U.S.-China Relations
NextFin News - While Wall Street still debates whether the artificial-intelligence boom is overvalued, China's AI bubble has already burst. The mainland's STAR50 index of AI and semiconductor leaders has fallen roughly 40% in less than four months, wiping out about $480 billion in shareholder value — even as the Nasdaq has slipped only about 6% from a record high set three weeks ago. The divergence is the point: the world's second-largest economy has just run through a full boom-bust cycle in AI equities while American tech stocks merely wobble.
The sell-off carries no single obvious trigger, but it arrives with valuations that had detached from any earnings reality. The STAR50 still trades at around 100 times trailing earnings, down from a peak of nearly 250 — more than three times the Nasdaq 100's multiple of 30. Individual names are more extreme still: Cambricon Technologies, China's flagship AI-chip designer, changed hands at 521 times earnings compared with 50 for Nvidia, LSEG data showed, even after its shares fell 14% in a week. The bubble did not merely deflate. It popped first, and hardest, in China.
The Anatomy of a Popped Bubble
The numbers trace a classic speculative arc. China's AI rally ran for two years, lifted by stimulus promises and a state-backed push to steer household savings into equities that would fund the country's technology ambitions. AI-related stocks swelled to 42% of the combined market capitalisation of mainland A-shares and Hong Kong H-shares, up from 24% in 2024, and accounted for 57% of turnover, Goldman Sachs found. Turnover rising faster than market value — 57% versus 42% — is the fingerprint of a retail-driven market: the theme was not just owned, it was traded.
The retail footprint is visible in the IPO frenzy. Robot-maker Unitree's August listing was oversubscribed 8,000 times by individual buyers; the shares have since halved. Chipmakers MetaX Integrated Circuits and Moore Threads Technology, which surged 700% and 400% respectively on their market debuts last year, have each given back about half their value since July. Optical-parts suppliers Zhongji Innolight and Eoptolink Technology — once the favoured picks of the AI-datacentre trade — are down roughly 60% from their June and July peaks. Memory-chip maker CXMT has declined 27% from its post-listing peak.
"The stock market later evolved into a speculative crazy bull," said Dong Baozhen, a Beijing-based fund manager. "And now we see the bubble bursting."
The reversal has slammed the brakes on Beijing's effort to engineer a "slow bull" — a steady, reliable market meant to replace China's long boom-bust pattern. The broad blue-chip CSI 300 has dropped about 20% from highs made in June and entered bear-market territory after 10 consecutive weeks of losses, the longest losing streak in the index's history, evidence that the AI unwind is spilling into the wider market rather than staying contained in one overheated corner.
Why China Popped While the US Merely Wobbles
The first-order explanation is valuation. The mechanism runs deeper: China's AI rally was financed differently, by different investors, for different reasons than America's.
The US AI boom was built on actual cash flows from a handful of cash-rich hyperscalers — Microsoft, Alphabet, Meta, Amazon — spending hundreds of billions on datacentres against visible revenue growth. China's version was built on import substitution. With US export curbs blocking access to Nvidia's most advanced chips, Beijing framed domestic AI-chip makers as a national-security necessity, and investors priced them as monopolies-in-waiting rather than as companies. Cambricon's 521-times earnings multiple was not a bet on this year's profits; it was a bet on a geopolitical outcome.
That bet is now colliding with earnings reality. Cambricon reported third-quarter earnings per share of 2.23 yuan against a forecast of 2.79, a 20% miss, while revenue of 3.53 billion yuan fell 57% short of the 8.14 billion yuan consensus. A company trading at more than 500 times earnings cannot survive a revenue miss of that magnitude without a multiple reset — and the reset is what the 40% index decline represents. The earnings miss matters less for what it says about one quarter than for what it reveals about the pricing model: at 500 times earnings, there is no margin for error, only margin for disappointment.
The second-order transmission channel is the funding model itself. China's strategy was to use a buoyant equity market to channel household savings into technology companies, reducing reliance on bank lending and state subsidies. That model requires rising prices to keep working: retail investors buy because prices go up, and companies list because valuations are high. Once prices stop going up, the mechanism reverses. Fundraising by tech firms is likely to slow as retail investors nurse losses and rotate into defensive assets. "What's 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.
