China is preparing export restrictions on artificial intelligence technology, according to a report by The Straits Times, in a move designed to give Beijing greater control over a sector both Washington and its allies have spent three years trying to fence off.

The report describes a plan still in draft form. It says the curbs would cover AI-related technology and would serve two purposes: to assert regulatory control over what Chinese firms sell abroad, and to give the state leverage in a technology race that now defines trade policy on three continents. No implementation date, product list or licensing framework appears in the report.

What is Beijing actually proposing?

The plan would subject AI technology exports to government approval. That much is confirmed by the report. Everything else remains open. The scope of "AI technology" is undefined in the draft stage, and past Chinese control lists show that definitions can be narrow or sweeping depending on the ministry that writes them.

Three questions will decide the commercial impact:

• Which product categories fall under the rules: models, chips, training tools or full systems
• Whether the controls apply only to Chinese-origin technology or also to re-exported foreign components
• How quickly the Ministry of Commerce issues or denies licences once the rules take force

Chinese export control law already gives the state broad authority. Beijing added dual-use items to its control list in December 2020 and imposed export licensing on gallium and germanium in August 2023, then on graphite that December. An AI-specific regime would extend that toolkit to the most contested layer of the stack.

Why does this matter to the chip supply chain?

The move mirrors the structure of US policy in reverse. Washington has restricted advanced AI chips and semiconductor equipment since October 2022, and it tightened those rules in 2023 and 2024 to cut off workarounds. Beijing's draft suggests it intends to control the technologies where it holds strength, rather than only defend against restrictions imposed on it.

China dominates several inputs the AI hardware chain cannot replace quickly. It refines the majority of the world's gallium and germanium, metals used in radio-frequency chips and optical components. It also processes most global rare earth output, and its AI models, led by firms such as Baidu, Alibaba and DeepSeek, have closed part of the performance gap with US rivals at lower cost.

Export controls on AI technology would therefore give Beijing a bargaining instrument. The Straits Times report frames the plan explicitly as an attempt to "exert control over key tech", language that signals intent rather than a routine regulatory housekeeping exercise.

How do curbs on AI differ from curbs on metals?

The gallium and germanium cases offer the closest precedent, and the comparison is instructive. Metals are physical goods that cross borders in measurable volumes, so licensing slows trade in visible ways. AI technology is partly software, weights and know-how that move through networks, not ports.

That difference cuts both ways. Software controls are harder to enforce, because models can be copied and transferred outside any customs channel. But software controls are also harder to evade for a buyer who needs updates, support and hardware integration, which gives Chinese vendors a point of leverage that a metal shipment cannot match.

The draft's outcome may therefore depend less on enforcement at the border than on licensing decisions made case by case in Beijing, the same mechanism Washington uses through its own entity list.

What does this mean for semiconductor buyers and vendors?

For chipmakers outside China, the plan adds a second regulatory wall. A firm selling AI accelerators into China already navigates US export rules; if Beijing restricts Chinese AI technology flowing outward, customers in third markets may face pressure to choose between the two ecosystems rather than mix them.

For Chinese AI vendors, the curbs carry a cost. Export licensing adds friction for customers in Southeast Asia, the Middle East and Latin America, regions where Chinese models and inference hardware have won business on price. The report does not say whether Beijing plans exemptions for those markets, and that gap leaves the commercial picture incomplete.

For equipment and materials suppliers, the signal is indirect but real. Every new control layer pushes the two blocs toward separate stacks, from training silicon to packaging to model distribution, and firms with footprints in both camps will need compliance architecture on both sides.

What comes next?

The report offers no timeline, and Chinese regulatory drafts have historically moved at uneven speed: the rare earth regime took months from announcement to enforcement, while other proposals stalled. Companies exposed to Chinese AI exports should watch the Ministry of Commerce's formal consultation notice, because the final scope of the product list, more than the announcement itself, will set the price the market pays.