This contribution proposes a discussion on whether the European Union could consider treating foundation AI models as part of public digital infrastructure.

The article examines the structural asymmetry between large-scale model developers and downstream deployers, particularly in light of the AI Act’s documentation and accountability requirements. It explores whether a publicly guaranteed baseline model could reduce systemic risk, improve auditability, and strengthen competitive neutrality within the European AI ecosystem.

The objective is not to advocate for a predetermined institutional outcome, but to open a debate on governance design, incentive structures for data contribution, and the alignment between legal responsibility and epistemic capacity.

Feedback from technical, legal, economic, and policy perspectives would be very welcome.