Concentration risk magnifies every move. With AI stocks at 42% of market value and 57% of turnover, a decline in the theme drags the entire market down — which is exactly what the CSI 300's 20% fall from June shows. A 40% drop in a narrow index would normally be contained; in China's case, the narrow index is the market's engine room. The STAR50's weight in China's market is more than double what technology represented at the peak of the 2000 dot-com mania in most developed markets, leaving no diversified corner to hide in.
There is also a geopolitical overlay with no US equivalent. Washington is moving toward banning Chinese optical equipment from US datacentres, and the threat of broader restrictions on Chinese technology hangs over exporters. Chinese AI exporters gained 126% from early 2025 through June 2026 before falling 23% from their peak in the global AI selloff, Goldman Sachs noted — a reminder that the most globally exposed names were the first to give back gains when sentiment turned. Domestic-focused AI companies have held up better since the correction, which is small comfort when the domestic leg of the trade is the one carrying the richest valuations.
The Counter-Thesis: Shakeout, Not Collapse
The strongest case against the bubble-popped narrative is that this is a healthy correction in a structurally sound boom. Domestic AI demand in China is real and growing. Import substitution is not a fantasy — it is policy, backed by state capital and procurement. Some of the biggest names have already been punished enough to look cheap, and sophisticated money is still on the buy side: Asia funds remain generally overweight China despite being net sellers over the past three months, according to HSBC research, and Goldman Sachs has placed names including Innolight and CXMT on a revamped recommendation list. BNP Paribas Securities strategist William Bratton has said the market offers buying opportunities for the careful.
There is arithmetic behind the optimism. Since the "DeepSeek moment" in early 2025, nearly $5 trillion in market value has been created in Chinese AI-related public equities through share-price gains and IPOs, Goldman Sachs calculated. A market that has added that much value is not a mirage; it is a newly capitalised industrial base. The sell-off has also done something constructive: it has separated the companies with real revenue from the ones with only a narrative.
The counter-argument holds weight — up to a point. It correctly identifies that demand is real and that policy support will not disappear. But it mistakes a real industry for a fair price. An industry can be genuinely important and its stocks can still be uninvestable at 500 times earnings. The question is not whether China's AI sector will grow; it is whether the companies now carrying the theme's valuation can grow into it before the funding model breaks. And the funding model is already showing stress: fundraising by tech firms is likely to slow as retail investors nurse losses and rotate into defensive assets.
The valuation gap also cuts against the import-substitution premium. Supporters argue that Cambricon deserves to trade above Nvidia because it is the domestic alternative in a sanctioned market. Guangdong-based hedge fund manager Xu Qiongna has made exactly this case, arguing that potential rapid growth anchored by China's moves to replace foreign technologies could justify Cambricon's valuation even though it is "much more expensive than Nvidia." But a premium for scarcity is not a premium for profitability. If the domestic champions cannot convert policy support into earnings at anything close to their multiples, the "strategic value" argument becomes a value trap with a patriotic label.
Cyclical or Structural: The Call That Determines the Conclusion
Every market crash invites the same question: is this a cyclical fluctuation that will revert, or a structural shift that will not? For China's AI equities, the answer is both — and confusing the two is the most common error investors make here.
The cyclical leg is clear and will revert on its own. Sentiment has swung from euphoria to despair; retail investors have been burned; leverage has been squeezed out by regulators; and the index has fallen 40% in four months. Historically, sentiment extremes of this kind mean-revert. The CSI 300's 10-week losing streak is the kind of exhaustion that sets up a technical bounce, and the STAR50's valuation has already compressed from nearly 250 times earnings to 100 — a two-thirds multiple reset without the index going to zero.
But the structural leg is what makes this different from an ordinary correction, and it will not self-correct. The funding model that financed the boom — rising retail-driven prices feeding IPO supply feeding more retail participation — has broken, and it cannot be restarted by waiting. The concentration of AI at 42% of market value and 57% of turnover was itself the distortion, and unwinding it means capital flows out of the theme for a sustained period, not a quarter or two. The earnings reality at the top of the pyramid — a 57% revenue miss at the flagship name — shows that the pricing model was structurally wrong, not just temporarily pessimistic.
