Services
GPU Compute, Private Cloud & Hosting
A Romanian infrastructure programme for accelerated computing, private and hybrid cloud, high-performance storage, backup and disaster recovery.
The planned service is intended to support organisations that require controlled access to GPU infrastructure for artificial intelligence, high-performance computing, rendering, analytics and other data-intensive workloads.
IN DEVELOPMENT — NOT YET AVAILABLE
The infrastructure is currently in development. No GPU capacity, cloud environment or hosting service is operational or available for purchase today.
Registering your interest helps us understand technical and market requirements. It does not reserve capacity or create a commercial commitment for either party.
01
Planned service portfolio
On-demand GPU compute
Flexible access to accelerated computing resources for development, testing, inference, rendering and time-limited workloads.
The final allocation model, instance types, minimum commitments and billing units will be established after the infrastructure has been commissioned.
Dedicated GPU infrastructure
Single-customer GPU servers or reserved compute environments for workloads requiring predictable performance, stronger isolation or sustained utilisation.
Potential configurations remain subject to platform selection, procurement, power availability, cooling design and commercial agreement.
AI training and inference
Infrastructure designed to support model development, fine-tuning, evaluation and production inference.
Available frameworks, model sizes, precision formats and scaling options will depend on the final GPU architecture, interconnect, storage and software environment.
Private cloud
Isolated compute, storage and networking environments designed around a customer’s security, governance and operational requirements.
Planned options include dedicated tenancy, controlled administrative access, private networking and integration with customer identity systems.
Hybrid-cloud integration
Connectivity and operational integration between customer infrastructure, public-cloud services and NOVACORE AI’s planned compute environment.
Each implementation will be designed around explicit security boundaries, data flows, routing, identity, encryption and responsibility models.
High-performance storage
Storage services intended for datasets, model artefacts, checkpoints, analytics and high-throughput computing workflows.
Performance tiers, capacity, replication, retention and data-lifecycle options will be defined after the storage architecture is selected and validated.
Backup and disaster recovery
Planned backup, replication and recovery services for supported systems and workloads.
Recovery objectives, retention, replication location, restoration testing and customer responsibilities will be defined contractually for each service.
02
Target workloads
The planned environment is being assessed for:
- large-language-model training and fine-tuning
- AI inference and model serving
- retrieval-augmented generation
- computer vision
- scientific and engineering computing
- simulation
- rendering and media processing
- data analytics
- enterprise AI applications
- private research and development environments.
Workload acceptance will depend on technical suitability, resource availability, security requirements, applicable law and the future acceptable-use policy.
03
Compute platforms under evaluation
The technical assessment includes a range of current and emerging enterprise GPU platforms from NVIDIA and AMD, including PCIe servers, high-density multi-GPU systems and liquid-cooled rack-scale architectures.
Candidate technology families include:
- NVIDIA Hopper platforms
- NVIDIA Blackwell and Blackwell Ultra platforms
- NVIDIA RTX PRO enterprise systems
- NVIDIA DGX, HGX and NVL architectures
- AMD Instinct CDNA 3 and CDNA 4 platforms
- AMD OAM/UBB multi-GPU systems.
This list describes technologies under evaluation. It does not indicate that a particular platform has been ordered, installed or selected for a future commercial service.
04
Planned infrastructure architecture
The service architecture is being developed around the following components:
- accelerated compute nodes
- high-bandwidth GPU interconnects
- low-latency scale-out networking
- high-performance shared storage
- object storage for datasets and artefacts
- isolated customer networks
- secure management infrastructure
- workload scheduling and orchestration
- centralised monitoring and logging
- backup and recovery systems
- metering and capacity management
- air- and direct-liquid-cooling options.
Final specifications will depend on verified power availability, cooling capacity, facility design, equipment selection, procurement and commissioning.
05
Security and data governance
The planned service model is being designed to support controls such as:
- tenant and workload isolation
- role-based access
- multifactor authentication
- encrypted network communications
- encryption options for stored data
- secure administrative access
- network segmentation
- centralised logging and monitoring
- controlled change management
- vulnerability and patch-management processes
- backup protection
- incident-response and recovery procedures
- defined data-location, retention and deletion requirements.
The precise controls, responsibilities, service levels and data-processing terms will be specified in the applicable service documentation and customer contract.
No security control can eliminate every risk. The objective is to apply proportionate safeguards, maintain operational visibility and support effective response and recovery.
06
Data location
The infrastructure is planned for deployment in Romania.
Data residency, replication, backup location and permitted cross-border transfers will be defined for each future service. No specific residency guarantee applies until it is documented in a signed agreement.
07
Service and access models under consideration
Future access options may include:
- consumption-based GPU access
- reserved GPU capacity
- dedicated bare-metal servers
- isolated private-cloud environments
- managed AI infrastructure
- customer-managed infrastructure
- hybrid-cloud connectivity
- project-based compute allocations.
The final commercial, technical and support models have not yet been published.
08
Development and commissioning
Deployable capacity and rack density will depend on:
- confirmed electrical capacity
- cooling architecture
- selected GPU and server platforms
- network topology
- storage architecture
- resilience requirements
- regulatory and safety reviews
- successful system commissioning.
The service launch date will be announced only after the required infrastructure has been installed, tested and approved for operational use.
09
Procurement
Technical evaluation and procurement planning are in progress.
Platform names, quantities, deployment phases and availability will be published as commercial offerings only after the relevant decisions and records are confirmed and suitable for public disclosure.
10
Register your requirements
We are gathering technical requirements from organisations planning AI, HPC, private-cloud and data-intensive workloads.
Tell us:
- your preferred GPU architecture
- required GPU quantity and memory
- workload type
- expected utilisation period
- storage and network requirements
- data-residency requirements
- security and isolation needs
- desired deployment period
- support and management requirements.
Submitting these details helps inform capacity planning. It is not an order, reservation, quotation or guarantee of future availability.
11
Plan your next compute environment with us
Share your expected workload, scale, security requirements and target timeline. We will respond with a technical discussion based on the project’s verified development status.
Register your infrastructure requirements View company details
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.