What is the AI bubble? A simple guide to the boom and the doubts in 2026
The AI bubble is the fear that AI stock prices and data-centre spending have run far ahead of real profits. Here's the case on both sides.

The AI bubble is the worry that the prices of AI companies, and the hundreds of billions going into chips and data centres, have run far ahead of the profits AI will realistically earn. If that's true, prices eventually fall hard once investors lose faith, much as internet stocks did after 2000. Whether we're actually in one is disputed by serious people on both sides, and this month the argument got a lot louder.
On Thursday 5 November the Nasdaq fell 5.52% in one session, its worst day since April 2025, with AMD down 17.3% and Nvidia down 9.3%. Twelve days later Nvidia reported quarterly revenue of $115.8 billion, more than double a year earlier. Both facts are true at once. That's the puzzle.
Where do I land? I think parts of the AI market look bubbly, especially the way it's being financed, while the technology underneath is real and already earning money. Those two ideas can sit side by side, and most of this piece is about telling them apart.
The background: booms, busts and the dot-com comparison
A bubble, in plain terms, is a price that's held up mainly by the belief that someone else will pay more later. Economists can't agree on whether you can spot one before it pops. Robert Shiller published Irrational Exuberance in March 2000, right at the internet peak; Eugene Fama has long argued that "bubble" is a label people apply only after a crash. The two shared a Nobel prize in 2013, which tells you how unsettled the question is.
History does offer a pattern, though. A genuinely useful technology arrives, money floods in faster than revenue can follow, and too much gets built. Britain's railway mania in the 1840s left the country with real track and a lot of ruined investors. The dot-com boom left behind fibre-optic cable that sat unused for years (and later carried cheap broadband to millions of homes).
The Nasdaq peaked in March 2000 and lost roughly 78% by October 2002. Amazon's shares fell more than 90% in that stretch, and the company survived anyway. That's the part people forget: a bubble can burst while the underlying idea turns out to be right.
The dot-com era and today, side by side
| Feature | Dot-com boom (1995–2000) | AI boom (2023 to now) |
|---|---|---|
| Who's spending | Telecoms and young start-ups, often with borrowed money or fresh share sales | Mostly cash-rich giants like Microsoft, Alphabet, Amazon and Meta, plus a growing pile of debt |
| Profits at the leading supplier | Cisco was profitable but priced for decades of fast growth | Nvidia made $65.3 billion in net income in a single quarter |
| What's being overbuilt (possibly) | Fibre-optic cable | Data centres, chips and power capacity |
| Circular money | Vendor financing: equipment makers lent customers money to buy their gear | AI firms investing in customers who then buy their chips or cloud capacity |
| Fall so far | Nasdaq down about 78% from peak to trough | Nasdaq more than 10% below its 6 October record, a correction |
The comparison table is mine, built from the numbers in this piece. It's rough, but the two rows that worry me most are the last two.
Real examples of AI bubble warning signs
The spending is enormous
Google, Amazon, Microsoft and Meta set out plans this spring to spend about $725 billion on capital projects this year, up 77% from roughly $410 billion the year before. Microsoft alone put its calendar-year figure at $190 billion, and Alphabet raised its own guidance to as much as $190 billion. Much of that money goes into the giant buildings and chips described in our explainer on how big AI data centers are.
Is anyone paying for what comes out the other end? Some of it, yes.
Google Cloud's quarterly revenue reached $20 billion in that same round of results, up 63%, and its contract backlog hit $460 billion. But plenty of AI is still handed out free or bundled into products people already use, such as Google's AI Mode in search, and that makes the payback period hard to judge from outside.
The money goes round in circles
A recent Bank for International Settlements bulletin found that between 2021 and 2025, 55.2% of the money invested in AI firms came from other AI firms. By deal value, 28.7% of AI firms' own investments went into other AI companies, and nearly half of those AI-to-AI deals (46.4%) also involved a supply relationship between investor and target.
Picture a small town where the builder lends money to the plumber, who uses it to hire the builder, whose busy order book then convinces the bank to lend the builder more. Nobody's lying, exactly. It's just hard to see how much of the activity is real outside demand. The BIS authors say these links have sensible business reasons (securing scarce chips, for one), yet they add macroeconomic risk and make the system more opaque.
Official bodies are nervous
The IMF's October Global Financial Stability Report lists an AI-related market correction as one of the main risks to the financial system. Its worry is specific: if AI investment and profits turn out weaker or less durable than hoped, prices could fall, and rising leverage plus concentrated exposures could make that fall worse.
Even supporters of the build-out admit the bill is changing some companies. Dec Mullarkey of SLC Management put it bluntly about Meta: "Investors continue to be concerned about how Zuckerberg's once capital-light money machine may be morphing into a capital-intensive incinerator."
Meta has guided to more than $145 billion of spending this year.
