The fundraising rush out of China’s AI sector reads, on its surface, as a geopolitical sprint — six startups preparing IPOs in Shanghai or Hong Kong through 2027, ByteDance negotiating a $20 billion bond, Tencent raising $4.7 billion, DeepSeek seeking a valuation north of $70 billion on its way to a Shanghai listing next year. The Wall Street Journal presents this as a race, China versus America, a war chest for a computing-power arms race. The framing is familiar, the story almost comfortable — two great powers sprinting toward artificial general intelligence, with national prestige and security hanging on who gets there first.
It is worth being precise about what is actually happening, because the geopolitical narrative is doing real work that benefits specific people, and it is worth naming who.
The Journal’s own reporting contains the tell, buried in the fourth paragraph of the substantive analysis: “Executives at OpenAI and Anthropic are sounding the alarm about the rise of inexpensive AI, including some of China’s latest models, suggesting they present unacceptable security risks without regulation.” Read that sentence again. The alarm is not about AI that is dangerously powerful — it is about AI that is inexpensive. The security-risk frame is the packaging around a price war that OpenAI and Anthropic are losing. Chinese models have begun outperforming American ones at benchmarks, and the response from the companies with the most capital sunk into expensive infrastructure is to lobby for restrictions that would prevent Americans from buying the cheaper product. This is not new. It is the oldest play in the protectionist playbook, dressed in the vocabulary of national security.
The Doctorow analytical framework names this operation with precision. When a firm cannot compete on product or price, it competes on regulation — the four forces that constrain platform abuse include government regulation, and when a firm captures that regulatory apparatus, the force runs in reverse. OpenAI has raised more than $100 billion in funding commitments this year alone. That capital is deployed against a specific bet: that the compute-intensive, closed-weight, high-burn-rate model will dominate. Chinese firms offering comparable capability at a fraction of the cost — Moonshot AI’s latest release jolted global markets last week precisely because it demonstrated this — threatens to make that bet look like what it is: an overbuilt bubble whose investors need tariff protection to recover their costs.
The “bezzle” — Galbraith’s term for the interval between a fraud’s commission and its discovery, when the perpetrator has the money and the victim does not yet know he has lost it — operates across borders. The Journal documents CXMT, a memory-chip maker, doubling its fundraising target to $8 billion with an $85 billion valuation. It documents a securities regulator warning brokerages that “the AI rush might be getting out of hand.” It documents Z.AI’s market capitalization peaking at $150 billion before falling “sharply after local rivals released more powerful models” — a competitive dynamic that, in a genuinely functioning market, would be evidence of healthy rivalry, but in a bubble economy driven by state-directed patient capital and IPO mania, is evidence of a market where valuation is untethered from revenue. “There is sufficient liquidity for the strongest offerings,” a Saxo Markets strategist told the Journal, “but probably not enough to support every company at every proposed valuation.” The understatement does the work.
Here is what the engineering-substance analysis reveals beneath the geopolitical framing. The Chinese firms are raising money to secure Nvidia chips and computing power from overseas cloud-service providers before the United States tightens export restrictions. The American firms are lobbying for those restrictions to be tightened. The circularity is the point: restrict supply of computing hardware, raise the cost of competitive AI training, protect the margins of firms that already have the hardware — firms like OpenAI, which secured its $100 billion before the restrictions arrive. The chip-export-control regime functions not as a security measure but as a moat. It is John Deere’s parts-pairing lock on a continental scale: if you cannot buy the chip, you cannot compete, and the company that controls access to the chip controls the market.
Xi Jinping, for his part, is running a parallel extraction. His speech to the science congress — “we must smooth out corporate financing channels, and guide capital toward early-stage startups, smaller enterprises, long-term investments and core hard tech” — is the language of industrial policy. State-owned financial institutions pledged to hold long-term stakes in listed AI companies. Seventy companies are in the Hong Kong listing pipeline. A shop owner in southwest China won a lottery to buy shares of a memory-chip IPO and told the reporter, “I feel like I’m doing my part for China’s tech.” The open-source maneuver that Xi endorsed last week is the other side of the coin — while Washington restricts access to hardware, Beijing offers access to models, building dependency on Chinese infrastructure the way American platforms built dependency on American infrastructure. Two extraction machines, running the same playbook from opposite ends. On the Chinese side: a retail investor wins a lottery to buy 500 shares of a memory-chip IPO and describes the purchase as patriotic duty — the state directs capital into a bubble and distributes the downside to citizens who think they are funding national greatness. On the American side: OpenAI’s $100 billion war chest buys lobbying for export controls that lock cheaper Chinese models out of the market, and the security-risk frame gives the lock a public-interest gloss — the state restricts competition and distributes the upside to firms that arrived first. Same machine, different intake valves.
The structural analysis is not complicated. A technology that the field’s own researchers describe as expensive to train, unreliable in deployment, and incapable of meeting the specifications its promoters advertise has attracted, across two national economies, hundreds of billions of dollars in capital commitments. The Chinese side is channeling state money into a bubble through IPO mechanisms that let retail investors — shop owners who win lotteries — absorb the risk. The American side is channeling private money into a bubble and lobbying for export controls that function as a floor under valuations by eliminating cheaper competition. Both sides describe this as a race. It is a race, in the sense that two people running toward the same cliff are racing.
The Journal, to its credit, reports the facts that let you see this. The $150 billion peak-to-trough collapse of Z.AI. The regulator’s warning. The fundraising targets doubling on investor mania rather than revenue growth. The CXMT retail investor who describes buying chip stock as patriotic duty. What it does not do is connect these facts to the larger structure, because connecting them would require saying what the facts plainly show: that the U.S.-China AI competition frame is, in significant part, a mechanism by which firms on both sides of the Pacific extract capital from publics and governments while lobbying for regulatory architectures that protect their positions from competition.
China’s recent moves to limit overseas access to its own models completes the symmetry — Washington restricts hardware exports to China; Beijing restricts model exports from China. Two governments, acting in the name of national security, constructing a duopoly that serves the incumbents on both sides. The open-web principle — that data and services should flow between willing senders and willing receivers without a government or corporate intermediary inserting itself — is dead on arrival in this arrangement. The end-to-end principle was killed by the same forces that kill it everywhere: monopoly interests captured the regulatory apparatus and rewrote the rules to criminalize the competition.
The fundraising rush is real. The IPOs are real. The chip shortages are real. None of that changes the underlying arithmetic: the money pouring in exceeds the revenue coming out by orders of magnitude, and the security framing is how the people running the bezzle justify the extraction to the people funding it. Altman sounds the alarm about Chinese AI. Xi sounds the alarm about American restrictions. Both alarms serve the same function — they make the fundraising pitch easier.
The bezzle closes when the public stops funding it. When the Bureau of Industry and Security opens its next comment period on export-control rules — these dockets come around regularly — the comment portal is where that withdrawal can become a legal record. Submissions that name the specific economic interest being protected, not the security interest being invoked, are the ones that subsequent administrations have to read.