256 GPUs · DLC
Liquid-cooled AI SuperCluster
Supermicro liquid-cooled NVIDIA HGX H100/H200 SuperCluster: 256 GPUs in five racks with direct-to-chip cooling, around 40% lower cooling energy than air. Specs and UK/EU quotes.
256 GPUs32× 4U liquid-cooled nodes
5 racksvs 9 air-cooled
~40% lesscooling energy
CDU includedin-rack coolant units
Same GPUs, half the floor space
The liquid-cooled SuperCluster uses Supermicro's 4U direct-to-chip nodes, each with eight HGX H100 or H200 GPUs and cold plates on GPUs, CPUs and switches. Rack-level coolant distribution units are included. The result is the same 256-GPU cluster in five racks instead of nine, with lower fan power and a data-centre PUE that is often the difference between being allowed to deploy and not.
When liquid is the right call
- Your colocation provider limits you on power per rack or floor space.
- Energy cost is a line item you are measured on.
- You are planning for Blackwell: the GB200 NVL72 is liquid-cooled only, so building the plumbing now is not wasted.
What we check in the quote
- Facility water availability and temperature, or whether a liquid-to-air CDU is needed.
- Commissioning and maintenance terms for the cooling loop.
- Like-for-like comparison against the air-cooled cluster so the premium is visible.
Get three quotes for the Liquid-cooled SuperClusterComparable pricing and lead times from authorised integrators, plus a cloud and hosted alternative so you can see the trade-off.
Request quotes