From LLMs to compute clusters: inside China’s draft plan to control AI exports 🤖🌏

Cliquez ici pour lire en français

China is taking another step in its push to control artificial intelligence. On October 16, 2026, Beijing unveiled a draft regulation that would tighten export controls on AI technologies it considers strategic. The goal is clear: protect homegrown innovation and consolidate the country’s position in the global AI race. The move comes amid growing technological rivalry, with US restrictions on certain advanced technologies and ever-fiercer competition over large AI models, semiconductors and computing infrastructure.

Beijing puts its most advanced technologies under watch 🔐

The draft, presented jointly by China’s Ministry of Commerce and Ministry of Science and Technology, would add 23 AI technologies to the catalogue of technologies subject to export restrictions or bans. They span several key areas: large language models (LLMs), AI agents that can carry out tasks autonomously, large-scale computing infrastructure, and various information-processing technologies.

In practice, some innovations could no longer be transferred abroad, while others would reportedly require prior approval from Chinese authorities before any sale or licensing to international partners. For Beijing, this goes well beyond trade: these technologies would be treated as strategic assets, on a par with energy, critical infrastructure and sensitive resources.

Ever more powerful AI models are the focus 💻

Large AI models are among the most notable targets. According to the criteria described in the draft, some models could fall under a control regime once they reach a high level of performance or technological complexity. Models built on heavy computing power, advanced architectures or datasets deemed sensitive for national security would be among those covered.

Authorities also want to regulate the components that make those performances possible, including AI chips, advanced memory architectures and certain voice-recognition technologies. It points to a major shift: it is no longer only finished products that count as strategic, but also the knowledge, the training methods and the infrastructure needed to build them.

A clear line between open innovation and sensitive technology 🔍

Despite the tougher stance, Beijing is not closing the door on international technology cooperation. Technologies already made public through open source projects or scientific publications would stay out of scope. So would access to models through APIs (the interfaces that let developers plug a model into an app), mobile apps or SaaS platforms (software delivered online, usually by subscription), which would not count as a technology export.

That distinction matters most for foreign companies that rely daily on Chinese models hosted in the cloud. As long as the technology itself is not transferred or installed on third-party infrastructure abroad, it remains outside the core of the restrictions. The aim appears to be less about blocking the spread of AI than about controlling the flow of the most sensitive technology assets.

A response to the new global AI race 🌍

The announcement lands at a time when AI has become a question of technological sovereignty for major powers. For months, Washington and Beijing have been rolling out measures to protect their respective competitive advantages in chips, computing infrastructure and advanced models. In that context, China wants to prevent technologies developed on its soil from being used freely by foreign players without oversight.

Chinese AI heavyweights, whose models keep growing in power and scale, are now at the centre of that protection strategy.

What it means for businesses and developers 🔬

In the short term, international users should feel little impact. Cloud services, APIs and much of the already-published models remain available. But the change could shift the rules for companies that depend heavily on Chinese open-weight models (models whose parameters are published) or want to run these technologies directly on their own infrastructure.

More broadly, the move could speed up the fragmentation of the global AI ecosystem, with increasingly self-contained technology blocs emerging around the United States, China and Europe.

A new era of digital sovereignty 🌐

Beyond the regulatory detail, the Chinese draft reveals a deeper trend: AI is no longer just a technology tool but a lever of economic, industrial and geopolitical power. By seeking to control exports of its most advanced innovations, Beijing is sending the world a clear message. In the battle for AI mastery, data, algorithms and computing power have become strategic resources as valuable as the raw materials of the past. China’s next decisions could durably redraw the balance of the global digital economy.

Is export control a legitimate way to protect national innovation, or a brake on global cooperation in AI? Tell us in the comments.