China’s Ministry of Commerce has consulted domestic AI and semiconductor companies about a major expansion of technology export restrictions that could reshape the global AI supply chain, according to the Financial Times.
The proposed measures under discussion cover four areas: restricting foreign access to Chinese AI model weights, limiting cross-border transfers of training data, barring Chinese chip designers from manufacturing at TSMC and other foreign foundries, and tightening oversight of overseas technology acquisitions.
The consultations mark a significant shift in China’s approach to technology governance. For years, Beijing’s strategy focused on preventing US and allied export controls from cutting off Chinese access to advanced chips. Now, Chinese regulators are considering mirror-image restrictions of their own, keeping Chinese AI capabilities inside the country even if that means forgoing global market adoption.
The TSMC question
The most contentious proposal involves prohibiting Chinese companies from producing advanced processors at TSMC and other foreign foundries. Companies that would be affected include Alibaba, ByteDance, and Huawei, all of which design chips for AI workloads and fabricate them at TSMC’s leading-edge nodes.
The trade-off is stark. Moving designs to Semiconductor Manufacturing International Corporation (SMIC), China’s domestic foundry, would guarantee the state-backed company enough orders to fund its research and development. But SMIC trails TSMC by multiple process generations, and Chinese-designed chips would lose the performance advantage that TSMC’s fabrication delivers.
Sources familiar with the talks told the Financial Times that the proposal has generated significant pushback from companies that rely on TSMC’s manufacturing to remain competitive with US and Taiwanese chip designs.
Open-weight models at risk
The restrictions on AI model weights would affect companies such as DeepSeek and Moonshot, whose open-weight approach has become a hallmark of Chinese AI strategy. Unlike the closed models from Anthropic and OpenAI, Chinese models have typically been released with downloadable weights that anyone can deploy, modify, and fine-tune.
Under the proposed rules, foreign users could still access Chinese AI services remotely, preserving the ability to monetize API access, but downloading model weights for local deployment would be restricted. The change would eliminate one of the key advantages Chinese AI companies hold over their US counterparts in global markets.
The Ministry of Commerce has consulted Alibaba, ByteDance, and Zhipu about how to implement such restrictions without collapsing the open-weight ecosystem that has driven Chinese AI adoption abroad.
Parallel to US policy
The consultations come as the Trump administration revives its push to ban Chinese AI models from the US market. Tom’s Hardware reported July 21 that the administration is considering new restrictions in response to the Kimi K3 open-weight model from Moonshot, which demonstrated capabilities competitive with US frontier models.
The dynamic creates the possibility of a bifurcated AI ecosystem, where Chinese and US models are legally separated by export controls on both sides. China’s proposed rules would mirror the logic of US chip export controls: keep strategic technology within national borders and force foreign competitors to develop their own alternatives.
The final scope of the restrictions remains unclear. The Financial Times reported that the consultations are at an early stage and that significant industry opposition could narrow the proposals before they become law.
Sources: China is considering export controls on AI technologies (Tom’s Hardware, July 2026); China considers tighter export controls on AI models and chips (Reuters, July 2026)