Use Case · AI Infrastructure
The infrastructure profile AI training at scale demands: large uninterrupted power, high-density liquid cooling, EU jurisdiction, and room to grow.
Foundation model training runs continuously for weeks or months. Any power interruption corrupts training runs. The 650 MVA adjacent substation with three 400 kV corridors provides the redundancy and capacity headroom AI operators require.
NVIDIA H100 DGX systems produce ~10 kW/rack. GB200 NVL72 racks exceed 120 kW. Air cooling is insufficient. The Bârzava river provides 13,068 m³/h of cooling water — sufficient for 100 MW+ with >50× safety margin at mean flow.
Hyperscalers and sovereign AI projects increasingly require EU-jurisdiction compute for model training on sensitive datasets. Romania provides full GDPR compliance, EU data residency, and the legal framework of an EU member state.
~3 ha footprint within an industrial-zoned perimeter. No rezoning required. Municipal strategic partnership signed. Accelerated permitting pathway available through EU PCI corridor status.
At 100 MW IT load and PUE 1.2, total facility power is 120 MW. Annual energy consumption: ~1,051,200 MWh. At Romanian industrial pricing (~€0.14/kWh) versus Frankfurt/Amsterdam (~€0.22–0.28/kWh), the annual electricity savings are €84–147M per year — enough to justify significant infrastructure investment at the site.
