Digital Sovereignty and AI: From Health Data Hosting to Control Over Models

Digital sovereignty is no longer an abstract concept. It has become a political, industrial, and economic reality, as evidenced by recent debates surrounding the Health Data Hub and European initiatives to regulate artificial intelligence.

According to the European Commission, the global AI market is expected to reach $1,800 billion by 2030, with increasing concentration in the hands of a few U.S. and Chinese players. In this context, Europe—and France in particular—must rethink its technological autonomy to avoid structural dependence.

In 2026, as data becomes the new oil and AI redefines the balance of geopolitical power, the question is no longer whether digital sovereignty is necessary, but how to build it in practice.

Key Points

• Digital sovereignty is not limited to the location of data; it involves legal, technical, and economic control over the entire value chain.

• The case of the Health Data Hub illustrates the limitations of European data hosting that is subject to extraterritorial laws such as the Cloud Act.

• Dependence on U.S. hyperscalers (cloud, collaborative suites, GPUs, AI models) creates a structural vulnerability for governments and businesses.

• AI sovereignty rests on four pillars: legal, infrastructural, technological (models), and economic.

• Europe has promising players such as Mistral AI, but full sovereignty requires a cohesive industrial ecosystem, local infrastructure, and sustained political will.

Structural Dependence on GAFA: Beyond the Cloud

Our public and private organizations rely heavily on American tools:

• Email (Outlook, Gmail)
• Office applications (Excel, Word, PowerPoint)
• Video conference (Zoom, Meet, Teams)
• Collaborative suites (Microsoft 365, Google Workspace)
• Cloud infrastructure (AWS, GCP, Azure)

This dependence is not merely technical. It is structural: more than 80% of CAC 40 companies use Microsoft 365 or Google Workspace, and the majority of government data passes through or is hosted by U.S. hyperscalers. Every piece of data produced, every document edited, and every meeting recorded feeds into an ecosystem that is not European. In response, the French government has launched alternatives, such as the Suite Numérique, to offer open-source tools developed in France. This suite, combined with hosting in France, will help address the challenge of dependence on American tools. It’s a step forward, but it’s not enough.

Why? Because the next battle is no longer fought solely over data or collaborative software. It’s fought over intelligence itself.

AI Sovereignty: The New Strategic Frontier

Artificial intelligence is changing the nature of the debate. The question is no longer just: Where is my data stored? But rather:

• Who trains the models that process my data?
• Where are they staying?
• Under which jurisdiction?
• What are the hardware and software dependencies?

AI sovereignty has several dimensions:

• Legal sovereignty: compliance with the GDPR and the AI Act, as well as protection against extraterritorial laws (notably the Cloud Act).
• Infrastructure sovereignty: control over data centers and GPUs.
• Model Sovereignty: Developing European Large Language Models Trained on Local Data.
• Economic sovereignty: financing and control of key players by European capital.

However, today the global market is dominated by U.S. players, whether in terms of models (OpenAI, Google, Anthropic), GPUs (Nvidia accounts for 95% of the AI GPU market), or infrastructure (AWS, Azure, GCP). Europe must step up its efforts to avoid becoming merely a consumer of technology.

Mistral AI: European Hope, Persistent Challenges

In Europe, Mistral AI embodies this ambition. The company has demonstrated that it is possible to develop competitive models in France. This sends a strong signal, but the industrial reality remains complex:

• GPUs are manufactured outside of Europe (NVIDIA, TSMC)
• Mistral still uses AWS and Azure for some of its training, despite its partnership with OVHcloud
• Capital is international, which raises the question of strategic autonomy

This does not call into question the technological quality of Mistral, but it does show that AI sovereignty cannot rest on a single player. It must be based on:

• A well-managed infrastructure (sovereign data centers, secure networks)
• Local computing capacity (access to GPUs, green energy)
• A comprehensive ecosystem (startups, research labs, public funding)
• A clear strategic commitment (industrial policies, government procurement).

Mistral’s recent announcement that it will open a data center in Sweden is a positive step, but it will take years to implement. Sovereignty cannot be decreed; it must be built.

This means:

• Economic trade-offs (investments in local infrastructure)

• Increased engineering effort (development of customized solutions)

• Consistency between rhetoric and implementation (no “marketing sovereignty”)

In turn, this ensures:

• Legal expertise (GDPR compliance and the AI Act, protection against extraterritorial laws)

• Operational expertise (resilience, security, performance)

• Strategic coherence (alignment with national and European priorities)

The debate surrounding the Health Data Hub has opened a window of opportunity; La Suite Numérique has begun to address the issue of tools; and Mistral AI has proven that Europe can compete in the realm of models. But AI sovereignty must not be a one-off issue. It must become:

• A European industrial strategy (funding, regulation, cooperation among member states)

• A requirement in requests for proposals (mandatory criteria)

• A Key Criterion for IT Departments (Technology Choices Aligned with Sovereignty)

• A choice embraced by technology companies (transparency, commitment, local innovation)

The issue isn’t about cutting ourselves off from the world. The question is: which critical assets do we want to retain control over? Today, artificial intelligence is clearly one of them.