On 26 October, Reuters reported that the run in Chinese artificial intelligence stocks is over. The STAR50, the Shanghai board’s index of the country’s biggest AI and semiconductor names, has fallen nearly 40% from its July high. Individual favorites such as Unitree, MetaX, Moore Threads, MiniMax and Zhipu have dropped 50% to 60% or more.
A Beijing fund manager, Dong Baozhen, told Reuters the rally had turned into a “speculative crazy bull” and that the bubble is now bursting. Others quoted in the coverage still see buying opportunities, so “burst” is a judgment, not an official verdict. The numbers behind it are hard to dismiss.
How big the fall is
The broader CSI 300 index of blue chips is down around 20% from its June high and slipped into a bear market last week. The Nasdaq, by comparison, sits about 6% below its recent record. AI stocks make up so much of China’s market, 42% of its value and 57% of its trading according to Goldman Sachs, that the damage did not stay inside tech.
Why it popped
No single event did it. Several pressures arrived together.
Prices ran far ahead of earnings. At its peak the STAR50 traded at nearly 250 times trailing earnings. Even after the fall it sits near 100 times, more than three times the Nasdaq 100’s roughly 30.
Beijing leaned on the brakes. On 9 October China issued guidelines meant to stop investment bubbles in what it calls “new productive forces,” a category that includes AI. Analysts also point to tax bills for investors and the risk that the US bans Chinese data-center equipment.
Retail money piled in. The rally was driven heavily by individual investors, and the state’s push to lift equities added fuel. That kind of crowd moves out quickly. Fund manager Wang Zhuo expects chipmakers and AI suppliers to end up in a price war, like other Chinese manufacturers.
The warning from inside the industry
The stock crash follows months of unease about the physical buildout. In February, Zhao Haijun, co-CEO of China’s largest chipmaker SMIC, said companies want to build ten years of data-center capacity in one or two years. What those data centers will actually run, he said, has not been fully thought through.
China has seen this before. Under the “Eastern Data, Western Computing” plan, builders put large facilities in western regions with cheap power. Reports say many ran at only 20% to 30% of capacity. An MIT Technology Review investigation in 2025 counted more than 500 announced projects and at least 150 completed by the end of 2024, with local reports suggesting up to 80% of new capacity sat idle. Many sites were designed for model training, far from users, and poorly suited to the low-latency work that now dominates.
Not everything is empty
The picture is mixed. SemiAnalysis estimated in September that China runs more than 24 GW of data-center capacity, second only to the US, and that spending by ByteDance, Alibaba, Tencent and Baidu is heading toward $100 billion in 2026. The firm describes high vacancy sitting alongside a shortage of AI-ready capacity. Both can be true: the wrong buildings in the wrong places, and not enough of the right ones.
That is why several analysts separate the two stories. Share prices priced in a future that has not arrived yet. The companies are still training models and shipping products, and the government has not backed away from AI.
Why the rest of the world is watching
Western AI doubts have mostly come as warnings, such as the IMF flagging heavy debt and skeptics predicting a pullback. China is the first large market to show what a sharp unwind looks like. Morgan Stanley expects offshore Chinese stocks to fall another 10% to 15% before they bottom.
The test for the US comes this week. Microsoft, Alphabet and Amazon report earnings, with AI spending under close scrutiny. Alphabet, Amazon, Meta and Microsoft are expected to spend around $650 billion on capital projects in 2026. Amazon’s CEO, Andy Jassy, has said that even its higher spending will not meet all demand this year.
What to watch next
- Whether the STAR50 holds near its current level or the multiple keeps compressing toward the Nasdaq’s.
- Any new US export limits on Chinese data-center hardware.
- Utilization and pricing at Chinese cloud providers, which will show whether demand is catching up with capacity.
- This week’s Big Tech earnings, for signs that US demand is holding where China’s stocks cracked.
Markets do not need a collapse in the technology to fall hard. They only need a gap between what is promised and what is earned.
Sources
- In China, the AI bubble has already burst (Reuters analysis, via Global Banking & Finance Review)
- China’s AI stock bubble has burst, Reuters reports (Crypto Briefing)
- AI Bubble Bursts in China, Wiping $480 Billion: Is America Next? (BeInCrypto via Yahoo Finance)
- China’s AI Bubble Has Started to Burst (Startup Fortune)
- China’s Top Chipmaker Warns Rushed AI Capacity Could Sit Idle (Bloomberg)
- China has spent billions building too many data centers (TechRadar, citing MIT Technology Review)
- SemiAnalysis puts China’s datacenter fleet at over 24GW (FourWeekMBA)

Leave a Reply