Has the AI bubble burst? Why AI shares are falling in China and the US

For two years the trade was simple: if a company touched AI, its shares went up. October 2026 is the month that stopped working, in both of the countries that matter most. On Friday 30 October the Nasdaq 100 fell 3.4% and lost more than $1 trillion in a single afternoon. The same day, China's blue-chip CSI300 closed out its worst month since early 2016.

Headlines are calling it the AI bubble bursting. In China, some fund managers are using exactly those words. In the US the picture is more complicated. The two sell-offs arrived at the same time and feed off each other, but they are not the same story.

What happened in the US

The American sell-off built up over three weeks, and each step had a clear trigger.

• 14 October: inflation. September's consumer price index came in at 3.7%, above the 3.6% forecast, and traders began betting on another interest rate rise. Higher rates hurt AI companies twice: their valuations depend on profits far in the future, and much of their data centre building is now paid for with debt.
• 22 October: Meta. The Nasdaq fell 2.6%, its worst day since June, after Meta raised its 2026 capital spending forecast to between $135 billion and $145 billion while its profit missed expectations. The stock fell nearly 7%.
• 29 October: Alphabet. Google's parent had its biggest one-day percentage fall on record, losing about $500 billion in value, after weak search advertising results and a sharp increase in cash burn. Amazon dropped as much as 10% the same evening after confirming it expects to spend about $220 billion this year.
• 30 October: chips. Intel fell 29%, its worst day in data going back to the 1980s, after forecasting weak fourth-quarter sales and profits and blaming an "unprecedented memory crunch". The Philadelphia semiconductor index fell 7.2%. AMD, Arm, Qualcomm and Marvell each lost about 10%, as did CoreWeave and Nebius, the so-called neoclouds that rent out AI data centres.

By the close, the Nasdaq Composite had had its worst month since March 2025, and Reuters estimated that more than $2 trillion had come off the Nasdaq 100 in October alone.

The real worry: spending that grows faster than revenue

Nobody is arguing that the big US tech companies are failing. Revenues are strong, Amazon's cloud business accelerated, and Microsoft's results suggested its AI spending is starting to pay off. The problem is the size of the bet. J.P. Morgan estimates that this year's AI spending by Amazon, Alphabet, Meta, Oracle and Microsoft alone will roughly equal everything they have spent since ChatGPT launched in November 2022. It forecasts $900 billion of global AI capital spending in 2026 and $5.5 trillion by 2030.

That money increasingly comes from borrowing. Meta is planning a record $97.5 billion fundraising, which spooked shareholders worried about dilution, and Microsoft is reportedly preparing its first US investment-grade bond sale in more than nine years. When rates are rising and returns are still arriving slowly, investors start asking how long the spending can outrun the income. That is the question October answered with a sell-off.

What happened in China

China's fall is bigger, faster and more like a classic bubble. The Shanghai STAR50, which tracks the country's AI and semiconductor leaders, has dropped about 40% in under four months, erasing around $480 billion in market value. The CSI300 lost 8.5% in October and has now fallen for twelve weeks in a row, the longest losing streak on record.

The individual stocks tell the story better than the indices:

Why China fell harder

Valuations were extreme. At its peak the STAR50 traded at nearly 250 times trailing earnings. Even after the crash it sits around 100 times, against about 30 for the Nasdaq 100. When prices have run that far ahead of profits, it does not take much to reverse them.

AI became the whole market. According to Goldman Sachs, AI stocks make up 42% of the value and 57% of the trading across mainland and Hong Kong shares. When one theme is that dominant, a wobble in it is a wobble in everything, and there is nowhere inside the market to hide.

Retail money arrived late. Beijing has been encouraging household savers into the stock market to help fund the AI push, and many of them bought near the top, reading the rally as official support. Those are the investors who sell fastest when it turns.

Washington added pressure. The US is moving toward banning Chinese suppliers such as Innolight and Eoptolink from American data centres. Traders also point to tax bills and to Beijing itself, which has been cracking down on leverage and high-frequency trading, with state-backed funds selling into rallies to keep the market on what officials call a "slow bull" path.

And the US sell-off spread. Every bad day on the Nasdaq this month was followed by AI-led falls in Shanghai the next morning. Chinese suppliers depend on the same data centre build-out that American investors are now questioning.

"All bubbles inevitably burst." Wang Zhuo, Shanghai Zhuozhu Investment Management, who expects China's AI suppliers to end up in a price war.

So has the bubble burst?

In China, it is hard to call it anything else. Prices rose 400% to 700% on debut, valuations reached 250 times earnings, retail buyers piled in at the top, and the index has now lost 40%. That is what a bubble bursting looks like, even if the underlying technology is real.

In the US, not yet, at least not on the evidence of October. A fall of around 8% from a record high is a correction, not a crash. When the dot-com bubble burst, the Nasdaq lost almost 80% over two and a half years. What October did establish is that investors have stopped rewarding AI spending for its own sake. Companies now have to show the money coming back, and those that cannot, Intel this week, Alphabet the day before, get punished hard.

The next tests are close. Apple reports next week and Nvidia, the bellwether for the whole AI build-out, reports in mid-November. If Nvidia's order book holds up, October will look like a reset. If it does not, the American story starts to look a lot more like the Chinese one.

What it means if you build with AI

A falling share price does not switch off a model. The tools businesses use today will still work next month. What changes is the economics behind them. AI providers that have been pricing below cost to win customers face investors who now want margins, so expect fewer subsidised plans, more price changes like Opus 5.5 doubling on paper, and some consolidation among smaller providers.

The practical response is the same one that made sense before October: do not build anything that only works with one vendor at today's price. Route simple work to cheaper models, which is the subject of not every call needs the expensive model, keep the model behind an interface you can swap, and measure what each AI feature actually earns. Investors are now asking every AI company to prove its spending pays off. It is a fair question to ask of your own AI features too.

This article explains market events and is not investment advice.