Europe has spent three years building a sovereign AI compute strategy. The irony embedded in NVIDIA’s ISC 2026 announcement is that 90% of it runs on hardware from a single US company, at least according to that US company. The figures deserve scrutiny before they become received wisdom.
According to NVIDIA’s announcement at ISC 2026, 35 new AI supercomputers built on NVIDIA infrastructure are being deployed or announced across 23 European countries. The company states its infrastructure powers more than 90% of Europe’s AI factory buildout, representing a combined 800 AI exaflops. Two things about that 800 number matter: it includes both deployed systems and announced commitments, and the methodology behind it isn’t disclosed. The “deployed or announced” qualifier is load-bearing, announced infrastructure isn’t operational infrastructure.
The named systems are real. Barcelona Supercomputing Center’s MareNostrum, BavariaAI’s Blue Swan, NAISS’s Mimer EuroHPC AI Factory, HLRS’s HammerHAI, and IT4LIA are established European HPC institutions with publicly documented programs. Their NVIDIA connection in these specific configurations is sourced through the press release, which is dead. Derivative coverage from the LA Times and ISC conference reporting corroborates the announcement’s existence, but all figures trace back to a single source.
The part nobody mentions: NVIDIA’s 90% figure includes systems that are committed but not yet operational. That’s a meaningful distinction for any European institution evaluating its actual compute availability today versus its planned capacity in 12-18 months. Announced exaflops aren’t available exaflops.
Why this matters for infrastructure practitioners. European AI compute has been a bottleneck. The EU’s AI strategy explicitly identified compute availability as a constraint on European AI development, and the AI factories initiative, which these supercomputers are part of, is the structural response. NVIDIA’s dominant position in that buildout has two implications: first, it means European AI development is deeply dependent on NVIDIA’s supply chain, pricing, and export control exposure. Second, it means NVIDIA’s European infrastructure business is large and growing, which shapes how the company prioritizes European regulatory relationships.
The sovereignty angle is genuine. If Europe’s stated goal is reducing dependency on US technology infrastructure, a compute backbone that’s 90% NVIDIA, again, by NVIDIA’s own accounting, is structurally in tension with that goal. That tension isn’t resolved by the announcement; it’s illustrated by it.
Disputed Claim
What to watch. Watch for independent verification of the 35-system count and country coverage, ISC 2026 conference proceedings and EuroHPC Joint Undertaking documentation would be the places to look. Watch also for the “announced vs. deployed” breakdown: how many of the 800 exaflops are operational today, and how many are committed but not yet live? That number changes the picture significantly. And watch for European competitor responses, AMD, Intel, and European hardware initiatives that might be positioning against NVIDIA’s dominant market position.
TJS synthesis. NVIDIA’s ISC announcement is vendor marketing with real infrastructure substance underneath it. The named systems are real, the European HPC buildout is real, and NVIDIA’s hardware dominance in this space is real. What’s not independently verified is the precise scale of that dominance or the operational status of the infrastructure claimed. The sovereignty observation, Europe building “independent” AI on US vendor hardware, is the editorial hook, but it’s also the genuine strategic tension that European policymakers are actively navigating. Cross-reference the Fable 5 European response (BRIEF-05) for the geopolitical layer of that same tension.