Immersion Cooling
We run purpose-built immersion cooling data centres across the UK for AI, HPC and GPU-intensive workloads. Rack densities exceed 100kW with no thermal throttling - built for NVIDIA H100 and A100 clusters, LLM training and real-time inference at scale.
- 0.00
- Verified PUE or below
- 0kW–0kW
- Sustained rack densities
- 0W
- TDP per NVIDIA H100 SXM5 card
- 0–0 weeks
- Typical migration timeline
Why Air Cooling Fails Modern GPU Infrastructure
The NVIDIA H100 SXM5 operates at a TDP of 700 watts per card. A single 8-GPU server draws over 5.6kW. Multiply that across a standard 42U rack, and you are looking at densities that exceed 40kW before accounting for storage, networking, or redundancy.
Conventional computer room air conditioning (CRAC) units were not designed for this. They consume disproportionate energy, create uneven thermal zones, and require significant raised-floor infrastructure that adds cost without solving the fundamental heat problem. The result is thermal throttling, reduced GPU utilisation, and a PUE that routinely exceeds 1.5 in legacy facilities.
How Our immersion cooling Delivers Results
Built for UK enterprises running AI, HPC, and high-density compute
Verified PUE Performance
Measured PUE of 1.03 or below. You pay for compute, not cooling overhead.
Full TDP Support at Rack Level
Sustained rack densities from 50kW to 200kW. No thermal throttling. No derating.
Hands-On Engineering Support
Resident engineers handle installation, maintenance, and hardware changes. No remote-only ticketing.
UK Compliance and Certification
ISO 27001 and ISO 14001 certified. Infrastructure built to support your ESG reporting obligations.
AI and HPC Ready
Validated for NVIDIA H100 and A100 GPU clusters, LLM training, and real-time inference workloads.
Scalable from Day One
Start with a defined footprint. Scale capacity as demand grows, no migration, no contract renegotiation.
Frequently asked questions
What hardware can be deployed in an immersion cooling data centre?
Server hardware is compatible with a single-phase immersion cooling data centre environment, including NVIDIA H100, H200, and A100 GPU nodes, AMD EPYC and Intel Xeon CPU-based servers, and custom HPC configurations. We carry out a technical review with our engineering team prior to deployment to confirm dielectric fluid compatibility with your specific components, connectors, and thermal interface materials.
Is immersion cooling suitable for colocation, or is it only for private builds?
We offer immersion cooling as a colocation service. You do not need to own or build a dedicated facility. We provide the immersion tanks, dielectric fluid management, cooling infrastructure, power, and on-site engineering support. You bring your hardware. This model gives you the thermal performance of a purpose-built environment without the capital expenditure of constructing one.
How long does hardware migration into an immersion environment take?
For standard GPU server deployments, We complete initial migrations within two to four weeks from the signed agreement. This includes hardware compatibility review, fluid preparation, and on-site installation managed entirely by our engineering team.
Who manages the dielectric fluid maintenance?
We manage all fluid maintenance as part of the service agreement. This includes routine top-ups, quality monitoring, and full replacement cycles when required. No action is needed from your team, and there is no hardware downtime during routine maintenance.
Ready to upgrade your infrastructure?
Stop overpaying for legacy efficiency. Get a quote for colocation, immersion or a custom build in under 24 hours.