Modern data-centres — the backbone of cloud computing and AI — consume extraordinary volumes of water to stay cool. Yet the same heat they expel could warm homes or even generate electricity. By replacing evaporative cooling systems with sealed refrigerant and heat-pump networks, operators could eliminate most water use, cut carbon emissions, and turn an environmental liability into a community asset.
As large data-centres hum, towers in the buildings quietly steam. Those plumes aren’t smoke — they’re water vapour. Each wisp represents a few more litres lost to the air as the servers inside churn through trillions of calculations. As artificial intelligence and cloud computing surge, that vapour trail is becoming one of the tech industry’s least discussed environmental costs.
The figures are startling. A single 100-megawatt hyperscale facility can consume around 2.5 billion litres of water per year — roughly the annual usage of 25,000 people. In many locations, that water is potable, drawn from the same municipal supplies serving nearby homes and farms. Globally, data-centres now account for a measurable fraction of industrial water withdrawals, and most operators still rely on evaporative cooling — a technology that literally works by throwing water away.
Evaporative systems became standard because they are simple and energy-efficient. Water’s high latent heat of vaporisation makes it superb at carrying heat away. Spray a fine mist over a warm coil and the evaporation absorbs vast amounts of thermal energy, leaving the remaining air cooler. The downside is obvious: the water is gone. Up to 80 per cent of what’s circulated is lost to evaporation, meaning a large facility can effectively vent thousands of litres every minute on a hot day.
That’s where a new generation of engineers and energy planners are asking: why not capture the heat instead of wasting it?
Turning Waste Heat into a Resource
The alternative is to move from evaporating water to circulating refrigerant — in essence, applying the same principles that keep your kitchen fridge cold. Closed refrigerant loops transfer heat from one side of the system to another without consuming water. Add a heat pump, and that low-grade waste heat from servers (typically 25–45 °C) can be “upgraded” to useful temperatures for district heating or industrial use. In some cases, the concentrated heat can even drive small turbines or organic Rankine cycle units to produce electricity.
The potential scale is immense. A 100 MW data-centre, operating continuously, produces around 876,000 MWh of heat per year — enough theoretical energy to meet the space-heating and hot-water needs of roughly 70,000 average homes. Not all of that can be captured efficiently, of course, but even partial recovery could offset vast quantities of gas or electric heating.
Across Europe and parts of Asia, a few pioneers are already trying. In Denmark, data-centre waste heat warms residential blocks in Odense. In Finland, municipal networks feed low-temperature server heat into local grids. In the UK, several councils have begun exploring how to integrate future data-centre developments with planned heat networks — though the practice is far from mainstream.
The Challenge of Temperature
Using heat pumps to raise waste heat to the 60–90 °C needed for traditional district heating costs energy. The greater the temperature “lift”, the lower the efficiency — measured as the coefficient of performance, or COP. Modern industrial heat pumps, however, can achieve COPs of 3–5, meaning they deliver three to five units of useful heat for every unit of electricity consumed. If that electricity is renewable, the overall footprint can still be far lower than gas combustion.
Newer data-centre designs help, too. Direct-to-chip liquid cooling and immersion cooling bring coolant directly into contact with processors, capturing heat at higher temperatures. That makes the heat easier to reuse, while eliminating the fans and chillers of conventional air-cooled halls. The approach also suits refrigerant-based cooling loops, closing the circle between IT and HVAC engineering.
Practical and Policy Hurdles
So why isn’t this happening everywhere? Because retrofitting a giant data-centre is not like replacing a home boiler. Sealed refrigerant systems require high-pressure piping, complex heat exchangers, and robust controls to prevent leaks of greenhouse-gas-intensive refrigerants. The initial capital cost can be steep, and many sites were sited in remote or industrial zones far from any heat customers.
Even so, the case for integration is growing. Water scarcity is expected to affect large parts of southern England by the 2030s, and public pressure on industrial users is rising. Governments could encourage the shift through mandatory water-use reporting, incentives for heat-recovery retrofits, and planning rules that link new data-centre approvals to local heat networks. Some of these measures already appear in policy recommendations from sustainability alliances and trade bodies.
