American AI giants are facing fresh pressure as China pushes open-weight models that can be downloaded and run on users’ own servers. The debate is exposing a split inside Silicon Valley over whether strict control of model code and weights is helping, or hurting, U.S. competitiveness.
U.S. frontier AI companies such as OpenAI and Anthropic have kept their top models closed, charging high fees and relying on third-party platforms. Critics say that approach limits access for companies and developers, while Chinese labs are releasing open-weight models that can be used for free and customized for local systems.
That shift has narrowed the gap between the two sides. Moonshot’s Kimi K3 ranked fourth in a major intelligence evaluation, alongside Anthropic’s Opus 5 and OpenAI’s GPT-5.6 Sol, challenging the idea that only closed models can deliver top-tier performance. Developers and startups worried about data leaks and cost are increasingly turning to Chinese open models for routine work and tailored applications.
The fight is also dividing U.S. industry leaders. Nvidia, Microsoft and Palantir have warned that a rushed crackdown on open-weight models could weaken American competitiveness and push innovation overseas. They argue that concentrating powerful AI in a handful of closed systems can create single points of failure, rather than guaranteeing safety.
