The draft rules, released on October 16, distinguish between foreign customers using Chinese AI services and foreign companies obtaining the models, data or expertise needed to operate or reproduce them independently.
“The structure the leadership prefers, with IP and R&D kept onshore and operations and access sold abroad, is the one the market was already converging on,” said Alfredo Montufar-Helu, Beijing-based managing director at consultancy Ankura.
The new rules would broadly allow overseas users to access Chinese AI through websites and cloud platforms while subjecting some transfers of underlying technology or deployments overseas to government approval.
That distinction matters for developers that rely on computing facilities abroad, where they can gain access to powerful chips constrained by US export restrictions on China.
China is expected to adopt the proposals by the end of the year and issue guidelines on implementation later, according to a source familiar with the discussions who requested anonymity because the timetable was not public.
The draft would add 23 AI-related technologies to a list of technologies banned or restricted from export, including advanced models, agents that carry out tasks with limited human direction and technology used to design chips and run vast computing facilities. A temporary requirement to register certain transfers took effect immediately.
Companies seek clarification
The rules could impact how leading Chinese AI firms run advanced models on overseas data centres, structure international partnerships and handle foreign investment or ownership of AI assets.
DeepSeek, Moonshot and Z.AI offer models to overseas customers through websites or cloud platforms. Some Chinese AI developers have also turned to overseas computing facilities to access advanced chips restricted from export to China. Nvidia’s most advanced processors cannot be exported to China under US restrictions.
Researchers at major Chinese AI labs are seeking clarification on whether they can train or run their most advanced models on foreign-owned overseas servers, according to a person briefed on the matter who declined to be identified because the discussions were not public.
In a worst-case scenario, developers might have to maintain separate domestic and overseas versions of the same downloadable model, the person said.
Liu Jing, a senior partner at Beijing Dacheng Law Offices who participated in consultations with policymakers, told Reuters companies also wanted more clarity on the tests used to determine which models would be covered.
The draft specifies three tests: the computing work needed to train a model, with a threshold of 10 to the power of 26 calculations; its size, measured by at least 2 trillion parameters; or placing among the world’s top 10 in certain assessments of performance, use or efficiency.
Parameters are the numerical settings a model learns during training. The ranking test could cover smaller systems that excel at specialised tasks, but the draft does not identify which rankings would count.
Chinese law firm Han Kun, in an analysis, also questioned how regulators would count the size of models that activate only some of their components for each task, and said one proposed chip-design provision could reach beyond the most advanced products because it lacks a minimum performance requirement.
The proposals are open for public comment until November 15. China’s commerce ministry did not respond to a request for comment.
Overseas ownership
The rules could extend beyond model releases to corporate deals. Registration requirements, effective from October 16, cover transfers that put Chinese technology under foreign ownership or control, including qualifying transactions already underway, according to an analysis by law firm Fangda.
Beijing’s move earlier this year to order Meta to unwind its more than $2 billion acquisition of Chinese-founded AI startup Manus has already shown the risks facing foreign buyers.
For computing centres, the draft provides simplified approvals for Chinese companies building and operating their own facilities abroad, but not expressly for transferring technology to foreign customers to build equivalent systems, Han Kun said.
The controls also cover methods for coordinating more than 10,000 AI chips.
Montufar-Helu said security concerns, including keeping powerful cyberattack capabilities and sensitive data in China, accompanied a commercial aim to retain valuable technological advantages.
Yang Jingru, a partner at law firm Jincheng Tongda & Neal, described the emerging approach as a counterpart to US restrictions on China’s access to advanced technology and investment.
“The US controls the inflow of computing power, hardware and capital, while China is beginning to systematically manage the outflow of models and technological assets,” Yang wrote.
