DeepSeek is turning geopolitical fear into a $71 billion IPO.
It is true that DeepSeek has earned something most AI companies have not: a technical result that stands on its own. Its recent models demonstrated that a Chinese lab could produce frontier-competitive systems at a fraction of the cost Western labs were spending, and the efficiency gains were real — not marketing, not benchmark gaming, but actual engineering choices about training methodology and compute utilization that produced measurably better output per dollar of infrastructure. That matters. It should be said plainly.
The trouble is that the technical result is not what is being offered to Shanghai’s public markets. What is being offered is the narrative — that China’s AI industry is in a race it cannot afford to lose, and that DeepSeek, having demonstrated frontier competence, deserves public capital at a valuation that would place it among the most valuable AI companies on earth. The financing is the product. The technology is the packaging.
The numbers move faster than the product. Last month it was $50 billion. This month it is $71 billion. Next year it may be listed. In its first external funding round, DeepSeek raised $7.4 billion at a valuation exceeding $50 billion — a company that, until recently, ran entirely on founder Liang Wenfeng’s personal wealth and capital from his hedge fund, as we noted when the company announced plans to double its workforce alongside the raise. Liang is now seeking $71 billion or more in a new round while preparing IPO paperwork for Shanghai, targeting a filing by year-end and a listing as early as the second quarter of 2027. In a May investor pitch, Liang warned that DeepSeek could lose momentum without the funds to retain talent and expand computing infrastructure — a candid admission that the company’s competitive position depends on continued capital injection, not on self-sustaining economics. The valuation trajectory — from self-funded to $50 billion to $71 billion in weeks — is not a measure of revenue or a path to profitability. It is a measure of the gap between what the race narrative promises and what the financial record shows.
Chinese regulators have recently relaxed rules to allow AI startups to list on Shanghai’s Nasdaq-like board even if they are not profitable. This is the structural detail that matters most. The regulatory accommodation is not incidental to the AI capital race — it is the mechanism by which speculative private valuations get converted into public-market equity. When regulators rewrite listing requirements to lower the profitability threshold for a specific industry at the exact moment that industry is seeking billions in public capital, the rule change is serving the industry’s fundraising calendar, not the investing public’s protective interest.
To be fair, the competitive argument is not fabricated. China’s AI companies do need capital to sustain frontier research, and requiring profitability as a listing condition would slow their access to public markets while their competitors proceed without comparable constraints. But the argument proves precisely as much as it is worth and no more: it justifies relaxing the profitability gate. It does not justify a $71 billion valuation for a company that only recently began accepting external capital. The regulatory change opens the door. The valuation walks through it. And the difference between a reasonable regulatory adjustment and a mechanism for converting private speculation into public-market equity is the difference between what the rule says and what the rule enables.
The beneficiaries are identifiable. Liang Wenfeng converts a position built on personal wealth and hedge-fund capital into public-market equity at $71 billion or more. Early investors — including Tencent, a state AI fund, and battery maker Contemporary Amperex Technology, all backers of the record round — see a path to liquidity. The banks underwriting the listing collect fees on what could be the largest Chinese tech IPO in years. The risk falls on Shanghai’s public investors, asked to value an unprofitable company at a price that requires it to become one of the most consequential AI operations on earth — and asked to make that bet during the brief interval when geopolitical urgency makes the question feel irrelevant.
DeepSeek is not an isolated case. ByteDance is negotiating a $20 billion loan — potentially its largest outside China — to fund AI spending. Zhipu AI — already listed — raised $4 billion through a share sale following its January IPO and has since seen its market capitalization surge more than tenfold to exceed $90 billion — a trajectory driven not by revenue growth but by the AI narrative’s momentum in Chinese public markets. That tenfold surge is the template DeepSeek’s investors are chasing. The pattern across the Chinese AI industry is consistent: convert access to capital — debt, equity, public markets — at valuations that require future dominance to justify, backed by a geopolitical urgency that makes the question of whether that dominance is achievable or profitable feel like a secondary concern.
Here it is worth being precise about what the AI race actually requires, because the public discourse treats frontier AI as a software problem — something that scales at the margins the way a mobile application scales. It is not. Training and deploying frontier AI models requires compute infrastructure: servers, chips, cooling systems, electricity at industrial scale, network architecture. These are physical assets with physical costs, not software with marginal costs approaching zero. Joe Tsai, chairman of Alibaba, said it plainly in June: “In the China context, we’re very underinvested in infrastructure and in the AI supply chain.” The Alibaba chairman is describing a construction problem — a capital-intensive, real-estate-and-energy buildout — while the industry’s marketing describes an intelligence revolution. The gap between the two descriptions is where the valuation sits.
The regulatory relaxation of the profitability requirement is a de facto bezzle-enabling mechanism: it lets the story run long enough for private capital to exit before the financial record forces a repricing. Galbraith’s concept of the bezzle — the interval between a confidence trick’s execution and its discovery, when the perpetrator has the gain and the victim has not yet realized the loss — names what these companies are engineering. Private-market valuations that depend on narrative momentum are being converted to public-market equity through IPOs, debt facilities, and share sales, at the exact moment when the race framing makes it difficult to ask what the underlying businesses are actually worth. The structure is older than the technology. It is the same pattern that converts productive capacity into financial instruments at the moment of maximum narrative leverage: value built by people who know how the thing works, extracted by capital at the peak, risk transferred to whoever holds the asset when the story ends. The bezzle runs as long as the narrative holds. The IPO is the exit from the bezzle for private investors. The public market is where the narrative becomes someone else’s liability.
The corrective is straightforward. Require what every other public listing requires: disclosure of unit economics, revenue trajectory, and the gap between current valuation and demonstrable cash flow. Let the race narrative be tested against the same financial transparency that governs every other company on the Shanghai board. If the AI race is real, the numbers will support it. If the numbers do not support it, the race framing is doing the work that disclosure should do.
The timeline is the tell. IPO paperwork by year-end. Listing as early as Q2 2027. A new funding round at $71 billion in the meantime. The urgency is not driven by DeepSeek’s readiness to be a public company — the company only recently began accepting external capital at all. It is driven by the window: the brief interval in which the race narrative carries enough momentum to sustain valuations that the financial record cannot. Windows close. When this one does, the question that geopolitical anxiety was built to defer will be the only one that matters. What is the business worth, in the absence of the fear? The answer, by every metric other than anxiety, is considerably less than $71 billion. But the bezzle does not need the answer to be right. It only needs the question to be late.