SINGAPORE/BEIJING — China's proposed artificial intelligence export curbs will push a business model favoured by Beijing and the industry's biggest players: keeping valuable technology at home while selling access to it abroad, as rivalry with the United States intensifies, Reuters reported in an analysis published on Wednesday.

Source: Reuters

The draft rules, released on October 16 by the Ministry of Commerce and the Ministry of Science and Technology as Announcement No. 48, distinguish between foreign customers using Chinese AI services and foreign companies obtaining the models, data or expertise needed to operate or reproduce them independently. On October 23 the commerce ministry clarified that overseas API services are exempt — cross-border cloud calls processed inside Chinese data centres face no new limit — while physical transfers of a big model's weights abroad would need a state export licence.

Source: East Asia Brief

"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, speaking to Reuters. The rules would broadly allow overseas users to access Chinese AI through websites and cloud platforms, while subjecting some transfers of underlying technology and overseas deployments to government approval. A temporary requirement to register certain transfers took effect immediately on October 16.

Three tests decide which models are covered

The draft would add 23 AI-related technologies to China's catalogue of technologies banned or restricted from export, among 40 items added across three sectors. For large models, the draft specifies three tests, any one of which brings a model under control: training compute of at least 10 to the power of 26 operations, a size of at least 2 trillion parameters, or a place among the world's top 10 in certain assessments of performance, use or efficiency. The controlled material covers weights, checkpoints, source code, architecture and training methods, and the list also reaches AI agents that carry out tasks with limited human direction, technology used to design chips and run vast computing facilities, and methods for coordinating more than 10,000 AI chips.

Companies are asking what counts

The rules could change how leading Chinese AI firms run advanced models on overseas data centres, structure international partnerships and handle foreign investment. DeepSeek, Moonshot and Z.AI already offer models to overseas customers through websites or cloud platforms. Some Chinese developers have also turned to overseas computing facilities to reach advanced chips that US restrictions keep out of China — Nvidia's most advanced processors cannot be exported to China. Researchers at major Chinese labs are now 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 spoke to Reuters on condition of anonymity.

Liu Jing, a senior partner at Beijing Dacheng Law Offices who took part in consultations with policymakers, told Reuters companies also wanted more clarity on the tests used to decide which models are covered. Law firm Han Kun 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 sets no minimum performance level. The draft does provide simplified approvals for Chinese companies building and operating their own facilities abroad, Han Kun noted, but not expressly for transferring technology to foreign customers to build equivalent systems.

The reach extends to corporate deals. Registration requirements in force since October 16 cover transfers that put Chinese technology under foreign ownership or control, including qualifying transactions already under way, according to an analysis by law firm Fangda cited by Reuters. Beijing's move earlier this year to order Meta to unwind its acquisition of Chinese-founded AI startup Manus — a deal worth more than US$2 billion (yaani, 200 crore dollars se zyada ki deal) — has already shown the risks facing foreign buyers.

Yang Jingru, a partner at law firm Jincheng Tongda & Neal, described the emerging approach as a counterpart to Washington's restrictions: "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." For the Punjabi and Indian tech workers and students who build on Chinese open models and APIs, the immediate position is unchanged — access stays open; it is moving the model itself across a border that now needs Beijing's permission.