China Plans to Tighten Export Controls on AI Models and Chips
Miles Bennett
China's Commerce Ministry is consulting top tech firms on restricting AI model-weight downloads and chip-design exports — if enacted, the changes would mark the most significant overhaul of China's export-control technology catalogue in years, directly constraining how Chinese open-weight models spread globally.
What exactly is being restricted on the AI side?
The Commerce Ministry is consulting Alibaba, ByteDance, and Zhipu on two potential curbs: blocking the transfer of key training data abroad, and restricting foreign users from downloading model weights — the core parameter files that effectively *are* the model.
This means → overseas users could still access Chinese AI services online, but pulling the model onto their own servers may no longer be an option.
In plain terms = you can *use* it, but you can't *take it home* — China is trying to turn open-weight models from "downloadable" into "remote-access only."
Why does this matter for the global AI landscape?
Chinese star models such as DeepSeek and Moonshot's Kimi K3 use open weights, letting users download and customize locally; Anthropic's and OpenAI's flagship models are closed-source.
This means → open weights are the core differentiator driving Chinese AI adoption abroad. If these curbs land, that distribution channel narrows sharply.
This reflects a deliberate recalibration: China is redrawing the line between "openness for influence" and "preventing capability leakage."
What is changing on the chip side?
The ministry is also discussing a ban on foreign chip firms — Qualcomm and TSMC among them — manufacturing advanced chips based on designs developed by Huawei, Alibaba, or ByteDance.
In plain terms = a Chinese company can draw the chip blueprint, but can no longer hand it to an overseas foundry to build — the target is design capability leaking through contract manufacturing.
This means → the room for Chinese chip designers to partner with offshore foundries would be further squeezed.
How is Beijing plugging the acquisition loophole?
Regulators are also considering restrictions on foreign acquisitions of strategically important Chinese tech firms, with a focus on areas such as agentic AI — AI systems capable of autonomously executing complex tasks.
The immediate trigger: Meta's $2 billion acquisition of Manus, which Beijing viewed as exploiting a gap in existing rules. The deal has since been ordered reversed.
This means → China is not only restricting technology from leaving — it is also restricting the companies themselves from being bought.
What do the companies think — and what could change?
Most proposals remain under discussion. Tech firms have told regulators that some tighter controls would slow their own AI R&D and undermine China's competitive edge.
If enacted, the measures would be folded into an update to China's export-control technology catalogue — the most significant revision since lithium-battery manufacturing tech was added in 2025.
This reflects the central tension of this policy fight: tighter controls protect technology security but risk sacrificing the industry's global competitiveness — where the final line falls depends on how regulators weigh one against the other.
Content is for reference only, not financial advice.