“Apply AI”: Brussels’ sovereignty push meets industrial reality

• The Commission will unveil “Apply AI” on 7 October 2025, pushing European-made (often open-source) tools into healthcare, defence and manufacturing, with €1bn reallocated to jump-start adoption.

• The heavy lifting still hinges on compute (EuroHPC) and capital (InvestAI), while late-stage funding gravity continues to pull European AI toward the US.

What happened

Brussels will present an “Apply AI” strategy led by EVP Henna Virkkunen on 7 October 2025 to “strengthen EU AI sovereignty”. The plan warns that foreign AI stack dependencies can be “weaponised” and sets out to seed public-sector demand for European platforms, backed by €1bn from existing programmes. Defence C2 and other critical systems are explicit priority verticals.

Context & data

• Compute backbone: EuroHPC JU cites a €7bn budget, 13 AI factories, 13 supercomputers and 10 quantum systems as of 2025—central to any sovereignty story.

• Capital plan (announced): InvestAI (launched 11–12 February 2025) aims to mobilise €200bn, including a €20bn fund for AI “gigafactories”. Mechanism: blend EU instruments with national and private co-investment.

• Funding reality check: In 2025, US investors accounted for over 70% of European AI deal value ($14.2bn), reflecting Europe’s thinner late-stage pools and slower processes.

• Policy drumbeat: The Commission’s summer dialogue on “Apply AI” pre-signalled a pivot from rule-setting to deployment, positioning the AI Act as an enabler rather than just a constraint.

My take

I like the demand-side wedge. If public buyers standardise on credible European stacks, that’s durable ARR and lowers CAC via frameworks instead of one-off pilots. Defence and broader govtech can support multi-year contracts with usage-indexed pricing, improving LTV and smoothing payback. The explicit nudge toward open source is pragmatic for integration and avoids lock-in.

But the supply-side economics still dominate. €1bn of re-prioritised funds won’t move the needle against frontier training costs and compute opex (power + cooling + GPUs). The real enablers are EuroHPC capacity actually reachable by startups (not just research allocations) and InvestAI’s ability to write growth cheques at US-style speed and size. Until then, Europe’s cost of capital gap will keep pushing top assets to incorporate or list in the US, diluting the sovereignty goal the strategy is chasing. For now, I’d underwrite theses in vertical AI (manufacturing, health, public safety) with proprietary data rights and middleware (observability, evals, safety, workflow) where gross margins are less compute-sensitive and switching costs accrue.

Risks & watch-list

• Execution risk: Will “Apply AI” and InvestAI convert into actual GPU hours and disbursed capital, or stay in announcement mode? Track disbursements and SME access to EuroHPC time.

• Procurement friction: Fragmented national frameworks can elongate sales cycles and depress GM if vendors must bundle compute at thin mark-ups.

• Funding gravity: With US investors supplying the majority of late-stage euros, ownership and IP control risk shifting offshore.

• Supply chain exposure: EU remains dependent on non-EU advanced chips/HBM/packaging; sovereignty is fragile without domestic capacity.