Why Healthcare May Be the Hardest Test for Sovereign AI

By Henry Duah

Artificial intelligence is spreading quickly across the world’s economies, yet in regulated sectors, adoption lags what the technology allows. McKinsey, in their recent article, argues the cause is not immature technology but unease about control: leaders worry about their ability to build and run AI with some independence from foreign providers over data, infrastructure, operations and legal structures. Healthcare sits at the centre of that tension. One McKinsey partner put it plainly: in regulated industries such as healthcare, broad AI adoption is not happening because sovereign offerings are not yet available. This article asks why health is such a demanding test, and what leaders can do about it.

What sovereign AI means

In a survey of 300 executives, investors and government officials, McKinsey found that 71 per cent described sovereign AI as an “existential concern” or “strategic imperative” for their organisation’s goals. There is still no official definition. McKinsey’s working view is the ability to build, run and govern AI in line with one’s own rules, security needs and values, across four dimensions: territorial (where data and compute physically sit), operational (who manages, secures and can switch systems on and off), technological (who owns the stack and intellectual property) and legal (which jurisdiction governs access and compliance).

Two points matter for what follows. Sovereignty is a spectrum, not a binary and not every layer of the AI stack will necessarily be sovereign. The question is never “sovereign or not” but “how much control do we need, for which use?”

Governments are already moving

It is estimated that sovereign AI could become a $600 billion market by 2030, driven by public-sector and regulated-industry use cases that could push up to 40 per cent of AI workloads into sovereign environments. It is worth noting that these are projections, not settled outcomes. The World Economic Forum notes that Denmark unveiled a national AI supercomputer in 2024, with early pilots including an AI care companion, while Italy, Sweden, the UAE and India have launched sovereign initiatives of their own. In France, the government has said its public sector will lead by switching cloud and software providers so that core systems are built with European players, while deploying AI use cases to improve service delivery. Health ministries and hospital groups should expect to be asked where they stand.

Four reasons healthcare is the hardest test

The stakes of failure are human.

WHO’s guidance on large multi-modal models documents the risk of false, inaccurate, biased, or incomplete outputs that could harm people making health decisions, and warns of automation bias, in which clinicians or patients overlook errors or improperly delegate difficult choices to a model. In most industries, a wrong answer costs money. In health, it can cost a patient’s well-being. 

Liability is moving towards the deployer.

McKinsey describes a “liability squeeze”: courts increasingly hold organisations that deploy AI liable for failures such as bias or hallucination, while vendors try to cap their own liability. Sovereign approaches argue that an organisation can build a “liability firewall” through audit and control mechanisms. For a hospital group or ministry, being accountable for a system it cannot inspect is a difficult position to be in.

Dependence becomes a continuity risk.

This points to a trend away from reliance on a few global providers because of “kill switches”: service denials during geopolitical conflict. In other words, a clinical workflow that depends on a service that can be withdrawn is as much a patient-safety question as an IT one.

The rules are tightening.  

Laws on data (such as the US Cloud Act) and on AI systems (such as the EU AI Act), as well as regional localisation mandates, are pushing governments to reassess their dependence on foreign infrastructure. WHO recommends that governments assign a regulator to assess and approve large multimodal models intended for health care, and to introduce mandatory, publicly available post-release audits by independent third parties for large-scale deployments. In Europe, the European Health Data Space already requires that the reuse of health data occur under a permit from a health data access body, within secure processing environments from which no personal data can be downloaded.

Three questions to take back to your organisation

• First, which of our workloads touch data or decisions where losing control would be unacceptable?
• Second, who can switch our AI systems off or change them, and under whose law?
• Third, can we audit what they do and present the results to a regulator or a patient? 

Healthcare will not be the first sector to settle these questions, but it may be the one that forces everyone else to answer them. As framed, in an AI-driven world, the question becomes: how much of your intelligence will you truly own?

References

[1]  Ustun, A., Bennici, L. and Krawina, M. (2026). What is sovereign AI? McKinsey Explainers, 6 March 2026. https://www.mckinsey.com/featured-insights/mckinsey-explainers/what-is-sovereign-ai

[2]  Ustun, A., Tournesac, A., Bennici, L., Takkar, K. and Drahmoune, N. (2025). The sovereign AI agenda: Moving from ambition to reality. McKinsey & Company, 18 December 2025. https://www.mckinsey.com/capabilities/tech-and-ai/our-insights/tech-forward/the-sovereign-ai-agenda-moving-from-ambition-to-reality

[3]  Letzing, J. (2024). What is ‘sovereign AI’ and why is the concept so appealing (and fraught)? World Economic Forum, 13 November 2024. https://www.weforum.org/stories/technological-innovation/what-is-sovereign-ai-and-why-is-the-concept-so-appealing-and-fraught

[4]  World Health Organization (2024). WHO releases AI ethics and governance guidance for large multi-modal models. News release, 18 January 2024. Full guidance: Ethics and governance of artificial intelligence for health: Guidance on large multi-modal models. https://www.who.int/news/item/18-01-2024-who-releases-ai-ethics-and-governance-guidance-for-large-multi-modal-models

[5]  European Commission, Directorate-General for Health and Food Safety. European Health Data Space Regulation (EHDS). Page consulted 21 October 2026. Regulation published in the Official Journal of the EU on 5 March 2025. https://health.ec.europa.eu/ehealth-digital-health-and-care/european-health-data-space-regulation-ehds_en

Get innovation insights from The FutureList weekly. Subscribe to our newsletter here