Use Case

GPU Cloud Infrastructure in Eastern Europe

RES1 Reșița — the most power-ready pre-development GPU cloud site in the CEE region. No grid queue. 650 MVA adjacent. EU jurisdiction. Water cooling. Available now.

GPU cloud providers — neoclouds, GPU-as-a-service platforms, and hyperscale AI compute operators — face a single defining constraint in Europe in 2026: grid access. The GPU hardware supply chain has recovered. The capital is available. The demand is real. The blocking factor is finding sites with immediate, substantive high-voltage power access that can be developed within 18–36 months. The RES1 site in Reșița, Western Romania is precisely this: a 650 MVA Transelectrica substation directly adjacent to a development-ready industrial site with institutional documentation in place.

What GPU cloud operators require — and what RES1 provides

GPU Cloud RequirementRES1 StatusSource
50–200+ MW sustained HV power650 MVA adjacent · ATR pendingTranselectrica · The Diplomat
Water cooling for high-density racks3.63 m³/s river flow · 50× marginINHGA · ASHRAE TC 9.9
EU jurisdiction for GDPR complianceRomania EU member since 2007European Commission
Industrial land — no rezoningIndustrial classification confirmedReșița Municipality
No multi-year grid queueATR = first step, not year 3vs Frankfurt 3–5 yr queue
Institutional counterpartyPartnership Agreement signed · Mayor Support LetterCity Hall Registry Nr. 28099
<10ms to Western Europe PoP~30ms to Frankfurt PoPNot a latency-optimised site

The GPU cloud market in Eastern Europe — why now

The GPU compute market in Europe is growing at rates that outpace available infrastructure. Hyperscaler GPU cloud capacity is projected to surge more than 6× by 2035 (GlobeNewswire, May 2026). The constraint is not GPU supply — it is power. Eastern Europe, and Romania specifically, is emerging as the region where power is available without the multi-year queues that define Western European markets.

Romania's data center market is growing at 19.93% CAGR through 2031. The country has established tech clusters in Bucharest, Cluj-Napoca, and Timișoara, with growing demand for AI compute from Romanian and EU enterprise clients. A GPU cloud operator establishing an Eastern European node in Reșița gains: Romanian client proximity, EU data sovereignty compliance, Balkans network connectivity (via the new Serbia 400 kV link), and a cost structure significantly below Western European alternatives.

Power cost advantage for GPU cloud economics

GPU cloud economics are dominated by power costs. A cluster of 4,000 H100 GPUs operating at 80% utilisation consumes approximately 32 MW continuously. At Romania's industrial electricity price of approximately €0.14/kWh versus the EU average of €0.19/kWh:

Annual electricity saving vs EU average: 32 MW × 8,760 hours × (€0.19 − €0.14) = €14M per year per 32 MW cluster. At 100 MW, this differential exceeds €43M annually. Over a 10-year infrastructure hold, this is a fundamental component of IRR.

Development path for GPU cloud operators

Step 1: NDA execution → data room access (grid documentation, legal instruments, site KML, cooling data). Step 2: ATR commissioning with Transelectrica — the first capital-deployed step. 6–12 months for ATR result. Step 3: Land acquisition via public tender (ANEVAR valuation minimum). Step 4: EIA, Building Permit. Step 5: Construction (18–24 months). Step 6: First power on. GPU cluster commissioning.

Can a neocloud operator develop RES1 directly?
Yes. The standard developer partnership structure (Structure A) gives a GPU cloud operator or neocloud majority equity in the Romanian SPV and full DBO authority. RDI retains a minority position and advises on municipal relationships only. The operator controls all commercial decisions: tenants, GPU configuration, power contracts, and operational management.
What GPU rack densities are supported?
Power density is limited by ATR outcome (total site power) and building design (per-rack cooling). The Bârzava water supply supports direct liquid cooling and immersion cooling at scale — enabling rack densities of 50–100+ kW per rack as needed for NVIDIA GB200 NVL and similar configurations. Specific cooling architecture is developer-designed.
Is the site NVIDIA DGX-Ready certifiable?
NVIDIA DGX-Ready Data Center certification is an operational-stage programme applied after facility commissioning. RES1 is a pre-development site. Once built and commissioned by a development partner, the operator can apply for DGX-Ready certification. The power density, cooling capacity, and EU jurisdiction profile are consistent with DGX-Ready requirements.

GPU cloud site — no queue, EU jurisdiction, water cooling.

Request GPU Cloud Briefing → vs Frankfurt →

Related: AI data center Romania · Hyperscale CEE · Europe grid crisis