itself can be real, useful and world-changing. Railways were real. The internet was real. Both had bubbles.

What people mean when they say it about AI

They mean roughly this: an enormous amount of money has been spent on chips, data centres and startups, in the belief that AI will generate enormous revenue. Some of it already does. A lot of it does not yet. The gap between spending and earning is the part that worries people.

Analysts tracking 2026 have flagged a likely deflation of that gap, with knock-on effects on the wider economy, at the same time as they expect the underlying “factory” infrastructure — the compute and tooling layer for companies going all-in on AI — to keep growing. Those two forecasts sound contradictory. They are not. One is about valuations. The other is about usage.

A bubble is a story about the future being repriced. It is not a verdict on whether the future arrives.

The order things deflate in

If it happens, it will not happen all at once. It goes in stages, and knowing the stages is more useful than knowing the date.

First, the story stocks. Companies whose valuation rests mostly on an AI narrative rather than AI revenue fall hardest and earliest. They are the most sensitive to mood.

Second, funding. Startups that raised on a promise find the next round is smaller, slower, or does not exist. Consolidation follows: good teams get bought cheap, weak ones close.

Third, jobs. Hiring freezes before it fires. Watch job postings, not layoff announcements — postings move months earlier.

Last, and least, actual usage. This is the counter-intuitive part. In 2000, web usage kept climbing right through the crash. The companies died; the behaviour did not. There is no obvious reason people would stop using tools that save them time because a share price fell.

What would make it worse

Two things. One: if the debt used to build data centres turns out to be secured against revenue that never shows up. Bubbles funded by borrowed money hurt far more than bubbles funded by optimism, because the damage spreads to lenders. Two: if a few giant companies are simultaneously each other’s biggest customers, revenue can look bigger than the outside demand actually supporting it.

What would make it milder

If the money being spent keeps producing measurable savings inside ordinary businesses — a finance team closing books faster, a support desk handling more tickets — then revenue slowly catches up with the spending and the gap closes without a crash. This is the boring, likely-enough scenario that nobody writes headlines about.

What to actually do about it

Almost nothing dramatic, and that is the honest answer.

If you are an employee: your leverage is being useful with the tools, not owning the theme. Skill in using AI holds its value whether or not valuations fall.

If you are investing: understand that “AI is the future” and “this specific company at this specific price is a good buy” are unrelated statements. The first was true of the internet in 1999 too.

If you are running a business: build on things you would still want if the hype vanished. A workflow that saves you six hours a week is worth keeping regardless of what the market does.

And be suspicious of confident timing. The people who called the last bubble correctly mostly called it several years early, which in practice is the same as being wrong.