This is the distinction that matters for investors. The cyclical leg says a bounce is coming; the structural leg says the bounce will not carry the index back to its peak in any reasonable timeframe. A 40% decline from a bubble is not a buying opportunity if the starting multiple was 250 times earnings; it is the market discovering where the floor actually is. The STAR50 at 100 times earnings is cheaper than it was, but it is not cheap — and a market that has been priced for monopoly profits in a competitive manufacturing economy has further to fall than one that was merely overbought.
What Comes Next: Three Time Horizons and Three Scenarios
Short term, the path is set by technicals and sentiment rather than fundamentals. The STAR50's quarterly index rebalancing could force up to 8 billion yuan (about $1.1 billion) of passive outflows from funds tracking the benchmark, according to Reuters estimates — a mechanical seller into a falling market. Cambricon's shares, down 14% last week before bouncing 10%, illustrate the whipsaw that index-driven flows produce. Expect volatility, not direction.
Medium term, the test is earnings. The sector needs a sequence of quarters in which revenue growth catches up to valuation, compressing multiples through the denominator rather than through falling prices. The falsifying signal for the bubble-popped thesis is specific: if the STAR50's forward price-to-earnings ratio compresses toward 50 times on rising earnings — rather than falling further on shrinking multiples — the correction would qualify as a healthy shakeout rather than a structural deflation. Until that happens, the burden of proof sits with the bulls.
Long term, the structural question is whether China can fund its AI ambitions through equity markets without recreating the boom-bust pattern it set out to escape. Here the verdict is still open, but the early evidence is unkind. "Overemphasising the 'technology content' — that is, the weighting of tech companies by market value in major indexes — leads to a deviation, even a distortion, in capital behaviour," said Leo Wang, a partner at Beijing-based Qianchuang Capital. "This created a severe bubble in certain sectors and a serious drain on others."
Three scenarios frame the path ahead:
- Base case. The STAR50 grinds lower or sideways as multiples compress toward 50–60 times on weak earnings, a drawn-out deflation rather than a V-shaped recovery. Tech fundraising stays muted, and retail participation remains subdued. This is the "bubble popped" outcome, and it is the one the current data supports.
- Upside case. A coordinated policy response — fresh stimulus, direct state buying of equities, or a relaxation of the crackdown on leverage — reignites the rally before earnings have caught up. This would be a policy-driven rebound, not a fundamentals-driven one, and it would recreate the very volatility Beijing promised to eliminate.
- Downside case. Washington escalates export restrictions into a broader ban on Chinese AI equipment in Western datacentres, hitting exporters and dragging the CSI 300 deeper into bear-market territory. The 20% decline from June would then look like the first leg of a larger unwind.
The China Securities Regulatory Commission has been cracking down on leverage and high-frequency trading, and chairman Wu Qing met market participants earlier this month to solicit suggestions on improving capital markets. But regulation arrives after the fact; it cannot manufacture the steady returns that a "slow bull" requires. Some in the investment community say the boom has been bad for the market's structure, concentrating capital in a theme that could not support its own weight.
The Bottom Line
China's AI bubble has popped because it was inflated by a different set of forces than America's: import-substitution geopolitics instead of hyperscaler cash flow, retail speculation instead of institutional accumulation, and index concentration so extreme that the theme became the market. The 40% decline and $480 billion wipeout are the price of that difference.
For global investors, the lesson is not that AI is a bubble everywhere. It is that bubbles are local. The same technology can be a genuine revolution in one market and a speculative mania in another — and the gap between the two is measured not in models or parameters, but in valuation, ownership, and the durability of the funding model behind the rally.
"Just like all other Chinese manufacturers, (chipmakers and AI suppliers) will eventually end up in a price war," said Wang Zhuo, a fund manager and partner at Shanghai Zhuozhu Investment Management. "There's no exception. All bubbles inevitably burst."
The question now is what grows in the cleared space. China's AI industry is not going away. But the market that priced it as if it already owned the future has just been reminded that the future has to be earned, quarter by quarter.
Explore more exclusive insights at nextfin.ai.
