The release of two Chinese artificial intelligence (AI) models ranking with the most advanced American systems, timed to coincide with the World Artificial Intelligence Conference in Shanghai, has provoked panic in Washington and Silicon Valley, exposing the fragility of the speculative bubble on which the entire American stock market now rests.
On Thursday, the day before the conference opened, the Beijing-based firm Moonshot AI released Kimi K3, which, in the words of the Wall Street Journal, the principal organ of American finance capital, “jolted global markets” and is “seen as ranking with the world’s best.” Three days later, Alibaba previewed Qwen 3.8 Max, claiming to have surpassed Kimi K3.
“China’s Top AI Event Delivers Message to the U.S.: We’re Coming for You,” ran the Journal’s headline from Shanghai. Anthropic’s head of national security policy, Tarun Chhabra, said only last week that the American lead would have run 12 to 18 months had Chinese firms not trained their models on the outputs of American systems. The lead is now measured in weeks: Only Claude Fable 5 and GPT-5.6 Sol, released in June and July, still rank above Kimi K3.
What has provoked the crisis is the openness of the new models. Both are “open weight”: Moonshot will publish the full weights of Kimi K3 by July 27, with Alibaba promising the same. Once the weights are released, anyone can download, inspect, modify and run them on their own hardware, owing nothing to the American AI corporations.
The entire aim of the American AI industry has been to monetize and monopolize this technology. The hundreds of billions invested and the trillions projected for data centers rest on the premise that frontier artificial intelligence will remain a scarce, monopolized commodity from which a handful of corporations extract rents. If the highest quality AI models can be freely reproduced, their price must fall toward the basic cost of the electricity and silicon needed to run them. What evaporates is the rent, and the rent is all that sustains the valuations of OpenAI and Anthropic, both of which are preparing public stock offerings.
Investors’ faith that the American companies would keep their hold on the frontier, the Journal writes, “has been at the heart of the boom, helping to justify trillions of dollars in spending on infrastructure.” Yet the entire edifice is unprofitable. According to figures compiled from SEC filings and earnings reports, the major American technology companies have poured roughly $1.4 trillion into AI against roughly $700 billion in cumulative revenue, and only Nvidia, which sells the hardware the others burn through, operates in the black.
AI-linked stocks now account for a record 45 percent of the market capitalization of the S&P 500 and have supplied roughly three-quarters of the index’s returns over the past three years. Excluding AI, the American stock market has barely risen at all: Virtually its entire advance in recent years has been a bet on the permanence of the AI monopoly—a colossal mass of fictitious capital, paper claims on future monopoly profits that the arrival of comparable open models is now rendering worthless.
The AI bubble can burst at any moment, and the events of the past week indicate that the moment may come sooner rather than later. On Friday, the Nasdaq fell 1.4 percent, leaving it roughly 6 percent below the record it set six weeks ago, while the Philadelphia Semiconductor Index completed its worst week in more than 15 months, down 12.5 percent. Whether the present slide marks the onset of the collapse or merely another tremor cannot be foretold.
Rarely do the defenders of private property state their case so frankly as Dean Ball, a former Trump administration official who is now OpenAI’s “head of strategic futures.” A world dominated by open-weight models, Ball declared on X, would amount to “full AI communism,” in which artificial intelligence would cease to be “a market product” and become instead a “public good” provided by the state, a prospect he denounced as a “dystopian hellscape.”
The statement is an inadvertent confession. What terrifies the AI oligarchs is that artificial intelligence will slip the leash of private ownership and become what it is already in essence: a social product, created from the knowledge, language, and labor of all mankind, and belonging by right to all mankind.
This is why the oligarchs now seek to criminalize open-weight models, a campaign that is already well advanced. Security officials in the Trump administration, the Journal and Axios report, have weighed trade blacklists, federal security warnings and an executive order targeting open-weight models, with internal disagreement so far preventing action.
The rise of Kimi has revived these efforts, and the national security faction appears to be ascendant. Short of a formal ban, procurement rules and public pressure campaigns are being prepared to drive American companies and engineers off the Chinese models, which they have adopted as cheaper and nearly as good. The aim is to force them back onto American companies’ models in order to prop up their valuations.
The state has already exercised this power against an American firm. In June, a Commerce Department export directive forced Anthropic to withdraw Claude Fable 5, its most powerful model, for 19 days, demonstrating that the capitalist state will assert direct control over this technology whoever owns it.
The hypocrisy of these maneuvers is boundless: American imperialism has spent decades accusing its rivals of unfair trade practices and state subsidies, from steel to aerospace, yet the American AI corporations are themselves creatures of the state, built on decades of publicly funded research and fattened on federal contracts, chip subsidies and tax abatements.
Sam Altman, Dario Amodei and Demis Hassabis, who head OpenAI, Anthropic and Google’s DeepMind, built their fortunes on the mythology of the untrammeled free market and now clamor for heavy handed state intervention against competitors they cannot defeat. The Trump administration, a government of oligarchs, will shape whatever measures emerge to loot the public and enrich the financial aristocracy.
The alarms have set the oligarchs publicly against one another, with David Sacks, the White House AI adviser, denouncing Ball’s post as the confession of a “regulatory capture strategy.” Behind the dispute stands a conflict of material interests between the factions whose wealth is bound up with the valuations of the closed-model incumbents and those who stand to profit from cheap, abundant intelligence whatever its origin.
Confronted with this crisis, the American ruling class responds as it has to every challenge to its dominance, with economic nationalism and preparations for war. But an open-weight model crosses every border on earth in the time it takes to transfer a file, and no tariff can be levied on the propagation of knowledge.
Precisely because Chinese technology cannot be defeated in open competition, the campaign against it is inseparable from the export controls, the trade war and the military buildup in the Pacific. Down this road lies the breakup of the world’s scientific culture into hostile blocs and, at its end, war between nuclear-armed powers.
Mr. Ball, in his panic over “AI communism,” has inadvertently revealed a basic truth. The claim of a handful of billionaire oligarchs to own this technology, ration its benefits for profit, and weaponize it for war has no historical legitimacy. What Ball calls a “dystopian hellscape” is the alternative now posed: the transformation of AI into public infrastructure for the benefit of all.
As public infrastructure, AI would facilitate the rational and democratic planning of production on a global scale, directed to human need rather than private profit. The same systems now used for surveillance and war could be turned to the conquest of disease, the transition beyond fossil fuels, and the freeing of billions of people from deadening toil.
The Chinese capitalist state, which pursues its own national agenda, will not deliver that future. Only the international working class can. The International Committee of the Fourth International has already demonstrated the revolutionary application of this technology with Socialism AI, launched by the World Socialist Web Site on December 12, 2025, to turn AI toward the socialist education of workers and youth and place the accumulated lessons of more than 150 years of the Marxist movement at their disposal.
The significance of the past week lies not in the triumph of one national capitalism over another. Productive forces created by globally integrated labor have outgrown private ownership and the nation-state system alike, and the convulsions on the American exchanges register that contradiction in the language of the market. It will be resolved in one of two ways: by the capitalist classes, through economic collapse, dictatorship and war, or by the international working class, through the socialist reorganization of economic life on a world scale. A specter is haunting the AI oligarchs—the specter of “AI communism.” Their terror is well founded.