The explosion of artificial-intelligence investment in the United States has outrun the technology’s ability to generate cash. What began as a productivity story has mutated into a leveraged, debt-dependent capital cycle whose centre of gravity lies not in end-user demand but in a closed loop of supplier financing, hyperscaler capex and inflated equity valuations. For Chinese regulators, asset managers and policymakers, the imperative is no longer to ask whether a bubble exists, but to gauge how violently its bursting could transmit across borders — and to pre-empt the same pathologies taking root at home.
Market Distortion: When Seven Stocks Become the Market
Concentration risk in American equities has reached historic extremes. The seven mega-cap technology names — Microsoft, Apple, Nvidia, Alphabet, Amazon, Meta and Tesla — together account for an estimated 35 per cent to 40 per cent of the S&P 500’s total market capitalisation, and contributed roughly 85 per cent of the index’s gains in the first half of 2026. Nvidia alone is valued above US$5 trillion, representing about one-fourteenth of the combined market cap of every listed US company — a sum equivalent to the total value of the smallest 90–95 per cent of US listed firms.
South Korea offers an even starker mirror. Samsung Electronics and SK Hynix make up more than half of the KOSPI’s weight, effectively dictating the direction of the entire Korean market. In the first half of 2026 the two companies alone supplied about one-third of Korea’s real GDP growth, and nearly 80 per cent of the increase in the nation’s corporate operating profits. Strip them out and the rest of corporate Korea’s profit growth turned negative.
This “single-leg” posture means that any shock to a handful of balance sheets can instantly become a systemic event. The lesson for Chinese authorities is direct: monitor and, where appropriate, guide against the over-concentration of capital, liquidity and valuations into a narrow set of sectors, and encourage diversified allocation that lets traditional industries share in capital-market gains.
The Valuation Scissors: AI Sucks the Oxygen Out of Everything Else
While AI and semiconductor equities reach ever-higher sales multiples — OpenAI is said to trade above 30× sales, Nvidia around 20×, SpaceX near 100× at its June IPO, and select AI application names between 10× and 25× — the median sales multiple for traditional S&P 500 constituents sits at just 2–4×. This widening gap is not benign.
Resources are being pulled from traditional sectors on two fronts. On the capital side, investors crowd into AI both to capture upside and to hedge against the risk that AI disrupts the very businesses they would otherwise own. On the real-economy side, AI and semiconductors are consuming disproportionate shares of land, power, engineering talent and policy bandwidth. The result is a deepening “liquidity siphon” that starves non-AI firms of both financing and growth narratives.
Chinese regulators should resist the temptation to let AI hollow out the rest of the economy. AI is meant to empower all industries, not to empty them. The policy challenge is to channel capital market buoyancy toward a broad set of sectors rather than allowing one theme to monopolise liquidity.
The Mechanics of Financialisation: Circular Financing
Beneath the lofty valuations sits a structure that veteran analysts increasingly compare to the vendor-financing arrangements of the late 1990s. The pattern runs along the chain semiconductor vendor → cloud service provider → AI lab, with each party underwriting the other’s spending:
- Nvidia, acting as the “central bank” of the AI era, invests in cloud providers and backs their procurement with balance-sheet guarantees, enabling customers like CoreWeave to raise debt and buy Nvidia GPUs.
- In the “Stargate” project, Oracle, OpenAI and Nvidia have constructed a triangular loop of equity, leasing obligations and GPU purchase commitments measured in the hundreds of billions of dollars.
- Microsoft, Google and Amazon are simultaneously ramping capex and providing compute to AI startups, creating a self-reinforcing book of orders that translates into headline revenue for chipmakers.
Because the entire edifice depends on continuous access to equity and credit markets — and on the assumption that regulators will not allow it to fail — it is, in the words of one observer, “capital-driven, financially dominated, and decoupled from real consumption”. Should the downstream paying capacity of AI labs disappoint, the revenue, profit and valuation of the whole chain could unwind at once. China’s regulatory response must include strict scrutiny of related-party transactions, circular investments, and the independence of firms whose upstream and downstream counterparties are tightly intertwined.
Capex Runs Ahead of Cash Flow
The mathematics are stark. OpenAI, with annual revenue around US$20 billion, has already committed to more than US$1.4 trillion in compute procurement — a capex promise roughly 70 times its annual sales. The four largest US hyperscalers are on track to spend over US$1 trillion in 2026 alone. Across the next five years, total US AI infrastructure spending is projected at US$5 trillion or more; of the estimated US$3.5 trillion funding gap, about 40 per cent is expected to come from corporate bonds, 30 per cent from private credit, 20 per cent from asset-backed securities and leveraged loans, and only 10 per cent from equity. In other words, more than 90 per cent of the shortfall rests on debt.
This is the most dangerous and least discussed feature of the current cycle. Unlike the dot-com bubble of 2000, which lived almost exclusively in equity prices, today’s AI buildout — with its data centres, chips and power infrastructure — is fundamentally a credit phenomenon. Banks, insurers, pension funds and bond mutual funds are all exposed. When a bubble resides in equity, losses stay largely on the equity holders’ balance sheets; when it resides in credit, the damage radiates into the financial system itself.
