Washington Debates a Chinese AI Ban. The Market Already Signed the Contracts.

Alibaba did something last week that looked like a pricing decision. It shifted Qwen3.8-Max, its newest AI model, from free open-weight distribution to a revenue-sharing model: large commercial users now pay through Alibaba's cloud platform rather than downloading the model and running it themselves ^[1]. Moonshot AI had already set the template with its Kimi K3, which charges up to 30% of revenue from users earning over $20 million annually ^[1]. DigitalOcean, the American cloud provider, signed on. The move reads as monetization — Chinese labs spent billions training these models and now want a return. But it also changes something more structural. A free download is a file on a server. A government can criminalize its distribution going forward, and the legal question reduces to whether anyone still possesses a copy. A revenue-sharing contract is a commercial relationship. To ban it, a government would need to unwind signed agreements, terminate ongoing payment obligations, and order American companies to stop paying foreign counterparties with whom they hold valid contracts. The First Amendment obstacles that already complicate a download ban — the Supreme Court's Lamont v. Postmaster General precedent protects the right to receive foreign materials — multiply when the government must reach into the terms of private commercial agreements ^[2]. This is the product layer the Trump administration is now debating whether to ban. The debate was triggered by Moonshot AI's release of Kimi K3 on July 16, a 2.8-trillion-parameter open-weight model that set off a global tech sell-off — the Philadelphia Semiconductor Index dropped 12.5% — and forced Washington to confront how thoroughly Chinese models had penetrated the American AI supply chain ^[3][4]. The administration is split. National security officials want Entity List designations and procurement bans. The White House's own AI adviser, David Sacks, has called the effort something sharper than misguided.

the leading closed labs, already a duopoly in terms of AI model revenue, want the government to eliminate their open-source competition. — David Sacks

Five days before the ban debate intensified, the administration had already exempted open-weight models from government safety testing — including Chinese models like Kimi K3. National Cyber Director Sean Cairncross made the administration's reasoning plain.

A regulatory regime would not only strangle growth, development and innovation, and be enormously harmful to the industry, but it would be obsolete 48 hours after it was going through whatever process it had gone through. — Sean Cairncross

The exemption was pushed by a coalition of 25 companies led by Jensen Huang, whose Nvidia chips run the models in question, and included Microsoft and Palantir ^[5]. The administration opened the door on August 4, then convened a meeting to discuss whether to close it. American enterprises, meanwhile, have been walking through it. Pinterest integrated DeepSeek R-1 into its recommendation engines and reported that models trained on Chinese open-source techniques were 30% more accurate than leading proprietary alternatives.

They chose to open source it, and that sparked a wave of open source models. — Bill Ready

Airbnb deployed Alibaba's Qwen for customer service. The US-China Economic and Security Review Commission found that approximately 80% of US AI startups now use Chinese open-source base models.

Open model proliferation creates alternative pathways to AI leadership. — U.S.-China Economic and Security Review Commission

The scale extends beyond startups. Jensen Huang's coalition cited data showing Chinese models now account for over 70% of tokens served on inference platforms — up from less than 3% fourteen months earlier ^[5]. All six of the world's most-used AI models are Chinese. These are not downloads sitting on a server that a government can criminalize. They are integrations — into recommendation engines, customer service pipelines, inference platforms — and they are increasingly governed by contracts, not open-weight licenses. Alibaba's revenue-sharing pivot and Moonshot's 30% model are not anomalies. They are the direction of travel, and American firms are signing on before Washington has finished drafting the ban. The administration's internal debate will continue. In the time it takes to resolve, Alibaba and Moonshot will sign more American customers to revenue-sharing agreements — each one converting what could have been a criminal prohibition into a commercial relationship that would require unwinding. Washington is trying to criminalize a download. American enterprises are signing leases.