Discussions about “AI sovereignty” generally focus on two things: investment and capabilities. Countries are encouraged to fund national champions that can develop frontier AI models, build compute clusters and assemble domestic data pipelines. But while these steps do matter, they cannot, by themselves, deliver true AI sovereignty. For that, a fully contestable and interoperable AI stack is essential. An AI stack contains multiple interdependent layers. Energy infrastructure feeds compute clusters. Compute (processing power) enables the training and operation of foundation AI models.

Jayant Sinha shows how contestability, interoperability, and accountability can be built into every layer.

Discussions about “AI sovereignty” generally focus on two things: investment and capabilities. Countries are encouraged to fund national champions that can develop frontier AI…