Sovereign AI: why the Nordics are building their own infrastructure

In May 2025, Wallenberg Investments confirmed one of the most telling contracts in Nordic AI history: a consortium with AstraZeneca, Ericsson, Saab and SEB building a Swedish "AI Factory" together with NVIDIA — offering secure and sovereign access to AI compute. When the country's largest industrial owner, a defence company and a major bank share the cost of their own AI infrastructure, that is not a technology story. It is a strategic signal.

Three forces drive the demand

Regulation. GDPR sets the frame, but the detailed drivers are sector rules: DORA for finance, the national cybersecurity act for essential services, and the EU AI Act demanding risk management and data governance in high-risk systems. Being able to account for where data and compute happen is becoming a procurement question.

Geopolitics. Compute has become a strategic asset. Public and semi-public organisations no longer only ask "where is data stored" but "who can switch us off". That question was improbable five years ago; today it is standard in the public procurements we take part in.

Economics. The third force is the most underrated: predictability. Large AI workloads on variable cloud capacity produce hard-to-forecast costs. Own or shared capacity — like the AI Factory consortium — is as much about cost control as about sovereignty.

What it means for the public sector

Denmark aims to lead the world in public-sector AI, Norway's strategy highlights public-sector experimentation, and the OECD published AI-procurement templates in 2025 requiring transparency, privacy and explainability. The direction is shared: public organisations want to use AI — on terms where data, models and compute stay under European control.

For Swedish agencies and publicly owned companies we see three concrete consequences:

1. Procurement criteria are tightening. Data sovereignty moves from a ticket item to the specification. Suppliers who cannot account for where compute happens drop out early.
2. Hybrid becomes the norm. Sensitive workloads on sovereign infrastructure, the rest on commercial cloud. Architecture — not ideology — makes the decision.
3. Skills become the bottleneck. Sovereign infrastructure without people who can operate and govern it is an expensive file server. Competence, not capacity, sets the pace.

Our position

We are neither against the cloud nor against self-hosting — we are against decisions without numbers. For every workload we calculate data sensitivity, cost profile and regulatory requirements, and recommend from there. But the development is clear: the Nordics are building their own AI infrastructure because the requirements being placed can no longer be met anywhere else. It is a market unfolding in front of us — and one of the reasons data sovereignty sits in architecture reviews, not in the appendix.

Sources

• Wallenberg Investments — AI Factory consortium with NVIDIA (AstraZeneca, Ericsson, Saab, SEB), 24 May 2025, wallenberginvestments.com
• OECD — AI in public procurement, templates published September 2025, oecd.org
• Denmark's digital strategy 2026–2029; Norway's national AI strategy (updated April 2025)