What is VOLTANEUM immersion cooling?
VOLTANEUM immersion cooling cools servers directly in a non-conductive dielectric fluid to increase density, reduce airflow, lower noise and stabilize AI/HPC workloads.
GPU hosting for inference, fine-tuning and AI training
VOLTANEUM provides shared GPU capacity and dedicated GPU servers for intensive AI workloads.
4x H200 and 8x H200 immersion-ready profiles are designed for density, thermal stability and continuous operations.
VOLTANEUM key facts
- Published immersion colocation capacities: 10U, 20U and 37U.
- Target density up to 200 kW+ per tank depending on integration and electrical constraints.
- 1.03 target PUE for optimized high-density architectures.
- Immersion cooling billed at €0.07/kW, electricity excluding cooling metered separately.
- VOLTANEUM® proprietary dielectric fluid and France / Europe operations.
What the customer actually gets
VOLTANEUM GPU hosting is designed for inference, fine-tuning and AI training when thermal demand becomes continuous. GPU offers can be delivered as shared capacity or dedicated servers with immersion cooling.
Customers who do not want to manage a full cabinet can rent GPU capacity or a ready-to-use GPU server with visibility into power, operating system, management access and provisioning status.
VOLTANEUM offers and capacity
- RTX 3090 Cloud GPU — Capacité GPU dédiée, 0.6 kW, 0.39 € / hour, metered electricity excluding cooling 0 €/kWh. GPU infrastructure for inference, fine-tuning, AI training and HPC workloads.
- 2x RTX 3090 Cloud GPU — Capacité GPU dédiée, 1 kW, 0.74 € / hour, metered electricity excluding cooling 0 €/kWh. GPU infrastructure for inference, fine-tuning, AI training and HPC workloads.
- RTX 6000 Ada / L40S Cloud GPU — Capacité GPU dédiée, 0.8 kW, 1.01 € / hour, metered electricity excluding cooling 0 €/kWh. GPU infrastructure for inference, fine-tuning, AI training and HPC workloads.
- 4x RTX 3090 Cloud GPU — Nœud GPU 4x H200, 1.6 kW, 1.34 € / hour, metered electricity excluding cooling 0 €/kWh. GPU infrastructure for inference, fine-tuning, AI training and HPC workloads.
- 2x RTX 4090 Bare Metal — Capacité GPU dédiée, 1.2 kW, 1.41 € / hour, metered electricity excluding cooling 0.3 €/kWh. GPU infrastructure for inference, fine-tuning, AI training and HPC workloads.
- 4x RTX 3090 Bare Metal — Nœud GPU 4x H200, 1.6 kW, 1.74 € / hour, metered electricity excluding cooling 0.3 €/kWh. GPU infrastructure for inference, fine-tuning, AI training and HPC workloads.
- 4x RTX 4090 Bare Metal — Nœud GPU 4x H200, 2.4 kW, 3.02 € / hour, metered electricity excluding cooling 0.3 €/kWh. GPU infrastructure for inference, fine-tuning, AI training and HPC workloads.
- 4x RTX 6000 Ada / L40S Node — Nœud GPU 4x H200, 2.4 kW, 3.47 € / hour, metered electricity excluding cooling 0.3 €/kWh. GPU infrastructure for inference, fine-tuning, AI training and HPC workloads.
- H200 Shared Immersion Instance — Instance GPU H200 partagée, 3 kW, 3.9 € / hour, metered electricity excluding cooling 0.3 €/kWh. Hourly H200 shared GPU instance with 16 vCPU, 128 GB RAM, 500 GB NVMe, 10G networking and OS selection.
- 4x H200 Immersion Node — Nœud GPU 4x H200, 12 kW, 14.9 € / hour, metered electricity excluding cooling 0.3 €/kWh. Dedicated 4x H200 immersion GPU node with 96 vCPU, 768 GB RAM, NVMe storage, 25G/100G networking and OS selection.
- 8x H200 Ultra Immersion Node — Nœud GPU 8x H200, 22 kW, 28.9 € / hour, metered electricity excluding cooling 0.3 €/kWh. Ultra-dense 8x H200 immersion GPU node for AI training and HPC with 192 vCPU, 1.5 TB RAM, NVMe storage and 100G/200G fabric.
Key benefits
VOLTANEUM® combines immersion cooling, proprietary dielectric fluid, GPU density up to 200 kW+ per tank, tenant isolation, 1.03 target PUE, DCIM automation, 2D/3D monitoring and request/payment/provisioning workflows.
The VOLTANEUM® proprietary dielectric liquid provides non-conductive electrical insulation, stronger thermal stability, reduced exposure to dust, oxidation and vibration, and more predictable maintenance through fluid, filtration, acidity, moisture and dielectric strength monitoring.
The -40°C to 250°C range is indicated on VOLTANEUM® packaging. Quality control on the tested batch reports a -37°C pour point, a 196°C open flash point and observed dielectric breakdown voltage of 52 kV.
- Electrical insulation for compatible immersion-ready equipment.
- Direct heat transfer close to GPU, HPC and bare metal components.
- Operational protection against dust, vibration and thermal stress.
- Quality controls can cover fluid temperature, acidity, moisture, filtration and dielectric strength.
The VOLTANEUM® white paper download is available after entering a business email in the public page form.
Related pages
Frequently asked questions
Is VOLTANEUM suitable for HPC and GPU workloads?
Yes. The platform targets GPU clusters, intensive computing, inference, fine-tuning, dense storage and environments where classic air-cooled racks become limiting.
How does a datacenter colocation offer work?
The customer selects an offer, country and available datacenter, then pays the initial fees. After validation, the allocation is provisioned and visible in the DCIM portal.
Can I rent a bare metal server without renting a cabinet?
Yes. Dedicated physical server offers let customers rent a bare metal server with operating system selection, remote management access and customer monitoring.
What is an immersion cooling cabinet?
An immersion cooling cabinet is physical capacity in a dielectric bath, allowing a customer to rent a portion or a full cabinet for GPUs, ASICs, HPC or immersion-ready servers.
Can I bring my own hardware?
Yes, in BYOH mode, subject to technical validation. For immersion, the hardware must be dielectric-fluid compatible and validated before installation.