Apply for Early Bird Access.webp)
AI Sovereignty: Why Europe Can't Afford to Get This Wrong
The continent that wrote the AI Act now has to figure out how to fight with it. The gap between regulation and capability has never been wider — or more consequential.
In September 2024, the European Commission published its first-ever Defence Industrial Strategy. Buried inside was a sentence that should have prompted more alarm than it did: European defence, it noted, remains critically dependent on non-European AI infrastructure for everything from logistics optimisation to intelligence analysis.
That sentence describes a problem that is no longer theoretical. It is operational.
The Infrastructure Nobody Talks About
When European defence ministries and NATO commands discuss AI sovereignty, the conversation usually centres on algorithms — who built the model, who owns the training data, who can audit the decision logic. These are legitimate concerns. But the harder problem is the layer underneath: compute.
The vast majority of AI inference workloads across European defence — the processing that turns sensor data into actionable intelligence, that runs threat classification models, that powers autonomous systems — runs on hardware and cloud infrastructure with non-European ownership structures. In a crisis, the availability of that infrastructure is not guaranteed under European law.
This is not a hypothetical. It is the reason the EU's ReArm Europe initiative explicitly includes sovereign compute as a priority investment area. It is why France's defence AI programme centres on the GAIA-X framework. And it is why, at the 2026 AI in Defence Summit, the session on AI sovereignty drew the largest delegation of any working group — including direct representation from seven EU member state ministries.
"Europe cannot rely on infrastructure it does not control when its security is at stake. Sovereignty in AI is not a technical preference — it is a strategic necessity. ReArm Europe is not just about procurement. It is about ensuring that the intelligence layer of European defence is genuinely European."
Andrius Kubilius — Commissioner for Defence and Space, European Commission
Three Dimensions of the Problem
Data sovereignty
Training effective defence AI requires data — vast quantities of it, much of it classified or operationally sensitive. The challenge is not generating that data; European defence forces produce enormous amounts of it. The challenge is the infrastructure required to store, process, and use it in ways that comply with both national security requirements and EU data governance frameworks. Currently, the two sets of requirements frequently conflict.
Algorithmic accountability
The EU AI Act classifies defence AI systems in a complex grey zone. Systems used exclusively for national security are partially exempt — but dual-use systems, and systems operated by civilian defence agencies, are not. The practical result is that European defence organisations face a regulatory environment with no clear precedent and significant legal exposure. The companies building defence AI for European clients face the same uncertainty.
Supply chain exposure
The semiconductor supply chain that underlies European AI capability remains globally distributed, with significant dependencies on non-European manufacturers for the highest-performance chips required for real-time inference. Europe's Chips Act has begun to address this — but the timeline for meaningful domestic capacity is measured in years, not months, while the operational requirement exists now.
What a Sovereign AI Defence Architecture Actually Looks Like
The honest answer is that no European country has fully solved this yet. But the most advanced programmes share a common architecture: a sovereign cloud layer combined with interoperability standards that allow allied systems to communicate without creating dependency relationships.
Germany's Bundeswehr Digital Initiative and France's SCAF programme both point toward this model. The missing element, consistently, is the institutional coordination layer — the agreements between ministries, agencies, and allied commands that allow a truly interoperable European AI defence architecture to function.
That coordination problem — not the technology — is what keeps European defence AI chiefs awake at night. And it is the problem that the 2027 AI in Defence Summit is explicitly designed to address.
The Stakes in 2027
By March 2027, Europe will have spent two full years operating under the AI Act, three years under ReArm Europe, and — if current trajectories hold — at least two more years of active AI-enabled conflict on the continent's eastern border. The theoretical debate about sovereignty is becoming an operational one.
The organisations that will shape the outcome — the ministries, commands, and technology partners that are building European sovereign AI capability right now — will be in the room at AIDEF27.
The AI in Defence Summit 2027 convenes on 1 March in Brussels. Apply for early bird access at aidefencesummit.eu.