In under four months, China's AI stocks lost about 40% of their value, and roughly $480 billion went with them. If you use AI at work, that headline probably made you wonder whether you have been learning the wrong thing. You haven't. What crashed was the price investors were willing to pay for AI companies. The tools on your screen work exactly as they did in the summer, and the people who keep getting better at them through a moment like this usually end up furthest ahead.
Why the fall was so steep
The index at the centre of this is the STAR50, which tracks 50 of the largest and most actively traded companies on Shanghai's STAR Market, a board built for technology firms. AI and chip makers dominate it. Reuters reported the 40% drop in late October, while China's broader CSI 300 index fell about 20% from its June high. In the US, the Nasdaq was only about 6% below the record it had set three weeks earlier.
Reporters couldn't point to a single trigger. They listed doubts about AI demand, tax bills, the threat of foreign bans on Chinese chips in data centres, and a warning from Beijing on October 9 against investment bubbles in "new productive forces", a phrase that covers AI. All of that matters, but it explains the timing more than the size of the fall.
The size comes from how big the bet had become. According to Goldman Sachs, AI-linked stocks had grown to 42% of the value of all mainland and Hong Kong-listed Chinese stocks, and to 57% of their daily trading. At its peak, the STAR50 traded at nearly 250 times its companies' yearly profits. Even after the crash it sits at around 100 times, while the Nasdaq 100 trades at about 30.
Price-to-earnings ratios as reported in October 2026 coverage of the sell-off.
When people pay 250 times a company's yearly profit for its shares, every small doubt about the future becomes a large price move. We don't see a mystery in what happened next. A rally priced that far ahead of earnings was always going to fall back toward what the businesses actually make.
We have seen this chart before
Between March 2000 and October 2002, the Nasdaq lost about 78% of its value. Amazon's share price went from about $113 to $5.51, and a lot of sensible people decided the internet had been oversold.
The business underneath was doing something very different. Amazon's sales rose from $2.76 billion in 2000 to $3.12 billion in 2001, and the company posted its first quarterly profit at the end of 2001, right at the bottom of the crash. Over the same period the number of people online grew from about 385 million to 500 million, and passed 590 million the year after. For two years the share prices and the actual use of the technology moved in opposite directions, and the people who judged the internet by its stock chart missed the most important decade in business.
We think AI is following a similar path. Share prices will keep swinging, sometimes violently. The number of people using these tools to do their jobs will keep climbing anyway.
Read the headline before you react
The most useful habit we can suggest is to ask, every time an AI story makes you uneasy, whether it is about a share price or a product. Most crash coverage is about investors. It can be fascinating, and it still tells you nothing about whether Claude will summarise your contract this afternoon. DeepSeek, the best-known Chinese AI lab and the company many people link to this story, isn't listed on any stock exchange. Reports say it may list on the STAR Market one day, but for now this crash doesn't touch it.
The news that should change how you work is narrower and much less dramatic. It is a tool raising its price, removing a feature you rely on, changing how it handles your data, or shutting down. That kind of news comes from the tool's own announcements, and it rarely makes the front page.
What we would do this week
Start with the subscriptions. Write down every AI tool you pay for, and next to each one note the last real task it did for you. The ones that saved you time last week have earned their place. The ones you signed up for after an impressive launch video and haven't opened since can go. If two of your tools seem to do the same job, our Compare page will help you choose between them.
Then spend the money and attention you just freed up on going deeper with the one tool you use most. Most people use a small fraction of what they already pay for. In Claude, that might mean setting up a Project with your own documents so every conversation starts with your context. In ChatGPT, a Project for the report you write every week does a similar job. In Gemini, build a skill around your most common task. Our step-by-step lessons in Learn walk through each of these.
Finally, make your work portable. Downturns close smaller AI apps, and even the biggest companies rename and remove features. Google is turning Gems into skills from 17 November, and OpenAI retires custom GPTs on 11 December. Keep your best prompts and instructions in a document you own, keep important files in your own drive, and export anything you couldn't bear to lose. Then if a tool disappears, you replace an app instead of rebuilding your workflow.
The quiet advantage of a correction
Corrections are unpleasant for investors, but they are useful for everyone else. Money stops flowing to products that only had a good pitch, the number of launches drops, and the tools that survive are the ones people pay for because they genuinely save time. That is a much better environment for learning than the frenzy of the last two years.
Corrections also sort people. When the hype fades, plenty of people quietly stop putting effort into AI, and the ones who keep going build a skill their colleagues let go of. That usually becomes obvious about a year later, when companies start asking who can really use these tools well. If you want to see how strong teams combine them for everyday work, our Stacks are a good place to start.
The STAR50 will have more bad weeks, and so will the Nasdaq. None of that changes what you can do with these tools, or what happens to your skills when you keep building them.