Analysis
China's Ministry of Commerce and Ministry of Science and Technology proposed on Oct. 16 adding 40 technologies to the country's export-control catalogue, and 23 of them are AI-related -- the first time Beijing has moved to explicitly restrict exports of model weights and source code, not just chips. Crypto Briefing and the South China Morning Post report the full catalogue splits into 14 technologies marked outright prohibited and 26 marked restricted, spanning AI, biotechnology and solar power alongside large language models, autonomous agents, chip designs and computing clusters.
What the thresholds actually cover
The proposal would require a license for any model whose underlying technology surpasses specific performance thresholds -- covering models with at least two trillion parameters, or any model ranking in the global top 10 by performance, usage or safety benchmarks. The draft carves out exemptions that matter for US tech companies operating in China: access delivered purely through APIs, websites, apps or software-as-a-service platforms is excluded, and technology already public through open-source releases or academic papers is untouched. A public comment period runs through Nov. 15, and Beijing has already imposed an interim pre-export registration requirement on the affected technologies while the rules are finalized.
“15, and Beijing has already imposed an interim pre-export registration requirement on the affected technologies while the rules are finalized.”
Commerce Ministry officials cited national-security rationale, specifically the risk that frontier AI systems could be misused to attack critical infrastructure, defense industries or government systems. Officials have reportedly been consulting Alibaba, ByteDance, Zhipu and other domestic AI and chip developers on the shape of the rules since summer -- suggesting Beijing is trying to calibrate the controls so domestic champions aren't caught by their own government's restrictions.
Why this lands differently than prior chip controls
Washington's export-control fight with Beijing has mostly run in one direction: the US restricting Nvidia and AMD chip sales into China. This is China doing the mirror move on its own frontier AI output. For the US chip and AI-infra startups Pulse has tracked this month, including Mueon, which just raised $65 million to build data-center chips domestically, the practical effect may be limited: the exemption for API-delivered access means most US enterprise customers using Chinese models through a hosted API face no new friction, since the rule targets exports from China rather than into it.
What the headline misses: this is a draft open for comment until Nov. 15, not a final rule, and the thresholds -- particularly the two-trillion-parameter cutoff -- could move before anything takes effect. China has also shown before that it enforces export rules unevenly depending on the political moment, so a proposal with teeth on paper doesn't guarantee consistent enforcement once trade talks with Washington shift.
What to watch: whether the US Commerce Department treats this as a reciprocal, good-faith control worth matching with any easing of its own chip restrictions, or reads the carve-outs for domestic giants like Alibaba and ByteDance as evidence the rule is really about managing who inside China gets access to frontier capability.