The Sovereign AI Paradox: Why Owning the Infrastructure Is Not the Same as Controlling the Technology - a workload-level framework for control in a globally interdependent stack

This working paper argues that sovereignty should be assessed for a workload and its full service path under a stated disruption, not inferred from asset location or ownership. It proposes a five-part diagnostic (control, continuity, substitutability, data authority, economic viability) with two cross-cutting requirements, an ordinal five-rung evidence ladder from announced to independently stress-tested capacity, and a non-compensatory decision rule. A targeted, non-systematic review of 2024–2026 sources illustrates the framework with six country cases (United States, China, France, Canada, India, United Arab Emirates). The framework is a proposed diagnostic, not a validated index; three hypotheses and a staged validation agenda are set out. Not peer reviewed. Appendix B records the verification status of each reference.