Beijing is preparing to tighten controls on its most advanced open-source AI models, a policy reversal that could disrupt the cost-cutting strategies of hundreds of US companies that have adopted Chinese alternatives from DeepSeek and Moonshot AI.
Back
China weighs AI model restrictions as 23% of US traffic shifts to DeepSeek
Beijing is preparing to tighten controls on its most advanced open-source AI models, a policy reversal that could disrupt the cost-cutting strategies of hundreds of US companies that have adopted Chinese alternatives from DeepSeek and Moonshot AI.
Chinese officials have held discussions with domestic AI labs about imposing stricter regulatory reviews, public release deferrals and foreign investment restrictions on models including Alibaba's Qwen and Zhipu's GLM-5.2, according to people familiar with the matter. The measures would mark a sharp departure from Beijing's previous strategy of promoting open-source AI as a form of soft power.
"Beijing is concerned that sharing proprietary AI technology could help adversaries and ultimately get weaponized against China," said a person familiar with the discussions. The talks remain in early stages, the people said, and no final decisions have been made.
The shift comes as Chinese AI models have gained significant traction in Silicon Valley. DeepSeek's share of AI usage on the Vercel platform rose to 23 percent in June from 1 percent in April, while its share of AI spending stayed in the low single digits, the startup said. DoorDash co-founder Andy Fang said the company routes complex tasks to Anthropic's cutting-edge model Fable while delegating lower-level workloads to Kimi K2.6 from Moonshot AI — an approach that delivered better performance at lower cost than using two Anthropic models. Legal-services platform Harvey also uses DeepSeek and Zhipu for simpler work, according to Niko Grupen, head of applied research.
The proposed restrictions include a more stringent vetting process before labs can release models, with officials considering requiring labs to defer public releases and restrict access by foreign entities if a review determines their products contain sensitive technology. China is also examining tighter limits on exporting AI technologies and on Chinese AI companies accepting foreign investment, the people said.
A reversal of the open-source playbook
Until recently, Beijing encouraged the rapid global spread of Chinese AI models, viewing it as a form of technological influence. Many leading Chinese models are open-source, allowing developers worldwide to download and modify them at no cost. But the calculus has shifted as US-China technology rivalry intensifies.
Washington's own moves have reinforced Beijing's thinking. The White House previously banned foreign access to Anthropic's Mythos model — capable of detecting cybersecurity flaws automatically — before later allowing access for some users. To Chinese regulators, the back-and-forth underscored the need for governments to keep a tight grip on powerful AI technology to prevent misuse in areas such as cyberwarfare and bioweapons development.
Chinese companies are also accelerating efforts to reduce dependency on foreign hardware. DeepSeek is developing an inference chip designed to reduce reliance on Nvidia and Huawei, coinciding with its first external funding round. The initiative aligns with China's broader push for self-reliance in critical technologies after US export controls restricted access to Nvidia's high-performance GPUs.
What's at stake for global AI
Any move by China to restrict access to its open-weight models could disrupt supply chains for US companies that have come to rely on cheaper Chinese alternatives. Industry participants said tighter controls risk alienating foreign users and slowing global adoption of Chinese AI technology.
The potential restrictions could also intensify competition among AI providers. US companies that lose access to Chinese models may turn to alternatives from OpenAI, Anthropic and Google — potentially driving up costs. Smaller organizations with limited resources may find it particularly challenging to adapt.
China's top AI models still trail the best the US can offer, according to industry officials on both sides of the Pacific. But Chinese models have gained traction by offering good-enough capabilities at a fraction of the cost — a value proposition that could vanish if Beijing tightens the tap.
This article is for informational purposes only and does not constitute investment advice.