What would meaningful control of an AI stack require?
Sovereignty claims look different when compute, models, data, deployment, energy, and skills are examined separately.
Related work: Technology & AI StrategyThe question
When an institution claims an AI capability is sovereign or strategically autonomous, which forms of control does it actually possess and which dependencies remain decisive?
Why it matters
Ownership, domestic location, and sovereign branding can conceal dependence on imported accelerators, external cloud control planes, model providers, specialised software, data rights, or scarce operating skills. The claim becomes more useful when control is tested at each layer rather than asserted for the stack as a whole.
What we would need to know
- Decompose the capability into compute, cloud, models, data, deployment, security, energy, and talent.
- For each layer, test the ability to operate, inspect, modify, substitute, and recover under disruption.
- Separate present capability from announced investment and planned capacity.
- Identify concentration, switching costs, time to substitute, and the institutional owner of each response.
What the answer could change
An answer would inform procurement, architecture, partnership, skills, and resilience without assuming that complete self-sufficiency is possible or desirable.
Where the question comes from
This question grows out of Dipankar Sarkar's research on AI dependencies and technology sovereignty. A full public analysis would use official strategies, procurement records, technical documents, and evidence of systems in operation.
This is a strategy question and method, not a claim that Stratessence has completed a government mandate or assessed a named client.