Infrastructure

Data Centre Project

A data-centre development project in Romania, designed to support future artificial-intelligence, high-performance computing and secure digital-infrastructure workloads.

In development

In development — no operational capacity

This is a development project, not an operating data centre. No computing capacity or infrastructure services are currently available.

A building has been secured and development work is in progress. This page contains only information currently approved for public disclosure.

01 — Status

Current stage

A building has been secured in Romania, and the project is currently in the assessment, planning and development stage.

Technical studies are evaluating the facility’s electrical, cooling, connectivity, physical-security and operational requirements. The regulatory and technical requirements applicable to the proposed facility are also being assessed.

No part of the facility is operational or available to customers.

02 — Purpose

Project purpose

The project is intended to establish scalable digital infrastructure capable of supporting future workloads such as:

  • Artificial-intelligence development and inference
  • High-performance computing
  • Secure software development and testing
  • Data-intensive applications
  • Enterprise and research computing

03 — Engineering

Technical direction

The design process is evaluating modular infrastructure suitable for both conventional enterprise systems and high-density GPU computing.

  • PCIe-based enterprise GPU servers
  • High-density GPU platforms
  • AI training and inference systems
  • High-performance storage
  • Low-latency networking
  • Air- and liquid-cooling solutions
  • Resilient electrical and monitoring infrastructure
Schematic diagram of a modular rack layout showing standard and high-density zones separated by contained cold and hot aisles.
Schematic — modular rack topology under assessment. Not an as-built layout.

04 — Infrastructure

Infrastructure planning

The project assessment covers:

  • Utility and backup power
  • Electrical distribution and resilience
  • Air and direct-to-chip liquid cooling
  • Heat rejection and environmental monitoring
  • Internet connectivity and diverse fibre routes
  • High-performance compute networking
  • Racks and structured cabling
  • Fire detection and life-safety systems
  • Physical access control
  • Cybersecurity and operational monitoring
  • Backup, recovery and business continuity
  • Phased deployment and future expansion
Schematic diagram of a closed cooling loop: heat rejection, coolant distribution unit, rack manifold, cold plate and GPU, with a parallel air-cooled path.
Schematic — air and direct-to-chip liquid cooling under evaluation.

05 — Location

Location

The project is located in Romania.

The site address is not published at this stage for security and confidentiality reasons. It may be disclosed directly to qualified counterparties under appropriate confidentiality and due-diligence arrangements.

06 — Capacity

Capacity and equipment status

No public power-capacity figure is claimed at this stage.

Any internal design ceiling, power model or expansion scenario remains a planning assumption until it is supported by the relevant engineering studies, utility documentation and project approvals.

No GPU, server, rack-scale system, cooling platform or network configuration should be understood as purchased, installed or operational unless explicitly confirmed in a future project update.

Schematic single-line diagram of a dual-path electrical distribution: utility supply, two transformers, UPS systems, standby generator, distribution units and rack busway.
Schematic — dual-path distribution concept. Redundancy scheme subject to engineering studies and utility availability.

07 — Compliance

Compliance and resilience

Applicable Romanian and European requirements, together with relevant data-centre, information-security, energy-efficiency, safety and operational standards, are being assessed as part of the project-development process.

This statement does not represent certification, regulatory approval or confirmation of compliance. Any future certification or conformance claim will be published only after independent verification and supporting documentation are available.

08 — Due diligence

Due diligence

NOVACORE AI S.R.L.’s public corporate identity and registry information is available on this website and can be independently verified through the Romanian Trade Register.

Non-public project documentation may be made available to qualified counterparties under appropriate confidentiality terms and subject to internal authorisation.

Verified milestones and technical information will be published progressively when they are supported by documentation suitable for counterparty review.

09 — Evaluation

Compute technologies under evaluation

The project is evaluating a range of enterprise AI, inference, rendering and high-performance-computing technologies. The publicly available specifications below provide a reference for the platforms included in our technical assessment.

This list describes candidate technologies only. It does not represent a purchase, installation, partnership, final architecture or currently available capacity. GPU platforms from NVIDIA and AMD are among the technologies being evaluated; no manufacturer, model, quantity, supplier or final configuration has been selected or publicly committed.

