China considers export controls on its own AI models and chip designs, FT reports

China's Ministry of Commerce is consulting the country's leading AI and chipmaking firms about new export controls on advanced AI models and the semiconductors that run them, the Financial Times reported July 21, citing people familiar with the discussions. Regulators have reportedly been in talks with AI developers including Alibaba, ByteDance and Zhipu, and with chipmakers including Huawei, alongside firms in their supply chain such as TSMC and Qualcomm.

Under discussion: limits on foreign users downloading the model weights behind China's most advanced AI systems, restrictions on transferring the data used to train them abroad, tighter controls on overseas chipmakers manufacturing semiconductors to Chinese designs, and a broader review of the export list itself — clearer licensing criteria and tighter end-user checks, the same toolkit the U.S. has used against Chinese chip buyers since 2022. Models named in the reporting as potentially affected include Alibaba's Qwen, ByteDance's Doubao and Z.ai's GLM-5.2 — open-weight systems foreign developers currently download and self-host as lower-cost alternatives to closed offerings from U.S. labs.

Nothing has been decided, and the Financial Times' own reporting is explicit that specifics — which chips, what performance thresholds, how open-source releases would be handled — remain undefined; Reuters said it could not independently verify the report, and the companies named did not respond to requests for comment. The direction is still notable: after two years of the U.S. treating advanced chips as a national-security export risk, Beijing is now weighing whether its own AI models are the same kind of asset — one to protect rather than give away.