ResitaData (Resita Data Infrastructure SRL) is coordinating the RES1 site — a pre-development AI data center location in Reșița, Caraș-Severin, Western Romania, EU. The site is directly adjacent to a 650 MVA Transelectrica 400/220/110 kV substation, with an indicative connectable power range of 50–200 MW for a data center consumer, subject to the formal ATR grid connection study under ANRE Order 59/2013. For AI and HPC developers, this is the infrastructure profile that matters — not proximity to a legacy hub city.
Why AI workloads need this infrastructure profile
AI training and inference at scale are defined by three infrastructure requirements that are increasingly difficult to satisfy simultaneously in primary European markets: high sustained power (50–200 MW and above), water-based cooling (to achieve PUE below 1.3 for GPU cluster densities), and EU data sovereignty (required for GDPR compliance, EU AI Act obligations, and public sector contracts).
The RES1 site satisfies all three. The challenge in Frankfurt, Amsterdam, and Dublin is that grid access now requires 3–5 year queues or is under moratorium. A developer commissioning an ATR at RES1 in May 2026 can have a grid connection study result within 6–12 months — before a Frankfurt queue ticket would move to active study.
Power infrastructure for AI at RES1
The Transelectrica 400/220/110 kV substation at Reșița was upgraded in 2024–2025 to 650 MVA installed capacity. Three 400 kV corridors converge at this node: Porțile de Fier (117 km, operational April 2024), Pančevo Serbia (131 km, both circuits operational January 2025), and Timișoara (under construction). The node carries four EU PCI (Projects of Common Interest) designations under ENTSO-E TYNDP 2024 — providing accelerated permitting protection under EU Regulation 2022/869.
For AI training infrastructure, the 400 kV connection provides: high-voltage supply to the site transformer; N-1 security from three independent corridors; no single point of failure at the grid level; and direct interconnection to Romania's hydroelectric generation base (Porțile de Fier cascade) via the Corridor 1 link.
Cooling for GPU clusters — the Bârzava advantage
Modern AI GPU clusters — NVIDIA H100, H200, GB200 NVL configurations — operate at rack power densities of 20–100+ kW per rack. At these densities, air cooling is insufficient; water-cooled infrastructure (rear-door heat exchangers, direct liquid cooling, or immersion cooling) is required. Water availability at scale is a critical site selection criterion.
The Bârzava river at Reșița has a mean annual flow of 3.63 m³/s (13,068 m³/h), gauged by INHGA at the Reșița monitoring station. Per ASHRAE TC 9.9, a 100 MW IT load requires 30–80 m³/h of makeup water for evaporative cooling at PUE 1.2–1.3. The available flow exceeds this requirement by more than 50× at mean annual conditions. The climate advantage — mean ambient temperature 2–3°C below Frankfurt, London, and Amsterdam — extends free cooling hours and reduces chiller operating time.
EU jurisdiction and AI Act compliance
Romania has been a full European Union member state since January 2007. AI workloads operated from Romania-based infrastructure are within EU regulatory jurisdiction: GDPR applies for data processing; the EU AI Act (Regulation 2024/1689, fully in force from 2026) applies for AI system development and deployment; no cross-border data transfer mechanisms are required for EU-to-EU data flows. For AI companies working with EU public sector clients, financial institutions, or healthcare providers — all sectors with strict data residency requirements — a Romania-based training and inference cluster is a compliant EU-sovereign solution.
| AI Workload Type | Power Requirement | Cooling Type | RES1 Fit |
|---|---|---|---|
| LLM Pre-training (100B+ params) | 50–200 MW sustained | Water cooling required | ✓ Fits indicative ATR range |
| GPU Inference Farm | 10–100 MW | Air + DLC | ✓ Within range |
| HPC Scientific Compute | 20–200 MW | Water cooling | ✓ Fits |
| Edge AI / Latency-critical | 1–10 MW | Air cooling | ⚠ Not primary use case |
| Sub-5ms latency to Frankfurt | Any | Any | ✗ Reșița is regional, not hub |
Third-party validation: PPC Romania
PPC Romania, a subsidiary of Public Power Corporation (Greece's largest energy company, balance sheet >€5B), independently acquired 9.9 ha adjacent to RES1 in December 2025 for a 100 MW gas-fired power plant on the same Transelectrica 400 kV substation. This is the strongest available form of third-party grid validation: an institutional energy company with full due diligence capabilities found the same grid node worth committing capital to — independently.
Evaluate RES1 for AI infrastructure development.
Grid documentation, legal instruments, site KML, and cooling data available under NDA. ATR as first development step — not year 3.
Request AI Developer Briefing → Developer overview →Related: GPU Cloud use case · EU sovereignty · Grid infrastructure · Technical FAQ