China Imposes Export Controls on Advanced AI: The New Frontier in the Tech Cold War
On October 16, 2026, China's Ministry of Commerce (MOFCOM) and Ministry of Science and Technology released Announcement No. 48 of 2026, proposing the addition of 23 artificial intelligence technologies to its Catalogue of Technologies Prohibited or Restricted from Export. The draft proposal, open for public comment until November 15, 2026, introduces the most comprehensive technology export control regime China has ever implemented for the AI sector — with far-reaching implications for global AI development, the Sino-US tech rivalry, and the future of international scientific cooperation.
What was proposed
The controls target three broad categories: large-scale AI models, autonomous agents with planning and coordination capabilities, and AI chip design and compute cluster technologies. For large models, the restriction applies to any version meeting at least one of three conditions: trained with at least 10²⁶ floating-point operations, carrying at least 2 trillion parameters, or ranked in the global top 10 on public performance benchmarks. The restriction covers model weights, key checkpoints, source code architecture, and training and inference optimization know-how.
AI agents are treated independently from the underlying models. Technologies for planning, multi-agent coordination, tool use, and game-theoretic decision-making face restrictions above defined benchmarks. High-performance agents designed primarily for use in national critical infrastructure, defense, or core government operations face outright prohibition of export.
The announcement also includes 14 prohibited technologies — including large-scale models trained on China's "core data" and AI-driven malicious code generation tools — and 26 restricted technologies, each requiring a license for export.
The service vs. transfer distinction
A central feature of the proposal is the distinction between accessing a model as a service and transferring the technology itself. MOFCOM explicitly states that API calls, web applications, SaaS platforms, and lightweight on-device deployments do not count as exporting an AI model. This means companies like DeepSeek and Moonshot, which offer their models to international customers through cloud platforms, can continue operating without a license.
By contrast, local deployment of restricted model weights on infrastructure outside China counts as a technology transfer and requires the Chinese supplier to hold an export license. Companies already using Chinese models on their own overseas servers must assess whether they need to register their operations — even if the technology was acquired legally and has been in use for months.
The distinction is crucial for the business model China has cultivated: keeping proprietary knowledge and research and development within the country while selling access to those models 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," Alfredo Montufar-Helu, managing director at consultancy Ankura in Beijing, told Reuters.
Impact on the global ecosystem
The proposal directly affects companies and researchers worldwide. According to legal analysis by Fangda Partners, the registration requirements cover transfers that put Chinese technology under foreign ownership or control — including transactions already underway. This means foreign acquisitions of Chinese companies, such as Meta's acquisition of the Chinese-founded AI startup Manus, already show the risks facing international buyers.
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. The announcement also includes provisions requiring registration for intra-group code sharing and open-sourcing of restricted models.
The proposal also covers chip design technologies at process nodes of 14 nm or below and compute clusters with interconnect above 800 GB/s — directly affecting the AI hardware supply chain already under US sanctions.
A mirror of US restrictions
Observers have noted that China's controls function as a mirror of US restrictions on Chinese access to advanced hardware. 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." While the US controls the inflow of compute power, hardware, and capital, China begins to systematically manage the outflow of models and technology assets.
China is expected to adopt the proposals by year-end and issue implementation guidelines later, according to a source familiar with the matter who requested anonymity. The gap between public consultation and final implementation can be significant — companies are already operating under the interim registration requirement, which took effect immediately upon the announcement.
Meanwhile, China's AI and semiconductor markets continue to face extreme volatility, with stocks tumbling amid regulatory uncertainty. The proposal positions Beijing to leverage these controls as bargaining chips in future US-China trade negotiations, potentially offering export control concessions in exchange for relief on hardware sanctions.
The question remains whether China's new export control line will accelerate the fragmentation of the global AI ecosystem or serve as a negotiation instrument for a Sino-US agreement that could redefine the rules of the technology game for decades to come.
Sources: The News International, China Briefing, LavX News
✓ Independent sources cross-checked and verified before publishing
