China's AI Bubble Just Burst: $480 Billion Gone, and the Real Test Begins

Chen Wensheng, a small business owner from Hainan province, did everything the playbook said. He watched the government promote artificial intelligence as a national priority. He read about state-backed funds taking stakes in promising startups. He saw robot maker Unitree Robotics list on the STAR Market in August — a debut so hot that retail investors oversubscribed the offering eight thousand times and the stock closed its first session 460% above the IPO price. He bought in, along with his friends, because the signal from Beijing seemed unambiguous: this was the future, and the future had official support.

Last week, Chen was nursing losses he did not want to quantify. Unitree's shares have halved since August. His broader portfolio of AI names has fared worse. The government, it turned out, was willing to let the market fall.

Chen's story is one of millions. In under four months, China's STAR50 index — the Shanghai bourse's benchmark of the country's largest semiconductor and artificial intelligence companies — has collapsed roughly 40% from its July peak, wiping out approximately $480 billion in market value. The CSI 300 blue-chip index has fallen about 20% from its June high, entering bear-market territory after ten consecutive weekly losses. Chinese equities overall have shed an estimated $3.5 trillion since their May peak. And the index, even after the crash, still trades at around 100 times trailing earnings — more than triple the Nasdaq 100's multiple of roughly 30.

This is the sharpest unwinding of AI equities anywhere in the world since the current boom began. It landed with almost no warning, and it carries implications that stretch far beyond China's borders.

The Anatomy of the Crash

The STAR50's collapse did not happen in a single session. It was a slow-motion repricing that accelerated as three distinct pressures converged: valuations that had detached from any plausible earnings trajectory, a demand story that began showing cracks, and an export-control regime that turned from headwind to gale force.

The rally that preceded the fall was extraordinary. Following Beijing's 2024 stimulus package, the STAR50 climbed roughly 230% through its June 30 closing high. At the peak, the index traded near 250 times trailing earnings — a multiple that made even the most aggressive US tech valuations look conservative. Retail investors, encouraged by state media narratives about technological self-reliance and the promise of a "slow bull" market that would reward patient capital, poured savings into AI-linked equities. By October, Goldman Sachs estimated that AI stocks accounted for roughly 42% of China's total market capitalization and 57% of all trading turnover across mainland and Hong Kong markets.

Source: Reuters, Bloomberg, Goldman Sachs, Morningstar, October–November 2026.

The trigger was not one event but a cascade. Third-quarter earnings began rolling in, and while several companies posted impressive top-line growth, the numbers were not enough to justify the multiples. Unitree reported revenue up 70% year-on-year — genuinely strong — but the stock still fell 23% from its listing price in a single week because the starting valuation had priced in a decade of flawless execution. CXMT, the memory-chip maker that notched 132 billion yuan in profit in the first nine months, lost more than a quarter of its value in October alone.

Simultaneously, Washington sharpened its tools. The US State Department sanctioned five Chinese companies connected to the founder of Shanghai Zhisuan Technology, alleging trade-secret theft. More consequentially, US officials moved toward excluding Chinese-made optical components and networking equipment from American data centers — a direct threat to Zhongji Innolight and Eoptolink Technology, two of the STAR50's former darlings, whose shares have fallen roughly 60% from their June and July highs.

"We had three things hit at once: valuations that made no sense, earnings that couldn't keep up, and the US closing the door on export revenue," said Dong Baozhen, a Beijing-based fund manager. "The stock market later evolved into a speculative crazy bull. And now we see the bubble bursting."

Retail Favorites, Retail Carnage

The damage has been widest among the retail-favorite names — recent listings that embodied the narrative of Chinese technological ascendance and drew extraordinary levels of small-investor participation.

Unitree Robotics, the humanoid robot maker whose machines can dance the yangge folk dance, shipped over 7,000 units in the first half of 2026 — more than in all of 2025, according to Counterpoint Research. The company's revenue grew 70% year-on-year. None of that mattered. After a 460% first-day pop, the stock has given back half its value, caught in a broader repricing of anything trading at multiples disconnected from cash flow.

Sources: Reuters, Morningstar, Counterpoint Research, Bloomberg, October–November 2026.

The GPU designer twins — MetaX and Moore Threads, both completely unrelated to Meta Platforms or X despite their names — tell an even starker story. Each soared on debut by 700% and 400% respectively in late 2025, riding a wave of enthusiasm for domestic alternatives to Nvidia. Neither has a meaningful earnings record. Both have halved since July, and the selloff has accelerated as investors begin to price in what Shanghai Zhuozhu Investment Management partner Wang Zhuo calls the inevitable endgame: "Just like all other Chinese manufacturers, [chipmakers and AI suppliers] will eventually end up in a price war. There's no exception. All bubbles inevitably burst."

The newly listed AI model developers have not been spared. MiniMax and Z.AI (formerly Zhipu), which went public in Hong Kong amid enormous fanfare, have lost a combined $150 billion from their highs. The market is beginning to grapple with a question that was invisible during the rally: if open-weight models from DeepSeek, Qwen, and Kimi are free and increasingly capable, what exactly is the moat for a proprietary Chinese AI application?

