- Lead. China’s Ministry of Commerce is consulting domestic AI and semiconductor companies on a set of export control measures that would restrict the overseas transfer of advanced AI models, training data, and model weights — and potentially ban Chinese firms from using foreign foundries like TSMC to manufacture chips designed in China.
- Fact. The proposed controls, reported by the Financial Times on July 21, mirror the logic of US restrictions on Nvidia chips: Beijing increasingly views advanced AI as a critical national asset requiring the same strategic protections it has applied to rare earths and other strategic technologies.
- Stake. If enacted, the measures could reshape access to Chinese AI capabilities globally, complicate existing partnerships between Chinese AI groups and Western cloud providers, and open a new front in the technology war between Beijing and Washington.
Regulators led by the Ministry of Commerce have been in discussions with leading domestic AI and chipmaking groups — including Alibaba, ByteDance, and Zhipu AI — about how to prevent China’s advanced technologies from being acquired or replicated by foreign actors, according to the Financial Times report. The consultations cover several distinct categories: restricting the download of AI model weights by foreign users, limiting the export of training data used to develop frontier models, controlling overseas acquisitions of strategically important Chinese AI or semiconductor intellectual property, and — most significantly for the chip industry — examining whether to prohibit Chinese companies from using overseas foundries such as TSMC for the manufacture of chips designed on Chinese intellectual property.
The TSMC Dimension
The potential ban on Chinese firms using TSMC is the most consequential element for the semiconductor industry. Several Chinese chip designers currently use TSMC’s advanced nodes to produce chips that cannot yet be replicated at China’s domestic fabs. A formal prohibition would accelerate the push toward domestic production at SMIC and other Chinese foundries, while also creating difficulties for companies like Qualcomm, which has revenue exposure to Chinese customers who design chips using its architectures and then manufacture them at TSMC. The measure would in effect be a mirror image of the US foreign direct product rule that restricts TSMC from manufacturing chips for certain Chinese entities without an export licence.
That symmetry is intentional. China’s rationale, as described in the FT report, is that advanced AI has become a critical national asset and that the tools for protecting it should resemble those the United States has used to restrict Chinese access to Nvidia hardware. The argument gained credibility after DeepSeek’s V4 model demonstrated that China could train frontier-class AI on domestic Huawei Ascend chips, reducing the asymmetry in hardware capability that had underpinned US export control strategy.
Consultations, Not Yet Policy
No final decision has been made. The Ministry of Commerce is gathering industry feedback before incorporating any measures into the next revision of China’s catalogue of technologies prohibited or restricted from export. The process mirrors the consultative approach Beijing has used for rare earth and critical mineral controls — both of which ultimately became policy after similar rounds of industry engagement.
These developments follow earlier reports that China was consulting Alibaba and ByteDance specifically on curbing overseas access to AI models. The current FT report suggests those consultations have broadened in scope to include chip manufacturing restrictions and training data controls — a more comprehensive architecture than the model-access limits previously discussed.
Global Implications
Western AI developers that have integrated Chinese model APIs or datasets into their products would face potential disruption if model weight exports are restricted. Cloud providers that offer Chinese AI capabilities to enterprise customers would need to restructure those arrangements. For the semiconductor industry, the TSMC dimension adds another variable to what is already one of the most fragmented supply chains in technology history — with geopolitical boundaries increasingly dictating which companies can manufacture where.