Where AI Factories Get Built
Build economics are governed by power, cooling supply chains and social license. Europe's sovereign build-out is funded but lagging compute availability, while commercial gigafactories in the Nordics and Iberia race ahead.
Lock CDU capacity early, favor brownfield sites, and price permitting risk. Rack densities are trending toward 1 MW — liquid cooling is no longer elective.
19 + 13
EU AI Factories
Operational + associated Antennas, ~25k chips each
up to 5
EU AI Gigafactories
>100,000 processors each · €20B InvestAI
46%
Liquid cooling market share
By 2024 — now standard for new builds
12–18 mo
CDU lead time
Critical supply-chain bottleneck
75 (~$130B)
Projects disrupted
By community opposition, Q1 2026
1.05
Best-in-class PUE
Compal/ZutaCore two-phase cooling
From 10 kW racks to the megawatt rack
Rack power density is climbing two orders of magnitude — redefining the entire facility.
10 kW
Legacy
35 kW
Inflection
120 kW
Current AI
1 MW
2027–28
5 MW
Specialized
From elective to mandatory
Liquid cooling market
$B · $4.8B (2025) to $27.1B (2035)
Cooling technology mix
Approx. share of new AI builds
Commercial gigafactories in construction
| Project | Partners | Capacity | Power / notes | Timeline |
|---|---|---|---|---|
| Stargate Norway (Narvik) | Nscale, Aker, OpenAI | 100,000 NVIDIA GPUs | Renewable hydropower | Phase 1 H2 2026 |
| Start Campus (Sines, PT) | Microsoft, Nscale | 66,000+ Rubin GPUs | — | Late 2027 |
| Deutsche Telekom (Munich) | NVIDIA, Deutsche Telekom | ~10,000 Blackwell GPUs | ~0.5 ExaFLOPS | Operational |
| Mistral AI (Essonne) | Mistral, NVIDIA | 18,000 Blackwell GPUs | €830M project | In build |
Tracked build lifecycle
Each announced build is tracked across seven independent, separately-evidenced facets — announcement, site, grid, planning, procurement, construction and operations — so capacity is never confused with delivery.
Tracked projects are in review. Each build is published only once its facet evidence has a named reviewer and a primary source receipt.
The build constraints
Turn these signals into a placement decision
Discuss power, location, cluster architecture and TCO with a CUDO AI infrastructure specialist.
