Answers are grounded in this paper and cite their sources.
In this position paper we argue that AI sovereignty is not technological autarky or a requirement that every component beEuropean. Rather, sovereignty is a graduated model of control, specific to a use case and not a binary choice between fullindependence and total dependence. Pragmatic AI sovereignty is the ability to understand and govern critical dependencies, maintain feasible alternatives, and make meaningful choices within acceptable costs, risks, and timeframes. We propose assessing sovereignty through four dimensions: (1) jurisdictional control; (2) data legality, integrity, and residency; (3) algorithmic, architecture, and training/inference transparency; and (4) operational continuity—across the layers of the AI stack. Further, the paper introduces an AI sovereignty graph to represent dependencies, feasible deployment routes, mandatory constraints, and the cost and time required to transition between routes. The paper also highlights independent, system-level evaluation of end-to-end AI systems and their dependencies as a practical contribution that research infrastructures such as the Barcelona Supercomputin Center (BSC) can strengthen. For European policymakers, this implies investing not only in models, but also in compute and energy capacity, data governance, independent evaluation, adaptation and deployment skills, and credible migration options. For institutions such as BSC, sovereignty is best treated as a use-case-specific capability that can be strengthened across the stack—not as a label attached to a model, provider, or hosting location.
No takes yet. Share an insight, caveat, or question.
Perino et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: