Infrastructure, Compute and European Digital Sovereignty

This chapter opens Part III, on infrastructure, and asks where AI runs and who controls it, because every model and pipeline executes on someone's hardware under some jurisdiction's law. It covers scarce accelerator capacity, on-premises and hybrid inference, European cloud, Gaia-X and the EuroHPC AI Factories, data residency after Schrems II, confidential computing, sovereign model choices, and the infrastructure code that keeps placement a decision rather than an accident.

AI reverses part of the long trend towards abstracting where a workload runs. Training and inference depend on scarce, unevenly distributed accelerators; the data that flows through a model is often the most sensitive the organisation holds; and a European organisation's regulatory position depends on whether personal data leaves the European Economic Area, which cloud regions and support arrangements can cause without anyone intending it.

Digital sovereignty, in the sense this book uses, is the ability of European organisations to control their data, infrastructure, and models. It is not a synonym for on-premises but a set of properties: the organisation knows where its workloads run, can show which law governs them, can change provider without rewriting its estate, and can continue operating if a provider or jurisdiction becomes unavailable. The European Commission's AI Continent Action Plan frames compute capacity, data, and skills as the three resources Europe must secure to compete, and is explicit that compute is the scarcest (15, European Commission, 2025).

Primary topics explored

1. Why placement is an architectural decision
2. Accelerator scarcity and capacity planning
3. Hybrid and on-premises inference
4. European cloud, Gaia-X, and the AI Factories
5. Data residency and third-country transfer risk
6. Confidential computing