The economics are also improving. Energy prices and carbon accounting mean recovered heat now has tangible value. In some European cities, data-centre operators sell heat to municipal networks, reducing both parties’ costs. As refrigerant technology improves — particularly with low-global-warming-potential fluids — the safety and environmental concerns that once limited large-scale adoption are easing.
What a Redesign Would Entail
Moving to closed refrigerant and heat-pump systems would mean redesigning data-centres from the rack outward:
- Liquid-cooled servers to capture heat at higher, more useful temperatures.
- Refrigerant loops and heat exchangers instead of open evaporative towers.
- Large-scale heat pumps to raise temperature and deliver steady output.
- District heating connections or on-site conversion to electricity via small turbines.
- Monitoring and telemetry systems for both water and heat flows, enabling transparent reporting.
The transformation won’t be cheap — nor quick — but neither was the adoption of renewable electricity a decade ago. Once proven at scale, the economics of avoided water use, potential heat revenue, and reputational value may well outweigh the retrofit costs.
A More Sustainable Data-Centre Model
Ultimately, the shift from “cooling by wasting” to “cooling by capturing” could make data-centres more sustainable and less contested neighbours. Instead of competing with local households for scarce water, they could supply heat to them. Instead of releasing low-grade warmth to the sky, they could feed it back into cities and grids.
The digital economy has always promised efficiency and dematerialisation; yet its physical footprint is enormous. Rethinking the way we cool our computation — and what we do with the heat — may prove one of the simplest, most visible ways to make that promise real.
Further Reading
The Cloud is Drying our Rivers: Water Usage of AI Data Centers, Jennings, EthicalGEO (2025)
A well referenced study of AI’s additional ecological strain on rivers
Addressing water consumption in data centres amid AI expansion, Durie, New Civil Engineer (2025)
A UK-centric look at Water/Power Usage Effectiveness.
Data Centers and Water Consumption, Yañez-Barnuevo, EESI (2025)
From the Environmental and Energy Studies Institute, USA
Water Use in Data Centres and AI: A Review of Industry Practices and Sustainability Challenges, Government Digital Sustainability Alliance (2024)
A comprehensive overview of how AI and data-centre growth are driving water use — and what can be done to reduce it.
The Real Story on AI’s Water Use–and How to Tackle It, Ren and Luers, IEEE (2025)
Voices from California and Microsoft (who do use heat from data-centres to warm homes) go so far as looking at cooling with recycled water.
WaterWise: Co-optimizing Carbon- and Water-Footprint Toward Environmentally Sustainable Cloud Computing, Jia et al., Cornell University (2025)
An academic exploration of how intelligent workload scheduling can balance water consumption and carbon emissions in cloud systems.
Heat Reuse in the Digital Economy, European Heat Pump Association (2024)
Explores how recovered data-centre heat can supply local heating networks and industrial users across Europe.
Tracking Clean Energy Progress 2023, International Energy Agency (2023)
Part of the IEA’s flagship series examining the energy footprint of emerging technologies, including data centres and cloud infrastructure.
Heat Reuse: A Management Primer, Uptime Institute (2023)
A practical guide to heat-recovery projects in data centres, including case studies and the challenges of integrating with district heating systems.
Understanding Data Centre Water Use in England, techUK (2023)
A UK-focused study assessing the scale, efficiency, and future trends of water use in commercial data-centre operations.
International Review of Energy Efficiency in Data Centres, IEA / Energy in Buildings and Communities Programme (2022)
Global benchmarking of energy performance, cooling technologies, and policy frameworks for sustainable data-centre design.
A Water Efficiency Dataset for African Data Centers, Shumba et al., Cornell University (2024)
An emerging dataset mapping direct and indirect water use in African climates — useful for global benchmarking and model validation.
Source
Water Use in Data Centres and AI, UK Government’s Government Digital Sustainability Alliance (GDSA)
Boˆreal 