Chinese regulators responded to the July selloff in AI and tech stocks by citing 境外输入性风险 — risk imported from abroad. The phrase is an extraordinary admission from an official apparatus that normally frames market volatility as an external nuisance rather than a structural dependency. It acknowledges what the market had long shown: Chinese tech valuations move on U.S. and South Korean trading dynamics, not on domestic fundamentals.

The self-reliance narrative Beijing has promoted for years — a semiconductor supply chain that does not depend on Washington’s approval, AI models that do not rely on American chips or code — gained force when DeepSeek debuted its models to domestic acclaim. Moonshot AI’s Kimi carried a different emphasis: not just that China can compete, but that its cheap, open-source models might win. The narrative was always about technology. The market told a different story about capital.

The pricing gap

Chinese tech valuations rest on American AI benchmarks. Analysts told The Wall Street Journal that Chinese tech valuations are “more stretched than Korea’s,” resting on a domestic-substitution story that assumes China’s champions deserve to trade like Nvidia’s ecosystem without Nvidia’s numbers behind them. A Chinese fund manager told the Journal the mainland market’s swings are “driven by the AI trade in the United States.”

The valuation methodology is imported. Global market infrastructure — index providers, analyst consensus platforms, trading systems — is dominated by U.S. and European institutions that normalize U.S.-derived valuation methods. There is no independent domestic anchor, because the institutional infrastructure to produce one does not exist. China’s capital markets operate under a state-directed system where valuation benchmarks are externally sourced; the absence of an independent equity research culture reflects a legacy where market-based pricing is subordinate to state direction.

The capability underneath is genuine. China has invested roughly $142–150 billion in semiconductor capacity since 2014 — about 3.6 times the U.S. commitment, according to the Semiconductor Industry Association and Tom’s Hardware. The problem is not capital scarcity. It is the absence of institutional capacity to price Chinese AI on its own terms. The root cause is structural: the institutional architecture of China’s capital markets produces dependency on foreign pricing logic across cycles, not a one-off pricing error.

How the shock traveled

The transmission mechanism is sentiment, not money. Korean money is not a major presence in Shanghai or Shenzhen, and Chinese investors have little money in Korean stocks. When South Korean chipmakers, riding the same memory-chip windfall as CXMT, went into a speculative frenzy and then cracked, they rattled Chinese tech without a direct financial link. The shared valuation logic — the same global semiconductor-cycle data, the same AI-demand expectations — is the conduit. The relationship is one-sided: Korean swings affect Chinese markets, but Chinese developments do not symmetrically affect Korean chipmakers.

The state, through its control of the IPO pipeline, allowed CXMT — a company whose profitability depends on a commodity shortage — to go public at a peak-cycle valuation. CXMT’s $8.55 billion raise, the largest Chinese IPO in over 15 years, took the company to a $484 billion valuation. Blowout earnings already priced in barely move shares. The blowout profits came from a global supply shortage, not a business transformation. When the supply cycle normalizes — and capacity additions by Samsung, SK Hynix, and CXMT itself will see to that — the domestic-substitution narrative will face its first real test against the numbers underneath.

Who controls the price

The U.S. AI sector and Wall Street investors set the benchmarks; Chinese tech is priced in their image. South Korean chipmakers are dominant in the relationship — high power, low interest — because they affect Chinese markets through sentiment but have no reciprocal stake. The U.S. government’s export control apparatus occupies the same position: high power, low interest, the structural constraint that forced the self-reliance push, though the current market dynamics are mediated through the market rather than direct engagement.

Chinese tech companies — CXMT, DeepSeek, Moonshot AI — sit in the dependent quadrant. They have partial power (market cap, technological capability) and legitimate stakes, but no control over the sentiment drivers that determine their valuations. Chinese domestic retail investors are demanding stakeholders: no power as individuals, but their wealth is exposed to the volatility regulators acknowledged, having invested based on a self-reliance narrative that has not yet delivered financial independence. Chinese tech sector employees share that dependent fate — their livelihoods are at stake in the valuation correction, but they have no organized voice in the market discourse.

Chinese regulators and Huijin occupy the dominant quadrant formally — regulators wield authority and issued an urgent statement, while Huijin deploys billions in broad-based ETF purchases as the state’s plunge-protection team. But Huijin’s interest is partial: it responds to market stress rather than actively shaping the situation. Huijin’s ETF buying stabilizes the broad market; CXMT’s recovery depends on sentiment, not Huijin’s intervention. The real repair job — stabilizing the tech names that cratered — depends on sentiment holding steady in the United States and Korea. Huijin has a broad-market stabilization tool but no instrument that can decouple tech valuations from foreign benchmarks.

Several parties are absent or marginalized. Chinese semiconductor equipment suppliers — the companies that would need to exist for true self-reliance — are invisible in the narrative. Chinese academic AI researchers, the talent base for the entire enterprise, are marginalized by a discourse focused on commercial companies. Foreign investors in Chinese tech are filtered through intermediaries; the WSJ report focuses on the relationship between regulators and domestic sentiment, not on foreign capital flows.

Liang Wenfeng, DeepSeek’s founder, occupies an informal broker role between the AI industry and the market. His prediction at a four-hour investor session that AI could eventually account for 10% of global GDP, and his warning that companies hoarding the prize would be “abandoned by history,” is being read domestically “the way market participants once read Politburo readouts.” Liang is not a formal authority, but a voice the market treats as decisive in the absence of clear regulatory signals. That market-authority void is itself a symptom of a capital system that has not yet developed independent institutions to do the job of pricing.

