Quick Answer: Data centres use water primarily for cooling. As servers convert electricity into computation, they generate enormous amounts of heat that must be continuously removed to prevent hardware failure. Most facilities do this through water-based systems: evaporative cooling towers, chilled water loops, or direct liquid cooling applied to individual chips. Beyond what happens on-site, data centres carry an indirect water footprint through the electricity they consume, since the power plants generating that electricity require water of their own. Both sides of that equation matter when evaluating a facility's true resource profile.
The average person doesn't think much about water when they open a laptop or run a cloud application. Behind that request, a chain of physical systems is doing something very tangible: converting electricity into heat, then moving that heat somewhere else before the hardware fails. Water is the medium most facilities rely on to make that happen at scale.
Water consumption varies significantly depending on facility size, cooling design, and geography. A large data centre can consume up to 5 million gallons of water per day, though facilities running closed-loop or free-cooling systems in cold climates operate with a fraction of that.
The range matters, because it means the cooling infrastructure and location of a facility are not incidental details. For IT leaders evaluating colocation partners or planning new infrastructure, knowing how and why data centres use water gives you a more complete lens for assessing what you're actually buying.
The relationship between computation and heat is a fundamental law of physics. Every watt that flows into a server's processors is converted into computation, and the remainder becomes waste heat. That heat doesn't disappear; it accumulates in the rack, then in the room, and if it isn't continuously removed, the hardware throttles, fails, and shuts down.
Temperature management is therefore not an optional feature of data centre operations. It is the foundational constraint around which everything else is designed.
Air was the original solution to this problem. Cold air is pushed through server racks to carry heat away, and warm air is returned to be cooled again. For lower-density workloads, this worked reasonably well.
But as rack densities climbed and heat output grew, air began to reach the limits of its thermal capacity. Water conducts heat approximately 24 times more effectively than air, which is why water-based cooling became the dominant approach in modern facilities. For high-density compute, water is not a preference but a physical requirement.
When most articles discuss how much water a data centre uses, they focus only on what happens inside the building. That is the direct water footprint: the water moved through cooling towers, chilled water systems, and liquid cooling loops on-site. It is measurable, visible, and increasingly reported by operators.
What is far less discussed is the indirect water footprint: the water consumed upstream at the power plants generating the electricity that runs the facility. Fossil-fuel and nuclear power plants require significant volumes of water for steam generation and cooling.
Research published in IEEE Spectrum found that indirect water use through electricity generation often accounts for 80% or more of a data centre's total water footprint. In one analysis, a single GPT-3 response of 150 to 300 words consumed 16.9 millilitres of water total: 2.2 ml from on-site cooling, and 14.7 ml from electricity generation.
Different cooling architectures consume water in very different ways. Data centre cooling is not a single system but a layered one, operating at both the server level and the building level. The approach chosen at each layer determines how much water the facility consumes, how much energy it draws, and how it performs under high-density workloads.
Evaporative cooling towers are the most widely deployed water-based cooling mechanism in large data centres.
Hot water from the facility's heat exchangers is circulated through the tower, where a portion of it evaporates into the atmosphere. As it evaporates, it carries heat with it, cooling the remaining water so it can be recirculated. The process is thermodynamically efficient, which is why it became the default for so many large-scale facilities.
The mechanism that makes evaporative cooling work is also what determines its water demand. Between 70% and 85% of the water used in the process evaporates, which is exactly how the heat gets rejected: phase-change evaporation carries the thermal load away.
The remainder is discharged periodically as blowdown, a controlled process that removes the mineral salts and dissolved solids that accumulate as water concentrates through repeated cycles. Fresh water must be added continuously to replace both the evaporated volume and the blowdown, which is why evaporative systems have an ongoing makeup water requirement rather than a fixed one-time fill.
Chilled water systems circulate cooled water through distribution pipes to cooling coils within the facility, where it absorbs heat from server rooms before being returned to a central chiller plant to be cooled again. In a fully closed-loop configuration, the same water is continuously recirculated with minimal evaporative loss.
These systems consume significantly less water than open evaporative towers but require more mechanical energy to operate the chillers maintaining the loop temperature.
Closed-loop chilled water systems represent the primary way data centres genuinely recycle water rather than consume it. The water in a closed loop is treated, filtered, and reused rather than expelled or evaporated. This makes them well-suited to water-constrained environments, though the trade-off is greater electricity consumption compared to evaporative approaches at equivalent load.
Free cooling uses ambient outdoor air or cold-water sources to cool the facility without engaging mechanical chiller systems. In cooler climates, outdoor air temperatures fall below the facility's target temperature for significant portions of the year. When that happens, operators route outside air through the system or use water cooled naturally by ambient conditions to reject heat from the building, bypassing chillers entirely.
During free cooling periods, evaporative water loss drops sharply. The facility is borrowing cold from the outside environment rather than manufacturing it with electricity and water. Free cooling does not eliminate water use entirely, but it substantially reduces direct consumption during the hours and months when outside conditions allow. How often a facility can access this mode is almost entirely a function of where it is built.
