As “Eastern Data, Western Computing” and large-scale compute infrastructure roll out, recovering data-center waste heat for district heating is shifting from a policy talking point to a calculable engineering and operating opportunity. Many MEP contractors still assume that cooling for AI data centers belongs only to IT majors and state-owned EPCs. In practice, once cooling water leaves the machine-room wall, heat upgrading, secondary networks and district heat-exchange stations are exactly the territory of large water systems and heat-pump engineering.
Dual-track value: do not stare only at PUE — waste heat is an asset too
In liquid-cooling retrofit assessments, the industry often focuses on PUE (for example from 1.4 to 1.15) and electricity savings. From an energy-asset view that is only half the story: whether high-grade waste heat can be monetised is an equally important value line.
Almost all electricity used by server chips becomes Joule heat. Under conventional air cooling, exhaust air is only about 35 °C — a low-grade source that is hard to use at scale. With cold-plate or immersion liquid cooling on high-density racks, secondary-loop return water can stay around 45–60 °C. The data center then shifts from a pure energy sink to a steady heat source.
A 1MW baseline ledger: a winter monetisation model you can check
Use a “1MW IT load” winter model for cross-project comparison (a calculation framework only; real projects must follow site heat balance and local heat prices):
- Heat output: For 1MW of IT load, liquid cooling typically removes about 0.7MW of heat (the rest goes to air cooling and other paths) — roughly a continuous 700kW heat source.
- Heating-season hours: Taking a northern heating season of about five months (~3600 hours), 0.7MW × 3600h = 2520 MWh (about 2.52 million kWh of heat).
- Commercial valuation: Using a common northern distr