Beijing is considering regulating the export of its own AI models; Qwen, Doubao, and Z.ai are all named.

The Financial Times reported that China’s Ministry of Commerce is considering tightening export regulations for AI models and chip-related products. It has already been in talks with companies including Alibaba (Qwen Tongyi Qianwen), ByteDance (Doubao), and the startup Z.ai (GLM-5.2), with the aim of restricting overseas access to flagship systems.
(Background: Beijing is reportedly considering controlling DeepSeek and the Dark Side of the Moon, because China’s AI models are being widely “picked up for cheap” by Silicon Valley.)
(Additional context: An AI unicorn Manus founder was “restricted from leaving the country” by China! Beijing reviews Meta’s acquisition deal—does a $2 billion transaction have uncertainty?)

Table of contents

Toggle

• Who was named?
• What is China afraid of?
• Who will feel the pain first?

Over the past decade, Beijing has almost treated “export controls” as a synonym for the U.S. cracking down on China’s technology. But now the direction of the wind is unusually reversed. The Financial Times reported that China’s Ministry of Commerce is discussing with Alibaba, ByteDance, and the AI startup Z.ai to consider tightening export regulations for AI models and chip-related products, to prevent the country’s most advanced technology and leading startups from being acquired by or otherwise gaining access to it through the West.

This is not an idea that came out of nowhere. Half a month ago, there were already reports that Beijing was considering controlling DeepSeek and the Dark Side of the Moon, with the rationale that China’s AI models are being aggressively taken as low-cost substitutes by Silicon Valley companies. This time the scale is larger, and the target is the Chinese open-source model itself—the one that is used the most worldwide.

Who was named?

According to reports, the Ministry of Commerce has discussed restricting overseas access to their flagship systems with three companies: Alibaba (flagship model Qwen Tongyi Qianwen), ByteDance (Doubao), and Z.ai (model GLM-5.2).

The measures under consideration include: reviewing export lists covering AI and chip-related products, setting clearer standards for issuing export licenses, strengthening end-user verification—that is, confirming who the buyer is and what they will use it for—and raising the threshold for exporting technical products overseas.

Everything is still not finalized. The Ministry of Commerce has not released any formal proposal, nor has it published a timeline, and the three named companies have not publicly confirmed the rumors.

Some officials privately suggest that the scope of the controls may only fall on “future-generation models,” meaning new versions that have not yet been released. As for older versions that are already widely circulating, they would not be retrospectively handled for now.

What is China afraid of?

The Qwen family has more than 113k derivative models on Hugging Face. If all models tagged as Qwen are included, the number exceeds 200k—more than the combined total of Google and Meta. Qwen’s cumulative downloads are about 942 million times, already surpassing Meta’s Llama (about 476 million), making it the open-source model with the most users globally.

The speed of change is even more striking: in January 2024, the share of newly added fine-tuned models that use Qwen as the foundation was only about 1%. By February 2026, that figure has surged to 69%.

Even Singapore’s national-level AI project, AI Singapore, has chosen Qwen instead of Llama as the regional model foundation. The overall momentum of China’s open-source models (Qwen, DeepSeek, etc.) on the Hugging Face platform has now overtaken U.S. models.

In other words, Beijing is not looking to regulate the output of one or two labs. It is aiming at the most mainstream system set in the global open-source ecosystem. Once technology flows out, it is very difficult to pull it back. This is exactly why the Ministry of Commerce is tense right now.

In the past, “open source” was used by China’s camp as a weapon to counter closed-source U.S. tech giants—proving they were more open and more generous. But now that same weapon is causing Beijing to worry that the most advanced results will ultimately end up in someone else’s hands for practice.

Who will feel the pain first?

Over the past few years, European developers and many small and mid-sized companies have relied on China’s free and open-weight models as a cheap alternative to U.S. commercial services. If Beijing really tightens restrictions on overseas access, this group of dependents would be the first to feel the impact.

Some observers also believe that this whole discussion could just be bargaining chips—placed on the table in trade negotiations with Washington, to secure U.S. concessions on semiconductors or tariffs—rather than something that will necessarily be implemented in practice.

But the irony is that over the past decade, China has consistently heavily criticized export controls, calling them products of hegemony tools and unilateralism—unfair measures used by the U.S. to maintain technological advantages. Now the same script may end up being labeled as “made in Beijing,” and used in the very area China is best at—and also most wants to protect.