Direct liquid cooling (DLC) takes coolant fluid much closer to the source of heat: directly to the chip. In cold-plate configurations, liquid-cooled plates are mounted onto CPUs and GPUs, and the fluid absorbs heat at the component level rather than waiting for it to disperse into room air. In immersion cooling, servers are submerged entirely in a dielectric, non-conductive fluid engineered for heat transfer.
These techniques are increasingly relevant as AI workloads push rack densities beyond what air or conventional chilled water can handle efficiently. High-density GPU racks can generate 50 kW or more per rack, far exceeding the thermal ceiling of air-based cooling.
Direct liquid cooling systems are typically closed-loop, meaning they consume significantly less water than evaporative approaches. The trade-off is capital cost and the infrastructure complexity of delivering liquid directly into the rack.
|
Cooling Method |
Water Consumption |
Energy Use |
Best Suited For |
|
Evaporative Cooling Towers |
High (70–85% evaporated by design) |
Lower |
Traditional, lower-density workloads |
|
Closed-Loop Chilled Water |
Low (water recirculated) |
Moderate to High |
General purpose, water-constrained sites |
|
Free Cooling (Air/Water Economisation) |
Very Low |
Very Low |
Cold-climate facilities |
|
Direct Liquid / Immersion Cooling |
Very Low (closed loop) |
Lower at high density |
AI, GPU-intensive, high-density compute |
WUE, or Water Usage Effectiveness, is the industry standard for measuring how efficiently a data centre uses water. It was developed by The Green Grid, a non-profit industry consortium, in 2011 as a counterpart to PUE.
The formula divides a facility's annual water consumption in litres by the total energy consumed by IT equipment in kilowatt-hours. A lower WUE means less water consumed per unit of IT output.
A WUE of zero is theoretically achievable only in a fully air-cooled facility with no humidification systems. In practice, the industry average WUE sits at approximately 1.8 to 1.9 litres per kilowatt-hour.
Facilities reporting WUE well below 1.0 are typically running cold-climate or heavily free-cooled infrastructure. For perspective, a WUE of 0.2 L/kWh means the facility uses less than a cup of water for every kilowatt-hour of IT load.
WUE = Annual Site Water Usage (litres) ÷ IT Equipment Energy Usage (kWh)
A facility consuming 50 million litres of water annually to support IT equipment drawing 40,000 MWh would carry a WUE of 1.25 L/kWh. A cold-climate facility making heavy use of free cooling might achieve 0.4 or lower.
Berkeley Lab's 2024 US Data Centre Energy Usage Report projects that average WUE is expected to rise slightly as hyperscale and liquid-cooled facilities proliferate. Higher rack densities bring more energy intensity, and liquid cooling infrastructure introduces new water demands even as it reduces per-unit heat output compared to air.
The water-energy tradeoff in data centre cooling is one of the most consistently underexplained dynamics in the space. Evaporative cooling is energy-efficient because phase-change evaporation rejects heat without requiring large mechanical input, but it draws more directly from local water supplies as part of that process.
Air cooling reduces that direct water draw significantly, but it requires substantially more electricity to achieve equivalent heat rejection. The two approaches sit at opposite ends of a tradeoff, and facility design determines where on that spectrum a given site lands.
What changes the calculation significantly is the energy source. A facility powered primarily by hydroelectricity faces a fundamentally different indirect water equation than one drawing from coal or natural gas.
Coal power plants consume approximately 19,185 gallons of water per megawatt-hour of generation; natural gas plants around 2,800 gallons per MWh. Hydroelectricity requires a fraction of either. This means the choice between water-intensive cooling and energy-intensive cooling is not equally costly everywhere. The grid mix behind the facility changes the actual water impact of choosing more electricity over more evaporation.
Qu Data Centres operates exclusively within Canadian markets, where colocation infrastructure is backed by clean hydro grids across several provinces, and where cold ambient temperatures allow free cooling for extended portions of the year.
That combination means Qu's facilities can pursue lower direct water consumption without the indirect water penalty that follows in fossil-fuel-heavy markets. For Canadian enterprise buyers, the environmental and operational profile of colocation matters. Choosing a facility that operates in this structural advantage is meaningfully different from choosing one that does not.
Book a tour of our facilities today and see how Qu Data Centres prioritizes responsible usage of natural resources with our facility design.
For most of the history of enterprise computing, standard air cooling could manage the heat densities in typical server racks. A conventional enterprise rack running 5 to 10 kW of IT load is manageable with chilled air. A GPU cluster built for AI training can generate 50 to 100 kW per rack and beyond. Air simply cannot move enough heat fast enough to keep those chips within safe operating temperature.
The shift to direct liquid cooling infrastructure is a direct response to this, and it comes with water and capital implications that distinguish AI-ready facilities from legacy builds:
Microsoft reported a 34% increase in global water consumption during the early stages of its AI infrastructure buildout, which prompted the company to accelerate investment in liquid cooling and closed-loop systems. The broader industry response has followed the same logic: the cooling challenge created by AI is real, and facilities built to handle it properly are investing in designs that manage both the heat density and the water profile together.
