What EuroHPC actually put on the table
On July 30 the EuroHPC Joint Undertaking opened a formal call for tenders inviting industry and public consortia to build and operate up to seven AI gigafactories across the European Union. Each facility is meant to pair AI-optimized supercomputers with advanced data centers, high-capacity storage, ultra-fast networking, secure cloud access, and specialized AI support services. The stated purpose is to give European industry, academia, and public authorities a place to train and run frontier models at a scale the continent currently lacks.
The financial architecture is the headline. The initiative marshals roughly 30 billion euros, with up to 10 billion coming from EU and national government budgets and an expected 20 billion from private investors. That public-private split is the mechanism Brussels is betting on to close a compute gap it has watched widen against the United States and China. We read it as an admission that grant funding alone cannot fund gigawatt-class AI infrastructure, and that Europe must pull in commercial capital to compete at all.
The two-lot funding structure
The money flows through two lots designed for different scales. Smaller facilities in the first lot can access up to 100 million euros in an initial phase and up to 400 million euros in a second. Larger projects in the second lot can draw up to 200 million euros initially and as much as 800 million euros later. Bidding closes on November 12, 2026, award decisions are expected in early 2027, construction is slated to begin that year, and facilities target operation around mid-2028.
That timeline is the part enterprise planners should mark. A gigafactory that starts construction in 2027 and switches on in 2028 arrives well after the current wave of American capacity. Executive Vice-President Henna Virkkunen framed gigafactory-scale compute as a strategic necessity for Europe as AI development accelerates, and the urgency is real. We would temper expectations on speed, though: the bureaucratic runway from tender to running silicon is long, and frontier model economics move faster than procurement cycles.
The sovereignty pitch and its hardware asterisk
The explicit goal is to reduce European dependence on US and Chinese computing power, giving the bloc a compute base it controls under its own rules on data, security, and governance. For regulated European enterprises in banking, healthcare, and government, a sovereign AI training environment inside EU jurisdiction is a genuine draw, because it addresses data residency and legal exposure that the American hyperscalers cannot fully resolve on European soil.
The asterisk is silicon. The gigafactories rely on letters of intent with American chipmakers AMD, Nvidia, and Qualcomm for the accelerators that will fill them. Europe controls the buildings, the power, and the operating rules, and it still buys the compute engines from the same US vendors that supply the hyperscalers it hopes to counterbalance. We think that dependency limits how far the sovereignty claim can stretch. Owning the data center does not equal owning the supply chain, and enterprises should size that gap honestly.
Why enterprises should care now
For a European CIO, the gigafactory program is a future option that is worth tracking even before any concrete is poured. It signals that a publicly backed, EU-governed alternative to AWS, Azure, and Google Cloud for frontier training is coming, and that alternative may carry procurement advantages for public-sector and regulated buyers. It also gives negotiating leverage. A credible domestic option, even a nascent one, strengthens the hand of European customers pressing the hyperscalers on data residency, pricing, and lock-in.
For US-based enterprise leaders, the read is competitive intelligence. Europe is deliberately building capacity to keep its AI workloads and data on the continent, which will shape where multinational customers place regulated workloads. We expect this to accelerate the fragmentation of AI infrastructure along jurisdictional lines, with the practical effect that a single global cloud strategy becomes harder to run. Planning for a multi-region, multi-jurisdiction AI footprint is prudent regardless of where your headquarters sits.
The power question Europe has not answered
Gigafactory-scale compute needs gigawatt-scale power, and that is where Europe's plan meets its hardest constraint. Electricity prices across much of the continent run well above US levels, and grid capacity in the regions most attractive for data centers is already tight. A facility that wins funding in early 2027 still has to secure firm power at a price that keeps training economics viable, and that is a taller order in Europe than in Texas or the American Midwest.
We would watch the power arrangements in every winning bid as closely as the compute specs. The projects that pair their gigafactory with dedicated generation or favorable long-term energy contracts stand a chance of running competitively. Those that assume the grid will simply provide risk the same interconnection delays now stalling projects everywhere. Europe's compute ambition is credible, and its energy reality is the variable that will decide whether these gigafactories become working infrastructure or expensive announcements.
Our read for infrastructure leaders
We see the EuroHPC tender as a meaningful policy signal rather than an immediate capacity option. The money is real, the governance intent is clear, and the two-lot structure gives serious consortia room to build at scale. For enterprises, the practical value over the next 18 months is strategic leverage and forward planning, not workloads you can place tomorrow. Sovereign compute is arriving in Europe, on a 2028 timeline, with a US-supplied silicon core.
Our guidance is to fold this into any multi-year AI infrastructure roadmap that touches European operations. Track which consortia win, where they land, and how they solve power, because those winners will become procurement counterparties for regulated European workloads by the end of the decade. In the meantime, use the mere existence of a credible sovereign alternative to sharpen your terms with the incumbent hyperscalers. Optionality has value even before it ships.
