Overview
- U.S. companies are routing growing shares of AI usage to Chinese open‑weight models because they deliver near‑frontier performance at a fraction of the cost, with platform data showing Chinese models have taken over 30 percent of token use on some services.
- Chinese firms such as Z.ai and DeepSeek released widely adopted, low‑cost models this spring and early summer, prompting rapid uptake by startups and developers looking to cut expensive cloud AI bills.
- Beijing has held talks with Alibaba, ByteDance and Z.ai about restricting foreign access to top models and is considering treating unauthorized transfer of AI tech as a national security offense, though no final policy has been announced.
- The United States has already used export‑style controls on frontier models, with the Commerce Department ordering limits on Anthropic’s Fable and Mythos in June, and companies are responding with corporate bans and tightened vetting of partners.
- Open‑weight release practices make models hard to recall and raise IP and cyber risks, which could push governments toward 'trusted partner' access rules and force U.S. firms to choose between cost savings and supply‑chain and security tradeoffs.