Liquid cooling is no longer exotic, but it is not automatically the right answer either. A practical checklist for deciding which side of the line your next refresh sits on, before the hardware is specified.
What to check
- Start with the rack power you are actually allocated today, and what your site or colo contract allows you to grow to. That number decides more than the hardware does.
- Check inlet air temperature and containment before assuming air cooling has run out. Poor containment, not the servers, is the usual reason a hall cannot take another kilowatt.
- Confirm what your facility can accept: direct-to-chip liquid needs a coolant loop and a CDU, while a rear-door heat exchanger can often be added to an existing rack with far less disruption.
- Ask who owns the loop. In colo, the answer determines whether liquid is even available to you, and on what SLA.
- Account for maintenance reality. Liquid cooled nodes change how hot-swap, spares and hands-and-eyes support work, particularly on out-of-hours callouts.
- Price the whole change, not the servers. CDUs, manifolds, pipework, leak detection, commissioning and any building work usually dwarf the delta on the compute.
- Consider a mixed estate. Dense GPU or high-core-count nodes on liquid, general-purpose virtualisation and storage on air, is a common and sensible outcome.
The ETON view
Our honest position is that most UK enterprise and hosting estates are not yet at the point where liquid is the obvious answer, and a lot of the pressure to adopt it comes from AI reference designs rather than from the workload the customer actually runs. If your racks sit in the 6 to 15 kW range and your containment is decent, well-specified air cooled servers will serve you for this refresh and cost considerably less to deploy.
Where the calculation flips is density you cannot avoid. If a workload demands GPUs at a scale that pushes a rack past what your hall can remove as heat, liquid stops being an upgrade and becomes the enabling condition, and the right move is to design for it early rather than retrofit. The expensive mistake we see is the middle path taken by accident: buying liquid-ready hardware without a facility plan, then running it air cooled at derated performance for two years.
If you tell us the power and cooling envelope you actually have, we will spec to it. That usually means a more useful conversation than a vendor configurator, because the constraint is almost never the server.
Related infrastructure
Category: Servers · Vendor: ETON Technology · Technology: Direct-to-chip liquid cooling, rear-door heat exchangers, air cooling · Last verified: 02 Sep 2026 · ~2 min read
Sourcing this kind of infrastructure? Talk to ETON about availability, lead time and pricing.
