US threatens China AI sanctions as cheap models win 46% of US token traffic

Washington threatened sanctions on Chinese AI labs over alleged model theft, but the harder problem is already visible in startup invoices: Chinese models now handle 46% of US token traffic on OpenRouter, drawn by prices 97% below US rivals.

"Large-scale, covert industrial distillation aimed at stealing proprietary US technology is unacceptable," Michael Kratsios, director of the White House Office of Science and Technology Policy, said on X, accusing Moonshot AI of distilling Anthropic's Fable model for its Kimi K3 system. Treasury Secretary Scott Bessent told Fox Business the administration had found watermarks of US large language models on Chinese systems and could deploy sanctions or Entity List designations.

China's Ministry of Commerce fired back the same day, calling the accusations baseless double standards and typical AI hegemony. The exchange comes as US-China AI dialogue planned for September faces potential collapse, according to Paul Triolo, a partner at DGA-Albright Stonebridge Group. Anthropic earlier this year said DeepSeek, Moonshot and MiniMax created about 24,000 fraudulent accounts generating more than 16 million exchanges with Claude in what it called industrial-scale distillation attacks.

The stakes extend beyond diplomacy. DeepSeek V4 Flash lists at $0.14 per million tokens versus $5 for GPT-5.5, a 97% discount that has driven Lindy, an AI agent startup, to move all traffic from Anthropic to DeepSeek and cut inference costs by roughly 90%. DoorDash is pushing lower-level tasks to Moonshot's Kimi model, citing better quality at lower cost, Fortune reported. If Washington restricts access, companies that have quietly routed workloads through Chinese APIs face sudden disruption.

Price Did What Politics Couldn't

KuCoin, citing OpenRouter data, said the share of tokens used by US companies for Chinese models rose from under 5% at the start of 2025 to 46% in April 2026, with DeepSeek alone accounting for 17.6% of token traffic on the platform. Kimi K3's output pricing has been reported around $15 per million tokens, below Anthropic's higher-end Fable pricing in several industry comparisons. The last time a comparable price gap emerged in enterprise software — between on-premise and cloud infrastructure — it took less than three years for the cheaper option to capture a third of the market, according to Gartner data.

The Enforcement Gap

Entity List designations would still hurt. US companies would have to stop doing business with listed firms unless they received licenses, and cloud providers would face tighter compliance pressure. But open-weight models present a structural challenge: once weights are released, they can be downloaded, mirrored and run on infrastructure far from the company that trained them. Nvidia Chief Executive Officer Jensen Huang has supported open-weight Chinese models, arguing that learning from AI is part of how the field advances, Barron's reported. His commercial interest is obvious — more model use means more demand for Nvidia chips — but his technical point has force. The US-China AI talks expected in September may slow the first sanctions move, especially with Bessent set to lead the US side. The practical question for startups is already on the table: which AI workloads can survive a forced provider change, and which are now tied to a model Washington may decide you cannot use?

This article is for informational purposes only and does not constitute investment advice.