The Physics and Economics of Reject Heat

To understand why waste-heat recovery is both promising and difficult, you have to start with the temperatures involved, because temperature is what determines whether heat is useful or merely present. A data center rejects an enormous quantity of thermal energy, but it does so at a low grade. Traditional air-cooled facilities exhaust air in the range of roughly 30 to 40 degrees Celsius, which is warm but not hot. Heat becomes useful in proportion to its temperature above the surroundings, and lukewarm air is close to useless for most industrial or heating purposes without further upgrading. This is the central engineering tension of the entire field: there is a vast amount of heat, but it is of poor quality.

The economics follow directly from the physics. Low-grade heat cannot be transported far, because it loses its modest temperature advantage to the pipes and the environment over distance. That means a heat customer must be near the facility, typically within a few kilometers at most, and often much closer. It also means the heat frequently needs to be boosted before it is useful. The tool for boosting is the heat pump, a device that uses electricity to move heat from a cooler source to a warmer sink, raising the delivered temperature to something a district heating network or an industrial process can actually use, commonly 60 to 80 degrees Celsius or higher. The heat pump consumes power, so there is an efficiency penalty and a cost, but it converts an unusable stream into a saleable one.

Liquid cooling changes this calculus substantially and is the reason waste-heat recovery is becoming more viable just as it is becoming more necessary. As chips grow hotter and denser, particularly the accelerators used for artificial intelligence training, air cooling runs out of headroom. Direct-to-chip liquid cooling and immersion cooling capture heat in a fluid rather than in air, and they capture it at a higher temperature, because the coolant sits right against the silicon. A liquid-cooled loop can deliver water in the range of 45 to 60 degrees Celsius or warmer, which is close enough to useful that the heat-pump lift required is smaller or, in some district systems, unnecessary. Higher-grade heat is easier to sell, so the industry's shift to liquid cooling, driven by chip thermodynamics rather than by environmental concern, happens to make heat recovery far more practical. My read is that this coincidence is the most important tailwind the field has.

The efficiency metrics matter here because they are what regulators now require operators to disclose. Power usage effectiveness, PUE, is the ratio of total facility energy to the energy delivered to the IT equipment; a perfect score is 1.0, and the difference above 1.0 is largely cooling overhead. Water usage effectiveness, WUE, measures liters of water consumed per kilowatt-hour of IT energy, which matters intensely in water-stressed regions where evaporative cooling trades electricity for water. A newer metric, energy reuse effectiveness, ERF or ERE, credits a facility for heat it exports to a productive use, and it is the metric that turns reuse from an unmeasured good deed into a reported number. Once ERE appears in a mandatory database, as the Energy Efficiency Directive contemplates, an operator that exports heat can show a materially better profile than an identical operator that does not.

One more factor shapes the economics: the heat must be available when the customer needs it, and demand is often seasonal. A district heating network wants the most heat in winter and comparatively little in summer, while a data center produces its reject heat at a steady rate all year. That mismatch means the reuse fraction is rarely one hundred percent, and it pushes some projects toward customers with flat, year-round thermal demand, such as certain industrial processes or indoor agriculture, rather than space heating alone. Thermal storage can smooth part of the gap, but the underlying point stands: a builder must match not only the presence of a heat customer but the shape of that customer's demand against the facility's steady output, and that matching is easier in some locations than others.

The practical upshot for a builder is a checklist that must be run before the site is chosen, not after. Is there a district heating network within economic reach, which in Europe often means yes and in most of the United States means no, because district heating is rare there. Is there a greenhouse operation, an aquaculture facility, a food processor, or an industrial user with a steady thermal appetite nearby? Can the facility be designed with liquid cooling from the start so the exported heat is high enough grade to be wanted? Each of these questions has a locational answer, and each answer is expensive or impossible to change once the concrete is poured. That is why the honest framing is not can this facility recover heat but rather should we build where recovery is possible in the first place.