Earlier this month I watched a demo of Kimi K3 in Beijing start‑up Moonshot’s Radiohead‑themed conference room. Founder Yang Zhilin named the company after Pink Floyd’s *Dark Side of the Moon*, a nod to uncharted, high‑reward territory. The release of Kimi K3 has been hailed as a chance for China to flex its tech expertise against U.S. rivals.

More U.S. firms trim costs by switching to cheaper, high‑performing Chinese open‑AI models such as Z.AI’s GLM, Deep Seek’s R‑1 and Alibaba’s Qwen 3.8 Max. Flo Crivello, CEO of San Francisco‑based Lindy AI, wrote that the switch “saves us millions of dollars and actually increases performance on many core use cases.”

The security risk hinges less on a model’s origin than on where it is hosted. Open‑source weights can be downloaded, fine‑tuned, and run on domestic or third‑party infrastructure, keeping data out of China. When hosted in the U.S., models can answer questions blocked inside China, and users can alter content restrictions. Even in China, cooperation with the National Intelligence Law is not guaranteed, as disputes with Pinduoduo inspectors show.

The U.S. needs a viable open‑source option of its own. Transparency from open weights can improve security, as Hugging Face recently used a Chinese model to analyse a cyber‑breach without data leaving its system.