The EU's own AI Gigafactories tender calls for seven sites and at least €4–5 billion each. Matched megawatt for megawatt against Umpireal's real, locked node economics, the compute costs the same either way — but Umpireal's own capital requirement comes in at roughly 18% of it, because the compute itself was never Umpireal's balance-sheet risk to carry.
Matched at 930 MW, the EU's own stated minimum full-scale target across all seven AI Gigafactories: Umpireal's real, locked facility cost is €5.58 billion. Fully loaded with compute hardware at current market pricing, the total lands close to the EU's own €28–35 billion programme estimate — confirming the chips cost the same wherever they sit. The saving is entirely in what surrounds them.
Every figure on the left comes directly from the published Call for Tenders. Every figure on the right is Umpireal's own locked figure, or a same-hardware market price, at matched scale.
| 7 EU AI Gigafactories (EUROHPC-2026-CEI-AIGF-01) |
Umpireal, Matched | |
|---|---|---|
| Sites / Nodes | 7 consortia 4 medium-scale (Lot 1) + 3 large-scale (Lot 2) |
930 nodes ≈ Umpireal's own real 9-region plan, fully built |
| Minimum IT Load, Full Scale | 930 MW 4×120 MW (Lot 1) + 3×150 MW (Lot 2) |
930 MW |
| Minimum Compute Scale | 600,000 H100-equivalent units performance-normalised, not a literal chip count |
Matched at current-generation hardware |
| Capital Required From The Builder | €28–35bn consortium carries CAPEX + OPEX in full |
€5.58bn tenants supply their own compute under Colocation |
| Public Funding, Phase 1+2 | Up to €5bn EU, matched ≥1:1 by states ≥€10bn combined public money, capped at 17% of IT CAPEX |
None required to build node operating revenue covers ongoing costs |
| Copper | ~31,150 tonnes sourced industry-median intensity |
~3,630 tonnes ~88% less, Umpireal's working estimate |
Why the total system cost looks similar (€28–35bn either way): the chips are the majority of the cost, and they're priced by the market the same way regardless of who's buying them. Umpireal doesn't make compute cheaper. It's the row above — who has to fund it — where the two models actually diverge.
Chip pricing based on current market rates for comparable hardware; Umpireal's copper intensity is a working engineering estimate pending vendor confirmation. "H100-equivalent" is the EU's own normalisation unit for compute performance, not a literal count of physical processors — full methodology available on request.
Concentrating Europe's sovereign AI capacity into seven locations concentrates the risk with it. This is what a single grid fault, supply disruption, or physical incident actually costs under each model.
Seven large sites deliver compute and nothing else to the communities hosting them. 930 distributed nodes carry the same compute plus a set of civic and economic benefits no centralised site can offer.
Every node also removes a dangerous signalised junction, replaces it with a grade-separated crossing, and gives its host council a direct financial stake in the infrastructure sitting under their own road — none of which a gigascale campus, wherever it's sited, gives back to the place that hosts it.
Umpireal's contribution here sits inside the EU's own current resilience framework: the 2025 Preparedness Union Strategy, the Critical Entities Resilience (CER) Directive, and the NIS2 Directive, all of which name energy and digital infrastructure among the sectors requiring resilience of vital societal functions. This is a civil infrastructure contribution to that resilience, distinct from and outside of defence procurement, and complementary to, rather than competing with, the AI Gigafactories programme itself.
Node economics are self-funding through operating revenue once built, not dependent on ongoing subsidy to remain viable. Patent application IE 2025/0540, PCT filing in progress.
If you work in critical infrastructure planning, civil protection, or energy resilience policy and want to talk through where distributed infrastructure fits alongside centralised investment, we'd like to hear from you.