What people mean when they say the AI bubble will burst

On November 5, the Nasdaq 100 fell 6.3% in a single day, wiping out about $2.3 trillion in market value. The Philadelphia semiconductor index dropped 12.3%. Nvidia lost 9.2%. Asked to explain it, the economist Ed Yardeni told NBC News, "I think it's an AI bubble losing some air very, very rapidly."

You have probably heard some version of that sentence for a while now. "The AI bubble will burst." "I can't wait for the bubble to pop." People say it as though everyone agrees on what it means. They don't. Some mean the stock market will fall. Some mean AI companies will go bankrupt. A smaller group means the technology itself will turn out to be useless and go away.

This post explains what a bubble is, why so many people think AI has one, and what a burst would and would not change.

What a bubble is

Every investment price combines two things: what something is worth today and what people expect it to be worth later. A bubble forms when the second part runs far ahead of anything the first part can support.

The key feature is that people keep buying even when they suspect the price is too high, because they expect someone else to pay more later. While new buyers keep arriving, prices rise and everyone looks smart. The bubble bursts when the new buyers stop arriving. Nobody wants to pay the inflated price anymore, sellers outnumber buyers, and prices fall quickly toward what the underlying thing can actually earn.

The usual historical example is the dot-com bubble. Investors poured money into almost any company with a website, whether or not it had a plan to make money. The Nasdaq peaked in March 2000 and then fell about 78% by October 2002. Hundreds of companies disappeared. The internet did not.

Why people think AI is a bubble

Nobody disputes that AI is useful. The argument is about money: how much is being spent compared with how much is coming back.

The spending is enormous. Alphabet, Amazon, Meta, and Microsoft are expected to spend roughly $660 billion to $690 billion on capital investment in 2026, most of it on AI data centres and chips. J.P. Morgan estimates that this spending will consume 90% to 100% of their operating cash flow this year, up from about 60% in 2025. In other words, some of the most profitable companies in history are now spending close to everything their businesses bring in, and increasingly borrowing to cover the rest.

The revenue hasn't caught up. AI revenue is growing fast, but not as fast as the spending. A Forbes analysis from June argued that investment is scaling faster than revenue, which pushes the payback date further out each year. The companies building the models are mostly losing money. OpenAI's own projections, as reported by The Information, show a loss of about $14 billion in 2026.

Some of the money is circular. In September 2025, Nvidia announced plans to invest up to $100 billion in OpenAI. OpenAI would then use that money to buy Nvidia chips. Oracle signed a $300 billion computing deal with OpenAI, and OpenAI also took a stake in AMD while agreeing to buy its chips. Deals like these make each company's numbers look stronger, but much of the money moves between the same few firms. The Wall Street Journal described this kind of circularity as something that "can be virtuous on the upside, and vicious on the downside."

The products are often priced below cost. Many AI tools have been cheap or free while companies compete for users, with investors covering the difference. That subsidy is starting to shrink. Through 2026, several AI coding tools moved from flat monthly fees toward usage caps and metered billing, because heavy users were costing more than they paid. If prices rise, some of the demand that justified all those data centres may turn out to depend on low prices.

Businesses are still looking for the payoff. A National Bureau of Economic Research study published in February 2026 found that 90% of firms reported no AI impact on employment or productivity so far. Many companies are paying for AI. Far fewer can yet show that it has paid for itself.

Even people inside the industry have said it. In August 2025, OpenAI's chief executive Sam Altman said he believed investors were in a bubble. The Bank of England and the IMF have both warned about a possible correction.

What a burst would actually look like

When someone says the AI bubble will burst, the most likely version looks something like this.

Investors decide that AI profits will arrive later, or smaller, than they had priced in. AI-related stocks fall sharply. Because a handful of companies such as Nvidia, Microsoft, Apple, Alphabet, and Amazon made up roughly 30% of the S&P 500 by late 2025, a fall in those stocks drags down the whole market. Anyone with a pension, a 401(k), or an index fund would feel it, even if they never bought an "AI stock" on purpose.

Funding then becomes harder to get. Startups that depend on raising new money to cover their losses either shut down, cut staff, or get bought cheaply by larger companies. Planned data centres get cancelled or paused. Lenders that financed them take losses. Workers in those companies and their suppliers lose jobs, and the slowdown spreads outward.

There would be a few upsides. Computer parts such as memory chips and graphics cards, which have become expensive because data centres buy so many of them, would likely get cheaper. Competition for electricity near data centres would ease.

How bad it gets depends on how much borrowed money is involved. A stock market fall mostly hurts investors. A fall that leaves lenders holding bad loans can spread much further, which is what made 2008 so damaging. That is why regulators are paying close attention to the growing role of debt and private credit in funding AI infrastructure.

What a burst would not mean

A burst would not mean that AI disappears.

The dot-com crash is the best guide here. Pets.com and Webvan failed. Amazon lost more than 90% of its share price and survived. Telecom companies laid so much fibre-optic cable in the late 1990s that most of it sat unused after the crash. Years later, that "dark fibre" carried the video streaming and cloud computing that people now take for granted. The investors who paid for it lost money. Everyone else got the infrastructure.

Something similar is the most likely outcome for AI. The models people already use would keep working. Some of today's companies would vanish or be absorbed, and the survivors would probably charge more and push AI into fewer places where it isn't wanted. The people hoping a burst will make AI go away entirely are likely to be disappointed.

The case that it isn't a bubble

Not everyone agrees there is a bubble at all. Federal Reserve Chair Jerome Powell has pointed out that, unlike many dot-com firms, today's AI leaders earn real revenue. Goldman Sachs argues that the big gains in AI stocks are backed by profit growth and that valuations remain well below dot-com levels. Morgan Stanley notes that large US companies have far more cash than companies did before earlier crashes. Cloud revenue at the major platforms is growing quickly, and most data centre capacity under construction is already leased.

There is also a middle view. John Higgins of Capital Economics told Fortune that if you judge a bubble by stretched share prices, the bubble has arguably already burst, because Big Tech valuations came down from their peak. He worries about something rarer: that the profits themselves may be inflated by AI spending that cannot last. If the spending slows, the earnings that justify today's prices could shrink too.

So is it happening now?

Nobody can say for certain, and you should be wary of anyone who claims to know the date. Bubbles are usually obvious only in hindsight, and the trigger is often something few people were watching.

The November 5 selloff had several causes at once: weak results from chipmakers, the midterm elections bringing in hundreds of candidates who pledged to restrict data centres, and growing doubt about whether AI spending will pay off. Stocks recovered some ground the next day. One bad day doesn't prove a bubble has burst. It does show how quickly confidence can move.

If you want to judge for yourself, a few signs are worth following:

• whether AI companies' revenue grows faster than their spending
• whether the large tech companies cut their data centre budgets
• whether AI startups struggle to raise new funding
• whether AI tool prices keep rising and customers keep paying
• whether lenders start charging much more to finance AI projects

The short version

When people say the AI bubble will burst, most of them mean that investors have paid more for AI companies than those companies will be able to earn, and that prices will fall when investors accept it. Companies would fail, jobs would be lost, and retirement savings would take a hit. The technology would stay. It would just cost more, and there would be fewer companies selling it.