Four futures, one already in motion

The question of whether China’s AI sector achieves genuine sovereignty turns on two axes that are genuinely orthogonal.

Axis X — Technological Autonomy: Does China produce genuinely independent innovation stacks, or does it remain a domestic-substitution story where valuations assume parity before it exists?

Axis Y — Market Autonomy: Do Chinese capital markets develop valuation frameworks independent of U.S. and Korean sentiment, or do they remain — as regulators themselves acknowledged — driven by “risk imported from abroad”?

A country can achieve real technical breakthroughs while its equity markets remain correlated. Japan achieved semiconductor parity in the 1980s while remaining a sentiment-taker on Wall Street — the 1987 crash’s outsized impact on the Nikkei, despite Japan’s industrial strength, illustrates the gap between technical independence and market independence.

“Genuine Sovereignty” (high autonomy on both axes). Sustained state investment yields competitive domestic chips; CXMT wins export licenses to U.S. and European customers. DeepSeek’s open-source models build installed base in non-aligned markets, creating dual revenue streams. Foreign capital returns on the strength of a genuine tech story. Huijin’s ETF purchases become unnecessary as private demand returns. CSI AI Index correlation with the Philadelphia Semiconductor Index falls below 0.4.

“Innovative Satellite” (high technological, low market autonomy). Chinese AI models win benchmarks and enterprise contracts; CXMT’s innovations scale domestically and globally. Yet correlation with Nasdaq stays elevated because investor frameworks still price Chinese tech as a derivative of U.S. peers. Huijin remains active in markets years after technical milestones are met. The fund manager’s observation — mainland swings are “driven by the AI trade in the United States” — remains accurate. The irony persists: Beijing’s independence push is “financed and validated by market logic developed in America.”

“Walled Garden” (low technological, high market autonomy). Chinese capital markets develop independent pricing logic — but that logic is the domestic-substitution narrative untethered from technical reality. Valuations assume China’s champions deserve Nvidia-like multiples without Nvidia-like numbers. CXMT’s trajectory becomes the template. Markets are autonomous but fragile, priced on story rather than substance. The independence is built on fiction.

“Double Dependency” (low autonomy on both axes). Neither independence goal is achieved. Each Korean memory shock transmits proportionally into Chinese tech. Huijin intervention frequency increases rather than decreases. The self-reliance narrative delivers neither technical autonomy nor financial stability. Every dollar of that $142–150 billion semiconductor investment sits inside a market whose prices are still made in New York and Seoul.

The current evidence — regulators’ own admission, Huijin’s reactive buying, Korean sentiment transmission without capital flows — places China squarely in the low-market-autonomy half of this matrix. The open question is whether it lands in the “innovative satellite” or “double dependency” quadrant, and the answer depends on whether CXMT’s chips actually work outside China’s borders.

Leading indicators:

  • CXMT export licenses to U.S. or European customers — if two or more are granted within 12 months, the high-autonomy integration scenario is materializing. Source: Commerce Department license databases.
  • CSI AI Index correlation with Philadelphia Semiconductor Index — sustained above 0.7 for two consecutive quarters marks the sentiment-dependence quadrant; sustained below 0.4 for two consecutive quarters marks the autonomous-narrative quadrant. Source: 90-day rolling Pearson correlation.
  • Huijin intervention frequency — increasing frequency signals the double-dependency quadrant; decreasing frequency signals the autonomous-narrative quadrant.
  • U.S. restrictions on advanced lithography equipment — new restrictions would position for the double-dependency scenario to accelerate.

Wild cards outside the matrix:

  • Taiwan contingency. A military escalation or severe coercive action in the Taiwan Strait would collapse all four scenarios simultaneously — severing access to TSMC-fabricated chips, triggering capital controls, and converting the sentiment-transmission channel from a trading phenomenon into an existential financial shock. Both axes lose their logic. Every scenario assumes the Taiwan Strait remains at its current level of managed tension. Indicator: accelerated PRC military buildup in the strait, U.S. carrier redeployment, or public pronouncements that change the status-quo language both sides have maintained since 1979.
  • Photonic or optical chip breakthrough. A breakthrough in Chinese optical computing would obsolete the current CMOS-centric framework. CXMT’s $484 billion valuation would be stranded. DeepSeek would run on entirely different hardware. The U.S.–Korea sentiment link becomes irrelevant. Indicator: sustained government-funded research publications in optical computing from Chinese Academy of Sciences labs, coupled with patent filings for photonic chip architectures.

The architecture

What is premature is the price. What is structural is the pricing logic that set it. Beijing’s response — Huijin buying broad-based ETFs to stabilize the market — treats the symptom. For the structural dependency to change, Chinese capital markets would need their own pricing logic — a domestic anchor built on independent institutional capacity. Building that capacity would require an independent equity research ecosystem and valuation methodologies separate from U.S.-derived frameworks — institutional infrastructure that does not currently exist within China’s state-directed capital system.

The structural dependency is not a market event to be weathered. It is the architecture. Until Chinese capital markets can price their own tech sector without waiting to see what Nvidia did overnight, every claim of technological self-reliance is a story Beijing tells itself while its market takes instructions from someone else.

Analytical techniques used in this piece

This analysis applies the methods below. Each links to a short, plain-English explainer you can read and reuse.

Root-Cause Analysis
Traces a symptom back along its causal chain to the conditions that actually generated it.
Scenario Planning
Builds a small set of distinct, plausible futures to plan against.
Stakeholder Mapping
Charts the parties to a situation — their interests, power, and alignments.