The Ghost Speaks: Liang Wenfeng Breaks His Silence to Tell America Its Lead Is an Illusion

"America's advantage comes only from having more computing power." After 17 months of silence, DeepSeek's founder resurfaces with a thesis designed to unsettle Washington — and it lands on a nerve

The Return of the Invisible Man

A month ago, watchchina.ai profiled Liang Wenfeng as "the ghost who controls China's AI" — the DeepSeek founder who had not appeared publicly since February 2025, whose own company's image-recognition AI could not identify his face. This week, the ghost spoke. And what he chose to say was a direct message to the United States about the true nature of its lead.

A four-hour closed-door investor talk attributed to Liang Wenfeng, published July 23, argues that America's lead comes only from having more computing power — with China now running roughly 20,000 H-equivalent cards under severe acquisition constraints. Liang reportedly said Nvidia's CUDA moat is crumbling. For a man who has built his entire mystique on silence, the decision to surface with this specific argument, at this specific moment, is itself a strategic act. The thesis is precise and deliberately deflationary: strip away the raw compute advantage that export controls are supposed to protect, and the American lead, in Liang's telling, largely disappears.

The Argument Beneath the Provocation

watchchina.ai has spent two months examining exactly this question — whether America's edge is real capability or merely raw compute — and Liang's intervention sharpens it to a point. His claim is not that China has more or better chips; he openly concedes China operates under severe acquisition constraints with a fraction of America's hardware. His claim is more subversive: that the entire remaining American advantage reduces to that hardware gap, and nothing more. No superior algorithms. No deeper talent. No fundamental research lead. Just more cards.

If true, it is a devastating framing for US strategy, because it means export controls are not protecting a durable lead — they are the only thing propping up a lead that would otherwise already be gone. And Liang's second claim, that Nvidia's CUDA software moat is crumbling, strikes at the one advantage watchchina.ai has repeatedly identified as America's most durable. As documented when Jensen Huang admitted Nvidia holds "zero percent" of China, CUDA lock-in was the deepest reason to believe American AI dominance could persist even as Chinese hardware improved. Liang is now asserting that this final moat is failing too. Whether he is right or engaging in strategic messaging, the claim is calibrated to erode confidence in precisely the two pillars — compute superiority and CUDA lock-in — that the American position rests on.

Why the Ghost Chose to Speak Now

watchchina.ai will not take Liang's self-interested framing at face value, because the timing reveals the motive. His remarks surfaced the same week Chinese Vice Foreign Minister Ma Zhaoxu visited Washington seeking clarity on upcoming US-China AI talks, as the Trump administration escalated sanction threats — and the same week that, at the APEC ministerial in Chengdu, all 21 economies including the US endorsed "trusted open-source approaches" in a statement chaired by China's own MIIT minister. Liang's message serves Beijing's negotiating position perfectly: it tells Washington that its leverage is weaker than it believes, that its controls merely delay an inevitable convergence, and that China negotiates from a position of underestimated strength.

The honest reading holds both possibilities at once, as watchchina.ai always strives to. Liang may be right that America's lead is thinner than it looks — Stanford's 2.7% gap and China's frontier releases on domestic chips lend the thesis real support. Or he may be doing exactly what a brilliant strategist does before a negotiation: projecting confidence to shift the other side's expectations. Most likely, it is both — a genuine analytical conviction, deployed at a moment calculated for maximum strategic effect. What is not in doubt is the significance of the messenger. When the most secretive, most respected figure in Chinese AI breaks 17 months of silence to tell the world that America's advantage is an illusion propped up by chips alone, it is worth listening carefully — not because he is necessarily right, but because he clearly believes the moment has come to say it out loud. The ghost does not speak often. When he does, it is because he has calculated that the words themselves are a weapon.

Sources: AI Weekly, Bloomberg, CNBC, South China Morning Post