Capturing waste heat from a liquid-cooled facility

The temperature is the whole story

Heat is only worth recovering if something can use it. That single constraint explains why waste heat from computing has historically gone to the atmosphere: a chilled-water loop running at 7 to 12 °C produces a heat stream too cool to be useful for almost anything, and lifting it to a useful temperature costs more energy than the heat is worth.

Warm liquid cooling changes the arithmetic at the source. A loop supplying fluid at around 40 to 45 °C and returning it warmer is producing heat at the low end of what district heating, space heating and process pre-heat can accept. The heat did not get more valuable; it got warmer, and warmth is what determines usability.

That is also why this has become a design conversation rather than a sustainability footnote. As of September 2026, high-density compute is moving to liquid cooling, and the rejected heat arrives at a temperature where the question "what could use this?" has answers — and in at least one major market, a legal obligation to find them.

Germany's Energy Efficiency Act requires new data centres to demonstrate that a defined share of their waste heat is reused: 10 % from July 2026, rising to 15 % in 2027 and 20 % in 2028. That converts heat recovery from an optional efficiency measure into a design requirement for any new facility in that market, and it creates the heat buyer that the rest of this section assumes.

What the fluid has to do at those temperatures

The loop fluid's role in a recovery scheme is to move heat without losing it and without degrading faster because it is warm.

It has to operate at the recovery temperature, not just the cooling temperature. That is the point most easily missed. A loop designed to reject heat may never need to run above 45 °C. The same loop serving a recovery circuit may be asked to run hotter in order to deliver useful heat — and the fluid, the seals, the hoses and the heat exchanger all have to be rated for that higher figure.

Its stability has to hold at that temperature. Corrosion and oxidation rates rise with temperature, and inhibitor consumption rises with them. A fluid that would hold its specification for years in a cool loop is working harder in a warm one.

It has to be compatible with both circuits. If the recovery side is a separate circuit — which it usually should be, so that building-side water quality cannot contaminate the compute loop — then there is a heat exchanger between them, and its materials are in contact with both fluids.

The data-centre grades in the range are designated across 0 to +80 °C, which leaves headroom above the temperatures a warm-water loop normally reaches. The full operating window, including any period when the loop runs hotter to serve recovery, is what the selection has to cover.

Storing the heat to match supply with demand

A facility produces heat continuously. Most heat buyers do not want it continuously. That mismatch is what a thermal store addresses, and it is why phase-change materials appear in these schemes at all.

The heat-storage grades in the range transition between +7 and +55.5 °C — which brackets the temperatures a liquid-cooled facility produces:

Note the spread in latent heat across the series. The +7 °C grade stores less than half what the +36 °C grade does, which means a store at the low end of the range needs substantially more material for the same energy. That is a sizing consequence, not a quality judgement — the transition temperature still has to suit the duty. The arithmetic for arriving at a mass is set out in how much phase-change material do I need.

The question that decides the project

Every recovery scheme comes down to one question that has nothing to do with the fluid: is there a buyer, and does their demand coincide with your output?

Three situations recur, and they are not equally favourable:

• A nearby continuous demand — a district heating network, a greenhouse, a swimming pool with steady draw. The best case, because supply and demand track each other and little storage is needed.
• An intermittent demand — a process that needs heat in batches. Workable, but it needs either storage or a rejection path that carries the load in between.
• No demand, or a distant one — the heat is available and there is nowhere for it to go. This is the common case, and no amount of fluid or material selection changes it.

That third case is why the rejection path has to stay sized for the full load. A recovery circuit is an addition to heat rejection, and a scheme designed on the assumption that recovery will always be available will underperform in exactly the conditions it was built for.

What to establish before specifying anything

Four figures, and three of them come from outside the cooling system:

1. The facility's return temperature at the point where heat would be extracted, across the load profile rather than at one operating point.
2. The receiving system's required temperature, and how much of the year it needs it.
3. The timing overlap between the two, which decides whether a store is needed and how large.
4. The full-load heat rejection capacity that remains necessary for the hours when recovery cannot take the heat.

Those four decide whether a recovery scheme is worth building, what the fluid has to tolerate, and whether a thermal store earns its place. The fluid selection follows from them — it is the medium, not the decision, and the same thermal storage principles apply to sizing the store as to any other.