Politics just got involved
The 3 November midterms added a new risk. Voters rejected data-centre projects in 40 of the 41 local ballot measures on the subject, and Gina Hinojosa won the Texas governor's race after campaigning for a pause on new data centres. Investors read that as a threat to the construction pipeline that chip forecasts depend on, and the 5 November sell-off followed.
Then Nvidia lost Jensen Huang as chief executive, and Colette Kress is now running the company as interim CEO. By 16 November the stock had closed about 35% below its 6 October record. The next day's results were strong: data-centre revenue of $107.6 billion and guidance of $130 billion for the current quarter.
Strong results didn't end the debate, which is telling in itself.
The case for and against calling it a bubble
The bear case is easy to state. Valuations assume years of fast growth, a lot of demand is financed inside the industry itself, and more of the build-out now runs on debt just as borrowing costs climb (10-year US Treasury yields were above 5% in mid-November, with markets pricing in a likely Fed rate rise in December). The research house TS Lombard has war-gamed a serious unwinding that knocks at least 30% off US shares and causes as much as $200 billion in credit losses.
The bull case is just as concrete. Nvidia's gross margin was 74.8% last quarter, and its net income of $65.3 billion was more than double a year earlier. Microsoft has said it expects to stay short of capacity through at least the end of this year, which is the opposite of a glut. Brent Thill, an analyst at Jefferies, summed up that camp in April: "The AI economy is healthy." He went further, too. "The bear thesis is garbage."
I'm not convinced by either extreme. The firms doing most of the spending can afford it, which is a real difference from 2000, when many of the buyers were start-ups with no profits. But affordability isn't the same as a good return, and Mark Zuckerberg himself said there aren't many agents "that I would want to give to my mother." That's an odd thing to hear from someone spending $145 billion a year.
The bubble label has limits as well. A 10% correction isn't a crash, and a crash isn't proof the technology failed. People also use "AI bubble" for different things: share prices, start-up valuations, data-centre debt, or simply the hype. Those can deflate on different timetables, and probably will.
The AI bubble and your money, job and tools
If you own a pension or an index fund, you probably own a slice of the AI trade already, because a handful of tech giants make up a large share of the big US indices. I won't tell you what to do with that; it's a question for a regulated financial adviser who knows your situation. It's still worth knowing that "I don't hold AI stocks" may not be quite true.
For people working in or near tech, the effect is more direct. A funding squeeze usually hits hiring at young AI companies first, so I'd expect some of the frantic recruiting you see in posts about urgently hiring remote AI engineer jobs to cool if the slide continues.
The skills themselves won't stop being useful, though, and the steps in our guide to getting into AI still apply in a slower market.
And as a user? Free AI tools are free partly because investors are paying the bill. If that money gets tighter, my guess is you'll see more usage limits, more ads and higher prices before you see tools disappear. You may also notice the build-out locally, through planning fights over data centres and arguments about electricity and water.
Possible paths from here
Nobody knows the timing, and anyone who claims to is guessing. There are, I think, three broad paths. Prices could deflate slowly while profits catch up, the outcome bulls are betting on. They could fall sharply if a trigger hits, such as a big customer cutting spending or a debt-funded project failing. Or growth could simply outrun the worry, as it has after every scare since 2023.
A few dates will shape that. The Fed's December meeting matters because cheap credit has fed the build-out. The big four's next capital-spending guidance, due with results in late January and early February, will show whether anyone's actually pulling back. Nvidia's $130 billion forecast for this quarter will either land or it won't. If you want the wider picture behind the chip side, our guide to AI chip companies covers who supplies what.
My own view: the AI boom has bubble-like financing sitting on top of a technology that's genuinely being adopted. If a burst comes, I suspect it'll look more like the 2000s telecom bust, painful for lenders and over-extended builders, than like the end of AI.
FAQ
What is AI bubble?
An AI bubble is a situation where the prices of AI companies, and spending on AI infrastructure, climb far beyond what the technology is likely to earn. The concern is that once investors lose confidence, prices fall sharply. People compare it to the dot-com boom, when the Nasdaq lost about 78% between 2000 and 2002. Today's warning signs include record data-centre spending, AI firms funding each other, and an IMF warning about an AI-related correction. Supporters point to fast-growing revenue and profits.
What is bubble AI?
"Bubble AI" is usually just another way of searching for the AI bubble: the fear that AI stocks and investment are overheated. It isn't a specific product. If you came across it as an app name, check who makes it, because the phrase doesn't refer to any single well-known tool.
Has the AI bubble already burst?
Not by the usual definition. The Nasdaq fell into a correction in November, more than 10% below its 6 October record. That hurts, but it's nowhere near the 78% slide after 2000, and AI revenue is still growing quickly.
Is the AI boom different from the dot-com bubble?
Partly. Today's biggest spenders are highly profitable companies, and Nvidia earned $65.3 billion in one quarter. The similarities are the scale of building ahead of demand and the circular financing: the BIS found 55.2% of money invested in AI firms came from other AI firms between 2021 and 2025.
