Investors once wanted Unitree shares 8,000 times over. Then the robotics company lost roughly half its value. When the AI bubble in China burst, about $480 billion disappeared.
The promise was simple: AI would change everything and generate billions. Investors believed it and poured money into AI, chip, robotics and semiconductor firms. Beijing wanted China to be a technology superpower, and the money followed that ambition. Chip maker Meta-X jumped 700% on listing. Moore Threads, another AI chip company, rose 400%.
Prices ran far ahead of profits. Investors bet on what companies could earn tomorrow, not what they earned today. Then they started asking how much these firms really earned, what the technology cost and what return to expect. Why did those questions arrive only after the rally? Once they did, the mood turned. Unitree, Meta-X and Moore Threads each lost about half their value.
The damage spread to the indices. The STAR 50, which tracks China's largest technology and semiconductor companies, has fallen 40% from its July peak. The CSI 300, covering 300 major companies in Shanghai and Shenzhen, has declined for ten straight weeks. Even after the fall, investors pay about 100 times past earnings for the STAR 50. The Nasdaq 100 in the United States trades at around 30 times. Chinese prices may have further to drop.
Confidence is the next casualty. When investors lose faith, companies struggle to raise money. New listings get delayed and expansion plans get suspended. None of this means AI is finished. It means investors expected too much and paid too much for the promise.
Investors bet on what companies could earn tomorrow, not what they earned today.
That habit was not confined to Chinese retail enthusiasm. It was the pricing logic of a whole sector, and a 100-times multiple shows the market has not finished correcting it.
The pressure reaches beyond China. Microsoft spent about $175 billion on AI this year alone. Alphabet is expected to spend $205 billion and Amazon about $220 billion. Can AI earn enough to justify those sums? Building data centres is one thing; earning a profit from them is another. So Microsoft is hunting for savings. It is offering voluntary compensation packages to about 60,000 employees worldwide, more than a quarter of its workforce. That does not mean 60,000 job losses. It does show a company squeezing costs while it keeps spending on AI.
India shows the same strain in a different form. Six major IT companies, TCS, Infosys, HCL Tech, Wipro, Tech Mahindra and LTI Mindtree, recorded about 92,000 employee exits last year. Many people resigned, retired or moved to other firms, so the figure is not a layoff count. The telling number is the net one: these firms lost more people than they hired, and their combined workforce shrank by about 24,000. Clients want more work for less money. The companies want to protect their margins. AI lets them do both in some types of work, and hiring has slowed to match.
For a software engineer in Bengaluru or a Microsoft employee weighing a buyout, the AI debate is no longer abstract. Companies that must show returns on huge AI bills will look hardest at the one cost they control, which is payroll. Investors in China are already paying for the opposite mistake.
Building data centres is one thing; earning a profit from them is another.
The sentence explains why a $175 billion bill at Microsoft sits beside a buyout offer. Spending is a decision, while profit is a result that nobody has yet proved.
The AI story is moving from promise to proof. A technology boom has to pass two tests: it has to work, and it has to make money. China's investors have already seen what happens when the second test arrives late.