Beijing's Dilemma: The Market It Wanted, Crashing on Schedule

The crash puts Beijing in an extraordinarily awkward position. For two years, Chinese authorities have actively encouraged households to move savings from property and bank deposits into equities — partly to finance the technology buildout that is central to the country's strategic ambitions. Chinese households hold an estimated $11 trillion in financial assets, but only about 12% is allocated to equities, compared to roughly 40% in the United States. Every percentage point of rotation represents over $100 billion in potential buying power.

State-linked funds have responded with their heaviest buying since the April 2025 trade-war rout, stepping in to slow the decline in both the STAR50 and the CSI 300. But the intervention has so far failed to find a floor. Securities regulator Wu Qing met with market participants in late October to hear suggestions, but no new stimulus measures have been announced.

The deeper problem is credibility. Beijing wanted a bull market, but it wanted a *controlled* bull market — one that channeled savings into strategic sectors without the boom-bust cycles that have historically characterized Chinese equity markets. What it got instead was a speculative frenzy that enriched early movers and punished latecomers, many of whom bought at the top because they believed government backing was a floor under prices. Chen Wensheng in Hainan is the archetype: he invested because he believed the state supported Unitree, which had state-backed funds among its investors. That belief turned out to be worth less than he assumed.

"What is happening is now going to put a stop for many of the companies trying to get themselves listed," said Hong Hao, managing partner and chief investment officer at Lotus Asset Management in Hong Kong. The implication is significant: a pipeline of AI and tech IPOs, expected to include DeepSeek and Moonshot AI as soon as late 2026, may find the market far less receptive than it was six months ago.

The Global Mirror

For American investors, the natural instinct is to view China's crash as a contained, local phenomenon. The Nasdaq, after all, is only about 6% below its record high. US AI champions like Nvidia, Microsoft, and Alphabet trade at multiples that, while elevated, are nowhere near 250 times earnings. The systemic risk appears remote.

That instinct deserves scrutiny. The parallels are uncomfortable, and several prominent voices on both sides of the Pacific have been quick to draw them.

Sources: Reuters, Goldman Sachs, Morningstar, Motley Fool, November 2026.

Dean Baker, economist at the Center for Economic and Policy Research, put the warning plainly: "Many people investing in AI-related companies have no memories of the tech bubble and its collapse. China's plunge should remind everyone that bursting stock bubbles are not just ancient history. And yes, it can happen here."

Morgan Stanley's Asia equity team is even more specific, forecasting that offshore Chinese stocks could fall another 10% to 15% before finding a bottom. The bank's reasoning is straightforward: even at 100 times trailing earnings, the STAR50 remains three times more expensive than the Nasdaq 100, and the earnings trajectory needed to justify that premium has not materialized.

The counterargument — heard from Goldman Sachs, BNP Paribas, and HSBC — is that China's correction is a valuation reset, not a structural collapse. Goldman added Innolight and CXMT to a revised list of recommended Chinese AI stocks in early November. BNP Paribas strategist William Bratton sees buying opportunities for cautious investors. Asia-focused funds remain overweight on China even after three months of net selling, according to HSBC research. The bull case rests on a simple observation: China's AI demand is real, token consumption is growing explosively (175 trillion per day as of mid-2026), and the companies supplying the infrastructure will eventually grow into their valuations.

The bear case is equally simple: real demand and real earnings are not the same thing, and a sector that trades at 100 times trailing earnings in a market where the broader economy is grappling with excess capacity, soft consumption, and a property-market slowdown has further to fall.

What Most Coverage Missed: The Double Squeeze

Most Western coverage of the crash has framed it as a China story — an overheated local market correcting. That framing misses the more consequential dynamic: the crash is accelerating a structural transformation in how China's AI industry sources its compute, and Washington is tightening the vice at precisely the moment Beijing's capital markets are least able to absorb the shock.

Consider what happened in the first week of November, even as the STAR50 was finding its footing. Oracle, under pressure from the Trump administration, canceled a $7 billion chip-leasing deal with Tencent that would have given the Chinese internet giant access to as many as 100,000 advanced AI chips in Southeast Asian data centers. Nvidia, according to the Financial Times, told its cloud partners they must now sell at least 51% of their capacity to non-Chinese companies. And Beijing, for its part, had already pressured Tencent through Guangdong provincial officials to seek domestic chips instead.

Simultaneously — and this is the part that connects the crash to the broader AI narrative — DeepSeek confirmed plans to train its next-generation V5 model on 160,000 Huawei Ascend 950DT chips at a new data center in Ulanqab, Inner Mongolia. If successful, it would be the first frontier-scale model trained entirely on Chinese hardware, a milestone that would validate Beijing's decade-long bet on domestic semiconductor self-sufficiency.

The market crash, the export controls, and the chip pivot are not separate stories. They are three expressions of the same underlying dynamic: China's AI industry is being forced to decouple from American compute at exactly the moment when its domestic capital markets are least able to fund the transition. The $480 billion that vanished from the STAR50 was supposed to finance thousands of data centers, chip fabs, and AI startups. That capital has not disappeared from the economy — it has disappeared from the equity markets that Beijing was counting on to allocate it.