Graham Allison, the Harvard professor who coined the “Thucydides Trap”, calls America’s AI strategy a “civilisational wager” — the country’s economic growth, fiscal outlook and national security all riding on a single card. If the productivity miracle fails to materialise, he warns, the consequences could exceed those of 2008 and approach the scale of 1929. A New York Times estimate puts potential US wealth destruction at up to US$20 trillion.
The Policy Put and the Moral Hazard
The US Federal Reserve finds itself trapped. Normally it would tighten to cool an overheating economy; today, raising rates risks puncturing the AI bubble directly. Compounding this, the market broadly believes that Washington will not allow its strategic AI champions — be they Oracle, OpenAI or Nvidia — to fail. This implicit “policy put” encourages ever-more-aggressive leverage, precisely because investors reason that the government’s commitment to winning the AI race against China amounts to a blank cheque.
No previous American bubble — not the railways, nor the Manhattan Project, nor the internet, nor housing — has been so thoroughly fused with national security doctrine. The resulting moral hazard is profound: the more certain the market is of a bailout, the larger the bubble grows, and the heavier the eventual bill.
Spillover to China: Three Channels
Even with managed capital flows, China cannot insulate itself from a US AI collapse. The transmission operates through three channels:
1. Overflow (溢出效应). The narrative power of Silicon Valley and Wall Street is immense. A US-led frenzy easily breeds imitative speculation in Chinese markets, inflating valuations of domestic AI names beyond fundamentals and importing a bubble from abroad. Chinese policy has always positioned AI as a tool to empower all industries, not as a single dominant theme; the capital market should reflect that diversified reality.
2. Linkage (联动效应). Global capital is deeply interconnected. A sharp repricing of US AI and semiconductor equities would, through capital flows, sentiment, competitive expectations and supply-chain revisions, weigh on China’s technology, semiconductor and cloud computing listings in the short term — even if China’s real-economy position continues to strengthen.
3. Reflexivity (反身性效应). This is the most paradoxical channel. China’s own AI advancement could itself be the pin that bursts the US bubble. Chinese models, offered on open-source licences at prices 5–10 per cent — sometimes as low as 1 per cent — of their US counterparts’ API pricing, are already capturing share. DeepSeek alone accounts for 16.3 per cent of API compute consumed by American enterprises, and Chinese models collectively supply around 40 per cent of US corporate API usage. Through the “Digital Silk Road” and platforms such as the World AI Cooperation Organisation, Chinese firms are exporting full-stack AI solutions — models, cloud, chips, energy, infrastructure — to Southeast Asia, the Middle East, Latin America and Africa, directly attacking the “high-price, closed-source, global-toll” business model that underpins US AI valuations.
Yet reflexivity cuts both ways. A collapse in US AI and semiconductor valuations, triggered in part by Chinese competition, would immediately reverberate through China’s own AI and chip equities via global sentiment and capital-flow channels. Beijing must therefore run two plays at once: press the competitive advantage aggressively, while building buffers against the blowback.
Retail Investors and Social Risk
Korea provides a grim preview of the social cost. In the first half of 2026, buoyed by the memory-chip and AI narrative, the KOSPI doubled; retail investors piled in with record margin debt. From the June peak to August, the index fell nearly 40 per cent, erasing about US$2 trillion in market value. Nearly half of Samsung’s retail shareholders and almost 70 per cent of SK Hynix’s retail shareholders ended up in the red; thousands faced financial ruin. America is not immune — the wealthiest 10 per cent of US households own the bulk of equities and drive about half of consumer spending. A severe AI-driven drawdown would crush the wealth effect, depress consumption, and trigger a feedback loop of “stock decline → consumption decline → economic decline → further stock decline.” Even Americans who do not directly trade would feel it through pension allocations heavily weighted toward AI megacaps.
For Chinese authorities, the message is to monitor retail participation in concentrated sectors, conduct stress tests, and treat the mitigation of household wealth destruction as a policy objective in its own right.
What China Should Do
The prescriptions that follow from the analysis are concrete:
- Diversify, don’t concentrate. Guide capital toward a balanced set of economic engines — traditional, emerging and frontier — rather than allowing AI to monopolise liquidity, valuations and policy attention.
- Scrutinise circular structures. Tighten oversight of related-party transactions, cross-investments, and customer concentrations that mask a company’s true revenue quality.
- Anchor capex in returns. Resist the financialisation of industry; ensure that capital spending is justified by realistic payback periods and not by the imperative to sustain valuations.
- Watch the debt pivot. Track the extension of AI capex into credit markets — both direct and intermediated — and build appropriate risk controls.
- Preserve policy independence. Avoid creating an implicit government guarantee for any single industry; manage expectations to prevent moral hazard.
- Stress-test retail exposure. Map household leverage against concentrated themes and prepare circuit breakers before — not after — a drawdown.
The technology of artificial intelligence may indeed be boundless. But beneath today’s euphoria lies a financial lever of potentially infinite depth — and finite patience. As the US AI investment wave completes its journey from technological breakthrough to capital-driven, debt-laden financial game in the space of a few short years, China’s task is to harvest the productive promise of AI while refusing to import its financial fragility. The bubble across the Pacific is not yet burst. But it is no longer a question of if it will deflate, only of when — and how well-prepared Beijing will be when the tide goes out.