The single greatest lever on a data centre's direct water footprint is climate. Facilities in hot, arid regions must run evaporative cooling continuously through warm months and have no natural alternative to water-based heat rejection when outside temperatures are high.
Canadian facilities operate in fundamentally different conditions. Cold ambient temperatures for a significant portion of the year allow free cooling systems to carry the thermal load without evaporation, and without drawing heavily from municipal water supplies.
Research from Vaisala comparing cold-climate and hot-climate facilities found that a 10 MW data centre in a hot country may consume tens of millions of litres of water per year, while an equivalent facility in a cold climate would use just 10 to 20 cubic metres annually.
The difference is not marginal. Canada's climate does not merely reduce operating costs; it changes the resource profile of the infrastructure at a structural level.
The second advantage is the grid. The indirect water footprint of a data centre depends almost entirely on how the electricity supplying it is generated. Canadian provinces with hydroelectric-dominant grids allow facilities to run more energy-intensive cooling approaches without the indirect water penalty that follows in fossil-fuel-heavy markets.
The structural advantages Canadian data centres hold include:
As Water Canada noted in a recent interview with infrastructure experts, Canada's cold climate enables free-air cooling for a majority of the year in many markets, cutting evaporative water loss to a fraction of what facilities in warmer climates require. That advantage compounds with Canada's clean grid to change both sides of the water-energy tradeoff simultaneously.
Qu Data Centres operates nine carrier-neutral facilities across five Canadian markets: Calgary, Edmonton, Ottawa, Toronto, and London, Ontario. These facilities are purpose-built for enterprise and cloud workloads, which means the cooling infrastructure reflects the environmental realities of Canadian markets rather than designs borrowed from warmer climates.
Qu's TOR3 facility in the Greater Toronto Area, for example, runs a 1,400-plus-ton Liebert DSE cooling system with EconoPhase Free-Air Cooling, specifically designed to draw on ambient outdoor temperatures to reduce mechanical chilling and limit water draw during cold-weather periods.
Across Qu's national footprint, N+1 and N+2 redundant cooling configurations maintain system resilience without requiring overcooled environments. Several Qu facilities hold Uptime Institute Tier III certification, including TOR3, OTT3, EDM2, and CGY3, confirming that cooling systems meet the redundancy requirements for continuous availability without scheduled downtime. When you're choosing a colocation partner in Canada, you're choosing the full resource profile that backs the rack, including how its cooling is designed and where it sits on the water-energy tradeoff.
Book a facility tour to see the infrastructure firsthand, or speak with a solutions architect about matching your workload requirements to the right facility and market.
Data centres need water because servers generate significant waste heat, and water removes large volumes of heat far more efficiently than air. Every watt consumed by IT equipment becomes a watt of waste heat that must be continuously removed to keep chips within safe operating temperature. At enterprise and hyperscale scale, water-based cooling is the most practical way to achieve this without compromising performance or hardware reliability.
It depends heavily on the facility's cooling design and location. Evaporative cooling systems used by large facilities in hot climates draw more water by design, since evaporation is the mechanism that makes the cooling work. Facilities running closed-loop chilled water systems, free cooling in cold climates, or direct liquid cooling can operate with significantly lower water consumption. Design choices and geography are the dominant variables, not facility size alone.
Some do and some do not. Closed-loop chilled water systems circulate the same treated water repeatedly, replacing only minor losses from maintenance cycles. Evaporative cooling towers operate differently: the evaporation process that removes heat also removes water from the loop, so fresh makeup water is added continuously to maintain the system. The cooling architecture a facility uses is the primary determinant of how much water is recirculated versus replenished.
Seawater contains dissolved salts, minerals, and biological matter that accelerate corrosion in pipes, heat exchangers, and cooling equipment. Some large coastal facilities use seawater in specific heat exchange configurations with extensive treatment and filtration infrastructure, but this requires purpose-built engineering and significant capital investment. The operational overhead makes it impractical for most inland and urban facilities.
WUE (Water Usage Effectiveness) measures a data centre's annual water consumption in litres divided by IT equipment energy in kilowatt-hours. It was developed by The Green Grid as a companion metric to PUE. A lower WUE means less water consumed per unit of compute output. The industry average is around 1.8 to 1.9 L/kWh. For buyers evaluating colocation providers, WUE is most useful when read alongside PUE, since cooling designs that improve one metric often worsen the other.
Yes. Cold ambient temperatures give Canadian facilities access to free cooling for extended periods annually, sharply reducing or eliminating evaporative water loss during those periods. Hydroelectric grid dominance in several provinces also lowers the indirect water footprint of electricity generation compared to fossil-fuel-heavy grids. Both advantages operate simultaneously, giving Canadian facilities a structural position that facilities in hotter, carbon-intensive markets cannot easily replicate.