China's proposed artificial intelligence export curbs are designed to draw a line around the country's most valuable AI technology: models, data and technical expertise stay at home, while foreign customers get access through websites and cloud platforms.
That, at least, is the business model Beijing's leadership prefers — and according to a Reuters analysis published Wednesday, it is also the one the market was already converging on. "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. For more context on this story, see our ongoing breaking AI news.
The draft rules, released on October 16, would add 23 AI-related technologies to China's list of technologies banned or restricted from export. The list covers 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 with the government took effect immediately when the draft was published.
Three Tests Decide Which Models Are Covered
The draft specifies three tests for determining which AI models fall under the restrictions, and any one of them is enough:
1. Training compute. Models whose training required 10 to the power of 26 operations — a compute threshold that captures today's frontier systems.
2. Model size. Models with at least 2 trillion parameters.
3. Rankings. Models placing among the world's top 10 in certain assessments of performance, use or efficiency.
The ranking test is the most contentious. It could sweep in smaller systems that excel at specialized tasks, yet the draft does not identify which rankings would count. Liu Jing, a senior partner at Beijing Dacheng Law Offices who participated in consultations with policymakers, told Reuters that companies want more clarity on those tests.
Law firm Han Kun, in its own analysis, questioned how regulators would count the size of mixture-of-experts models, which activate only some of their components for each task. The firm also flagged that one proposed chip-design provision could reach beyond the most advanced products because it lacks a minimum performance requirement.
Services Abroad, Technology at Home
Under the new rules, overseas users would broadly be allowed to keep accessing Chinese AI through websites and cloud platforms. But transfers of the underlying technology — model weights, training data, know-how — or deployments of models on foreign infrastructure would be subject to government approval.
That distinction matters enormously for Chinese developers that rely on computing facilities abroad, where they can access the powerful chips that US export restrictions bar from entering China. Nvidia's most advanced processors cannot legally ship to the Chinese mainland, so some Chinese AI labs rent capacity at overseas data centers instead.
Researchers at major Chinese AI labs are now seeking clarification on whether they can continue to train or run their most advanced models on foreign-owned overseas servers, according to a person briefed on the discussions cited by Reuters. In a worst-case scenario, developers might have to maintain separate domestic and overseas versions of the same downloadable model.
DeepSeek, Moonshot and Z.AI Business Models in Focus
The rules could reshape how China's leading AI firms operate internationally. DeepSeek, Moonshot AI and Z.AI all offer models to overseas customers through websites or cloud platforms — a distribution pattern the curbs explicitly permit.
But the controls extend well beyond model releases. According to an analysis by law firm Fangda, the registration requirements that took effect on October 16 cover transfers that put Chinese technology under foreign ownership or control, including qualifying transactions already underway. Beijing's earlier move to order Meta to unwind its more than $2 billion acquisition of Chinese-founded AI startup Manus showed how much risk foreign buyers face.
For computing centers, the draft provides simplified approvals for Chinese companies building and operating their own facilities abroad — but offers no express pathway for transferring technology to foreign customers who want to build equivalent systems. The controls also cover methods for coordinating more than 10,000 AI chips, a threshold that captures the large-scale training clusters behind frontier models.
A Mirror of US Controls
Chinese lawyers describe the emerging regime as a deliberate 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," wrote Yang Jingru, a partner at law firm Jincheng, Tongda & Neal.
Security considerations accompany the commercial logic. Montufar-Helu noted that Beijing wants to keep powerful cyberattack capabilities and sensitive data inside China, alongside its aim of preserving valuable technological advantages.
The proposals are open for public comment until November 15. China is expected to adopt the rules by the end of the year and issue implementation guidelines afterward, according to a source familiar with the discussions cited by Reuters. China's commerce ministry did not respond to a request for comment.
When the rules take effect, the practical dividing line will be simple: foreign customers can use Chinese AI, but they will not get to own it.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
