China's AI open-source models are gaining global influence; the United States should respond with an open strategy rather than bans.

Foreign media: China's AI is winning the global developer ecosystem thanks to open-source weight models, but may soon tighten overseas access. Instead of responding with bans, the U.S. should do the opposite: government-funded training and permanent public release of advanced model weights, continuously iterated at a reliable pace, enabling global developers to build applications on American models.

Currently, Alibaba's Tongyi Qianwen is the most downloaded AI model worldwide, spawning over 100,000 specialized derivatives; Moonshot's Kimi K3 claims to be the world's largest open-source model, with performance comparable to top products from OpenAI and Anthropic; on the OpenRouter platform, the share of Chinese open-source models handling workloads has risen from around 1% in 2024 to over 40% by early 2025. U.S. companies such as Airbnb, Cursor, and DoorDash have deeply relied on Chinese models.

Switching AI models is like dismantling a foundation—if China cuts off supply and the U.S. lacks alternatives, global developers will be trapped with outdated versions, allowing China to control the lifelines of other nations’ systems. The U.S. should counter by packaging and exporting models, chips, computing power, and financing through export credit agencies, offering a comprehensive solution with higher security standards, rather than resorting to simple bans.

Original article: toutiao.com/article/1871522499159104/

Disclaimer: This article represents the personal views of its author(s)