OpenAI and Google are selling their advanced AI models to Singapore-based subsidiaries of Alibaba, Baidu, and Tencent, exposing a gap in the US' AI controls
US groups have been supplying AI services to Singapore-based subsidiaries of Alibaba, Baidu and Tencent.
Context & Ripple Effects
Related coverage shows Chinese authorities were already discussing limits on domestic companies’ access to foreign advanced models and penalties for AI theft. That makes US providers’ sales to Singapore-based affiliates a live issue in the wider contest over where model access is governed.
The same companies have also been associated with open-sourcing models as a way to continue development under US restrictions. The reported affiliate route highlights that access to proprietary US services and domestic/open model strategies can coexist.
First-order effects
- OpenAI and Google gain a channel to sell advanced AI services to Singapore-based subsidiaries of Alibaba, Baidu, and Tencent, while those groups retain access to foreign model capabilities through affiliates.
- The reported transactions expose a practical boundary in US AI controls: restrictions aimed at Chinese groups may not map cleanly onto overseas subsidiaries.
Second-order effects
- US policymakers and enforcement bodies face pressure to clarify whether controls should assess an immediate customer’s jurisdiction, its ownership, or its downstream access to model outputs.
- Chinese technology groups have greater incentive to organize AI procurement and deployment across overseas entities, while US model vendors must weigh revenue from such customers against heightened compliance and political scrutiny.
Third-order effects
- If affiliate-based access remains permissible, AI controls are likely to shift from country-based restrictions toward rules focused on beneficial ownership, end use, and access to sensitive capabilities.
- The episode reinforces a bifurcating AI market in which Chinese firms combine domestic and open-model development with selective foreign-model access, rather than relying on either path alone.
The trend: This is one data point in the move from broad AI export restrictions toward harder-to-enforce controls over cross-border model access, corporate structure, and downstream use.