AINEWS
Treasury wants sanctions on Chinese AI. Commerce is getting letters from startups. The president has to choose between OpenAI and the rest of Silicon Valley.
The Trump administration has a decision to make in the coming weeks, and it is being pulled apart by two of the most powerful US AI companies on one side and the rest of Silicon Valley on the other. The fight is over whether to keep China's free, open-source models flowing to American businesses.
The trigger is a model called Kimi K3, released in late July by the Chinese lab Moonshot AI. It is a large language model in the same family as OpenAI's GPT or Anthropic's Claude: a system that reads and writes text, answers questions, and writes code. Moonshot published it as an open-source download that anyone can run without paying Moonshot a licensing fee. Independent benchmarks put its performance on the tests that matter for production use within range of the leading US proprietary systems. The price on third-party AI routing services is a small fraction of what OpenAI or Anthropic charge per million tokens, the basic unit of text an AI model processes. For one wing of Silicon Valley, that is a market fact. For the other wing, it is a national-security problem.
The market-fact camp made its case publicly this week. Microsoft, Nvidia, Palantir, and Meta sent a joint letter to lawmakers urging them not to restrict access to open models. Nvidia CEO Jensen Huang carried the same argument into both parties' leadership offices on Capitol Hill on Tuesday, telling lawmakers that American businesses have already wired these models into their products and that cutting them off would punish US companies, not Chinese ones. Commerce Secretary Howard Lutnick has been receiving the same message from AI-startup founders who depend on cheap open models to compete with the well-funded frontier labs.
The national-security camp is smaller but closer to the levers of economic statecraft. OpenAI and Anthropic, the two US labs whose revenue comes from selling access to closed, proprietary models, argue that Chinese open-weight systems were trained on stolen American intellectual property and that letting them propagate through US infrastructure gives Beijing leverage in sensitive systems. Treasury Secretary Scott Bessent has publicly suggested the US could sanction Chinese AI firms on those grounds, treating the IP allegation as a basis for policy action rather than waiting for a court to adjudicate it.
The Trump administration is split in a way that maps almost exactly onto that industry split. The hawks sit at Treasury. The doves sit at Commerce, where startup founders have been writing directly to Lutnick. Between them is a White House that has spent the year oscillating between a hawkish posture on Chinese technology and a reluctance to disrupt the AI tools American businesses have come to rely on. The lobbying is live this week and the decision is near-term.
Microsoft, Nvidia, Meta, and Palantir have signed the same letter, even though they compete with each other on almost everything else. The reason is where each company makes its money. US frontier-model economics depend on customers paying premium prices for closed systems. Cheap, openly downloadable Chinese models put a floor under what US companies can charge for a unit of intelligence. The chipmakers and cloud companies make their money selling the picks and shovels regardless of which model wins, so an open-source outcome benefits them. The closed-model labs make their money selling the model itself, so an open-source outcome threatens them. The interest-group math is clean.
Silicon Valley has historically used Chinese competition as the political cover for resisting any US regulation that might slow its own labs down. That worked when the threat was abstract, a future Sputnik moment or a distant chip race. It stops working when the threat is a free download that any US startup can pull tonight, fine-tune by Thursday, and put into production by Friday. The open-versus-closed AI fight, long a philosophy debate among researchers and open-source advocates, has been forced into a concrete policy question with a deadline.
The decision in front of the White House isn't whether Chinese AI will matter to the US economy. It already does, and the models are already running inside American products. The decision is whether to treat cheap open models as a competitive pressure that US industry should absorb, or as a national-security exposure that US policy should choke off before it spreads further. Treasury and Commerce are about to file their answers, and the president will have to choose which side of Silicon Valley he is on.