China weighs tighter export controls on AI models and chips

China weighs tighter export controls on AI models and chips

Chinese regulators are considering a sweeping set of new export controls on advanced artificial intelligence models, semiconductor technologies, and strategic AI firms, signaling a shift in policy toward treating cutting-edge technology as a strategic national asset. The proposals, led by the Ministry of Commerce, would restrict the overseas transfer of model weights, training data, and chip designs.

According to the Financial Times, the Ministry of Commerce has held consultations with major domestic technology companies, including ByteDance, Alibaba, and Zhipu AI (now Z.ai), to discuss limiting the transfer of key training data and preventing foreign users from downloading model weights. Model weights are the core parameters produced during training that determine how an AI system operates; restricting them would prevent foreign developers from running, modifying, or fine-tuning models independently.

The proposed measures may be integrated into the next revision of China's catalogue of technologies prohibited or restricted from export. Technologies listed as restricted generally would require government approval before transfer, allowing regulators to review transactions on an individual basis.

The restrictions extend beyond software into hardware. Chinese officials are reportedly considering measures to stop overseas semiconductor manufacturers, such as Qualcomm and TSMC, from producing advanced chips based on designs created by Chinese firms, including Huawei, Alibaba, and ByteDance. Such controls could complicate existing international arrangements, as Chinese chip designers still rely on overseas supply chains and manufacturing facilities.

Beijing is also weighing restrictions on foreign acquisitions of strategic AI companies, with a specific focus on agentic AI—systems capable of planning and performing multi-step tasks with limited human intervention. Furthermore, regulators are discussing whether the unauthorized leak or transfer of advanced AI technologies should be treated as violations of national security laws.

These deliberations follow a public push for openness. Just last week, President Xi Jinping spoke at an international AI conference, positioning China as a global advocate for open-source AI. He stated:

"We should oppose overstretching the concept of national security in the field of AI or placing one country's security over that of others,"

Xi Jinping, President of China

Xi further argued that AI development should be a symphony of global collaboration rather than a solo performance by a single country. However, the reported export curbs suggest a more selective strategy: continuing to offer AI services through cloud-based APIs while preventing the permanent overseas ownership of the underlying technology.

This pivot comes as Chinese developers narrow the performance gap with American systems. Last week, Moonshot AI released Kimi K3, a natively multimodal model with 2.8 trillion parameters designed for complex reasoning and long-horizon coding. Other companies, such as DeepSeek, Alibaba, and Zhipu, have released powerful open-weight models that have gained international adoption due to lower operating costs.

To further reduce reliance on foreign hardware, some firms are pursuing domestic alternatives. DeepSeek is currently developing an inference chip to reduce dependency on Huawei and Nvidia, coinciding with the company's first external funding round.

The move mirrors actions taken by Washington, which has imposed limits on high-end semiconductor exports to China and scrutinized the availability of frontier AI models due to cybersecurity concerns. Consequently, both superpowers are increasingly treating AI and chip designs as national security assets rather than commercial products.

The proposed restrictions have not been without internal friction. Several domestic companies have warned regulators that tighter controls could reduce international competitiveness, slow innovation, and weaken China's standing in the global AI race. China has tightened internal oversight by requiring key professionals at private firms, including DeepSeek and Alibaba, to obtain government approval before traveling overseas. This policy now extends to researchers, startup founders, and senior executives, expanding beyond state-owned enterprises and nuclear research sectors.

No final decision has been reached. Regulators continue to evaluate industry feedback before deciding whether to implement the controls. It remains unclear if the measures would apply to technologies already available overseas or focus exclusively on future releases.