Sovereignty in brief
The issue. As AI becomes part of core operations, dependence on technology providers raises questions about business continuity, access, jurisdiction and control.
What is driving it. The reviewed sources point to geopolitical uncertainty, different national requirements, concentration in critical infrastructure and the difficulty of changing providers once technology becomes embedded.
How the thinking is developing. Selective control still dominates the expanded collection. The centre of gravity is managed interdependence rather than full-stack independence, but FTI and several newer sources make strategic dependency more prominent and give greater weight to regional or locally controlled capability for critical systems.
What we can learn
Sovereignty has several meanings. Data location, operational control, portability, recovery and the ability to replace a provider are different capabilities.
Similar mixed strategies can conceal different priorities. The same local/global architecture may be designed for compliance, continuity, bargaining power or access to innovation.
The evidence remains stronger on recommended architecture than demonstrated outcomes. The set does not establish a universally superior model or reliable financial return from greater independence.
The most consistent argument is that the degree of control should reflect the importance and failure mode of the business activity. Accenture, IBM, McKinsey / WEF, Strategy& / PwC and WEF / Bain all allow for combinations of local and global capability rather than uniform independence across the stack.
FTI strengthens the strategic-dependency argument by emphasising how hard external technology dependence can be to unwind once infrastructure and capabilities are embedded. The common ground is conditional dependence rather than isolation.
Where the thinking differs
PwC’s AWS-alliance article emphasises sovereignty controls within global cloud platforms. McKinsey / WEF argues that not all AI should be sovereign. EY’s Southeast Asian analysis, Red Hat and BCG place greater weight on locally controlled capability for critical infrastructure or decisions, while FTI gives more weight to strategic autonomy and regional infrastructure.
These are distinct responses, but they are not necessarily contradictory because they address different operating contexts and layers of the technology environment. Commercial perspectives also matter: PwC carries an AWS relationship; IBM and Red Hat supply technology; EY discusses its own offering. Those interests belong alongside the arguments when assessing them.
Plot’s interpretation: The common ground is conditional dependence. External technology remains valuable, while control becomes more important where loss of access, switching delay or jurisdictional exposure would materially affect the business.
In practice
Accenture’s authors describe AstraZeneca using locally controlled AI for pharmacovigilance in China and AWS public cloud for some research and clinical-development work elsewhere. The example illustrates how one enterprise can adopt different arrangements across activities and jurisdictions. It does not establish which arrangement produces better financial or operational results.
What we still don’t know
This collection is still strongest on published recommendations. It does not yet establish a systematic gap between what advisers recommend and what enterprise operators choose. Supplier thought pieces, company implementation accounts and a leader’s self-plot each provide different kinds of intelligence and should remain visibly separate.
Ideas to watch
The cost of control: migration expense, duplicated capability and the operating overhead of mixed environments.
Alternatives in use: evidence that switching, recovery and local fallback arrangements work under pressure.
Changes in the advice: whether institutions move toward more local capability or become more confident in safeguards within global services.
This edition is a snapshot. A change in the number of sources - or in the axes - does not demonstrate a change in institutional thinking. Tracking movement requires comparable, dated evidence.
Where does your organisation sit?
Plot one critical business activity, name the dependency it faces and state the control you would need if that dependency failed. The map should reveal how your definition of sovereignty compares with the published thinking, not prescribe a destination.
Follow this Plot for updates when institutions change their stance or material new evidence emerges. Plot your organisation to compare its assumptions with the published thinking.
Evidence note: 17 thought pieces are plotted at high editorial confidence; two relevant sources are held as context because they do not support both axes precisely enough. Source positions, supplier interests, policy proposals and operator examples remain distinct evidence types, and system-level accuracy is not yet independently calibrated.