What Comes Next

Three variables will determine whether the STAR50 stabilizes or falls further.

First, the state-fund bid. Beijing's buying has been aggressive but not unlimited. If the government escalates to more dramatic measures — margin requirement cuts, stamp-duty reductions, direct ETF purchases at scale — the floor may hold. If it holds back, Morgan Stanley's projected additional 10–15% decline becomes plausible.

Second, US export policy. The optical component ban, if formalized, would remove a meaningful revenue stream for Innolight and Eoptolink and set a precedent for excluding Chinese suppliers from Western data-center supply chains more broadly. Every escalation on this front pressures the STAR50's most internationally exposed constituents.

Third, and most fundamentally, earnings. The Q3 reporting season has produced genuine bright spots — CXMT's profitability, Unitree's revenue growth — but not at levels commensurate with 100x multiples. If Q4 results show AI-driven revenue acceleration, the market may find its footing. If not, the de-rating continues regardless of what the state funds do.

The longer-term question transcends the index level. China's AI buildout continues — 175 trillion tokens per day do not flow through an industry that has stopped building. The National Supercomputing Internet is pooling 100,000 domestic accelerator cards. DeepSeek is betting its V5 on Huawei silicon. Alibaba's Qwen family has 4 billion downloads and 330,000 derivative models. The *activity* is real. What has changed is the market's willingness to fund it at any price.

In that sense, the STAR50 crash is not a repudiation of China's AI ambitions. It is a repudiation of the idea that those ambitions could be financed by retail speculation at 250 times earnings. The companies that survive this correction — the CXMTs with actual profits, the SMICs with actual fabs, the Alibabas with actual cloud revenue — will emerge with cheaper capital costs and fewer competitors. The ones that were all narrative and no cash flow will disappear. That is what markets are supposed to do, even in a country where the state would prefer they did it more gently.

Chen Wensheng, the Hainan investor, has not sold. "I believe in the technology," he said. "I just don't know if I believe in the price anymore." In that sentence, he has articulated the question that every AI investor on earth — in Shanghai, New York, Taipei, and Seoul — is now asking themselves.

Voices from the Ground

@基民博士 · Weibo

"250倍市盈率的时候我说这是泡沫，评论区全是'你不懂国家战略'。现在跌了40%，那些人删帖跑光了。国家战略是对的，但国家战略不等于你的股票能涨。两码事。"

*"When the P/E was 250x, I called it a bubble and the comments were full of 'you don't understand national strategy.' Now it's down 40% and those people have deleted their posts and vanished. National strategy is correct, but national strategy doesn't mean your stock goes up. Those are two different things."*

@美股观察员 · Weibo

"STAR50跌到100倍市盈率还有人说便宜，说比250倍便宜多了。这逻辑就像说从30楼跳下来比从50楼跳下来安全一样。问题是地面在下面。"

*"People are saying the STAR50 at 100x earnings is cheap because it's much cheaper than 250x. That logic is like saying jumping from the 30th floor is safer than jumping from the 50th. The ground is still below."*

@AI陈大牛 · Zhihu

"作为一个在芯片行业干了十年的人，我想说：国产替代是真的，技术差距在缩小也是真的，但股价透支了未来十年的进步也是真的。这三个'真的'可以同时成立。散户的问题是把第一个'真的'当成了全部。"

*"As someone who's worked in the chip industry for ten years, let me say: domestic substitution is real, the technology gap is narrowing, and the stock prices have priced in the next decade of progress. All three of these things can be true at the same time. The problem with retail investors is they treated the first one as the whole story."*

@globalmacro_daily · X (Twitter)

"China's STAR50 crashing 40% while Nasdaq dips 6% is the most important market signal of 2026 and nobody's talking about it. Either China is 6 months ahead of the US in the AI cycle, or it's a different market with different dynamics. Either answer should worry you."

*(posted in English)*

@量化小散 · Xiaohongshu

"上周清仓了所有的AI股票，亏了将近40%。入市两年攒的钱半年亏光。不怪别人，就怪自己把'国产替代'四个字当成了护身符。市场没有护身符，只有风险。"

*"Cleared out all my AI stocks last week, down nearly 40%. Lost in six months what took me two years to save. I don't blame anyone except myself for treating 'domestic substitution' like a talisman. The market has no talismans, only risk."*

@northeastern_capital · X (Twitter)

"The oracle/tencent cancellation is the tell. If US cloud cos won't even *lease* compute to Chinese firms, the decoupling is further along than anyone pricing Nasdaq at 30x wants to admit."

*(posted in English)*

Related reading:

- The Open-Weight Escape Hatch: Why Kimi K3.1 Pro's 'Free' Model Is China's Cleverest Regulatory Move

- China's AI Just Crossed the Inference Threshold: 175 Trillion Tokens a Day

- The Compute Empire: How Sanctions Built China's AI Hardware Stack

- Wuzhen Summit: Open Source and the New Rules of Global Tech