Washington Eyes Chinese AI Models as U.S. Usage Nears 60%

The Trump administration is examining whether the rapid adoption of Chinese artificial intelligence models could create new intellectual-property and national-security risks for the United States. U.S. Trade Representative Jamieson Greer said Washington is closely reviewing how China is expanding its AI development, while Treasury Secretary Scott Bessent said foreign models found to be using stolen American intellectual property could face sanctions. Bessent also suggested that pressure could be placed on companies using Chinese AI and raised the possibility that businesses may eventually need to disclose that use to customers. However, Greer indicated that the administration has not yet reached a conclusion on what standards or restrictions it may pursue.

Any policy response could prove more difficult than previous measures targeting Chinese electric vehicles or Huawei Technologies, a Chinese telecommunications equipment company, because open-weight software can be downloaded, modified, and operated locally after release. Chinese models already represent nearly 60% of token usage by U.S. companies on OpenRouter, an artificial intelligence model marketplace, suggesting that the technology has become increasingly embedded in American business activity. DoorDash NASDAQ:DASH, a U.S. delivery platform, and Airbnb NASDAQ:ABNB, an online accommodation marketplace, have used Chinese models as lower-cost alternatives or complements to products from OpenAI, an American artificial intelligence developer, and Anthropic, an artificial intelligence company developing large language models. Restrictions could also affect Baseten and Fireworks, artificial intelligence infrastructure companies whose businesses depend on serving open-weight models, including a growing number developed in China.

Investors may view the debate as a potential source of regulatory uncertainty for artificial intelligence developers, infrastructure providers, and U.S. companies that have already integrated Chinese models into their operations. Some policy specialists believe Washington may not need to impose a complete ban, because higher legal, compliance, or reputational risks alone could encourage regulated businesses to reconsider using Chinese software. New restrictions could also complicate planned U.S.-China artificial intelligence discussions before Chinese President Xi Jinping's expected September 24 visit, after Xi warned against extending national-security restrictions too broadly in the AI sector. The situation may also matter for NVIDIA NASDAQ:NVDA, a U.S. chip designer whose processors have been part of Washington's strategy to keep Chinese developers dependent on American technology, as inexpensive Chinese models increasingly create their own commercial dependence among U.S. startups, researchers, and infrastructure providers.