China released draft rules on October 16 that would add 23 artificial intelligence technologies to a list banning or restricting their export, including advanced models, autonomous agents, and the technology used to design chips and run large computing facilities, Reuters reported. The draft distinguishes between foreign customers who merely use Chinese AI services and foreign firms that obtain the models, data or expertise to run or reproduce them, allowing the first while subjecting the second to government approval. A temporary requirement to register certain transfers took effect immediately, and the proposals are open for public comment until November 15.
The design encodes a business model Beijing and its biggest AI firms already favour: keep the valuable intellectual property onshore and sell access abroad. "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 of the consultancy Ankura. Companies such as DeepSeek, Moonshot and Z.AI already serve overseas customers through websites and cloud platforms, a route the rules would preserve.
Which models get caught turns on three tests, and lawyers say the thresholds are unclear. The draft covers a model trained with at least 10 to the power of 26 calculations, or with at least 2 trillion parameters, or ranking among the world's top 10 on certain measures of performance, use or efficiency, without naming which rankings count. The law firm Han Kun questioned how regulators would size models that activate only part of themselves per task, and said one chip-design provision lacks any minimum performance bar, so it could reach well beyond the most advanced products. Chinese AI labs are seeking clarification on whether they can still train or run their best models on foreign-owned servers abroad, a route some use to reach Nvidia chips that US rules bar from China.
The symmetry with Washington is the point. "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 of the law firm Jincheng Tongda & Neal. The registration rules also reach corporate deals that would put Chinese technology under foreign control, extending a stance Beijing signalled earlier this year when it ordered Meta to unwind its more than $2 billion purchase of the Chinese-founded startup Manus.
The timing is awkward for China's own market. The draft lands amid a selloff that has wiped roughly $480 billion from Chinese AI and semiconductor stocks, with the STAR50 index down 40 percent in under four months, driven partly by fears that Washington will ban Chinese equipment from data centres. Han Kun noted the rules offer simplified approvals for Chinese firms building their own computing facilities abroad, but not for handing the technology to foreign customers to build equivalent systems, a gap that could complicate the overseas expansion those firms are counting on.
The move cuts against the running US assumption that export controls flow in one direction. President Donald Trump said on September 13 that "whoever wins AI, wins" and that the US leads China, even as a Foreign Affairs analysis argued the decade-long US effort to slow Chinese innovation is "foundering" because Chinese firms keep finding workarounds. At their September summit, the two leaders agreed to extend a trade truce and set up an AI incident alert system but reached no deal on broader AI safeguards; they are due to meet again in China next month. What the draft does not yet settle is how aggressively Beijing will enforce it, and whether the top-10 ranking test will sweep in the specialised smaller models that were never its obvious target.