GPU accelerators

GPU Architecture Memory Bandwidth Max power Format
NVIDIA H100 SXM Hopper 80 GB HBM3 3.35 TB/s 700 W SXM
NVIDIA H200 SXM Hopper 141 GB HBM3e 4.8 TB/s 700 W SXM
NVIDIA B200 SXM Blackwell 180 GB HBM3e 8 TB/s 1,200 W SXM
NVIDIA B300 SXM Blackwell Ultra 288 GB HBM3e 8 TB/s 1,400 W SXM
NVIDIA RTX PRO 5000 Blackwell 48 / 72 GB GDDR7 ECC 1.344 TB/s 300 W PCIe 5.0
NVIDIA RTX PRO 6000 SE Blackwell 96 GB GDDR7 ECC 1.6 TB/s 600 W PCIe
NVIDIA L40S Ada Lovelace 48 GB GDDR6 ECC 864 GB/s 350 W PCIe
NVIDIA A100 SXM Ampere 80 GB HBM2e 2 TB/s 400 W SXM
AMD Instinct MI300X CDNA 3 192 GB HBM3 5.3 TB/s 750 W OAM/UBB
AMD Instinct MI325X CDNA 3 256 GB HBM3e 6 TB/s 1,000 W OAM/UBB
AMD Instinct MI350X CDNA 4 288 GB HBM3e 8 TB/s 1,000 W OAM/UBB

← Scroll to see the full table

Complete server platforms

Platform GPU configuration Aggregate GPU memory Reference power Type
NVIDIA DGX H100 8× H100 SXM 640 GB ~10.2 kW 8U
NVIDIA DGX H200 8× H200 SXM 1,128 GB ~10.2 kW 8U
NVIDIA DGX B200 8× B200 SXM 1.44 TB ~14.3 kW 10U
NVIDIA DGX B300 8× B300 SXM 2.304 TB ~14 kW 10U
RTX PRO AI Server up to 8× RTX PRO 6000 up to 768 GB GPU up to 4.8 kW OEM 4U–8U
AMD MI300X Platform 8× MI300X 1.536 TB GPU up to 6 kW UBB/OEM
AMD MI325X Platform 8× MI325X 2.048 TB GPU up to 8 kW UBB/OEM
AMD MI350X Platform 8× MI350X 2.304 TB configuration dependent OEM

← Scroll to see the full table

Reference power figures are manufacturer values for the equipment, not facility-capacity figures. A complete deployment calculation must also include CPUs, memory, storage, networking, power-conversion losses, cooling, redundancy and auxiliary systems.

Rack-scale systems under assessment

Platform Configuration Published memory / interconnect
NVIDIA GB200 NVL72 72 Blackwell GPUs, 36 Grace CPUs Rack-scale NVLink fabric
NVIDIA GB300 NVL72 72 Blackwell Ultra GPUs, 36 Grace CPUs ~20 TB GPU memory, 130 TB/s NVLink
NVIDIA HGX B200 8× B200 modules 1.44 TB HBM3e
NVIDIA HGX B300 8× B300 modules ~2.3 TB HBM3e
AMD MI325X UBB 8× MI325X modules 2.048 TB HBM3e
AMD MI350X UBB 8× MI350X modules ~2.3 TB HBM3e

← Scroll to see the full table

Rack-scale systems can impose substantially different requirements from conventional servers, including high rack power density, direct liquid cooling, coolant distribution, power shelves and specialised networking. Final platform selection will depend on engineering studies, workload requirements, software compatibility, energy availability, cooling design, procurement conditions and investment approval.

Infrastructure formats

PCIe GPU cards
Modular accelerators for compatible enterprise servers. Candidate platforms include L40S and RTX PRO 6000 Blackwell Server Edition.
SXM GPU modules
High-density NVIDIA accelerators designed for compatible HGX or DGX platforms with NVLink and NVSwitch connectivity.
HGX baseboards
Integrated multi-GPU modules used by qualified OEM server manufacturers.
OAM/UBB platforms
High-density accelerator infrastructure used for AMD Instinct MI300X, MI325X and MI350X systems.
DGX systems
Complete NVIDIA platforms integrating GPUs, CPUs, NVSwitch, storage, networking and the associated software environment.
NVL rack-scale systems
Integrated, liquid-cooled rack-scale platforms combining compute trays, processors, GPUs, NVLink switching, networking and power infrastructure.

Verify us independently

The legal identity of NOVACORE AI S.R.L. is published in full on this site and can be verified at the Romanian Trade Register.

Build your next critical system on a stronger foundation.

Tell us what you need to build, modernise or operate. We will respond with a technical point of view, not a sales pitch.