Data Center Water Usage: WUE Benchmarks and Cooling Trade-offs

Data center water use is measured as Water Usage Effectiveness (WUE): liters of water consumed per kilowatt-hour of IT energy. Evaporative-cooled facilities average 1.8-1.9 L/kWh, per U.S. Department of Energy figures, while dry, air-cooled sites run near 0 L/kWh and best-in-class hyperscale fleets — Microsoft at 0.27 L/kWh, AWS at a reported global average of 0.19 L/kWh — sit an order of magnitude below the evaporative baseline. The gap between these numbers is almost entirely a cooling-technology choice, not an unavoidable cost of computing.

Key takeaways

• Industry average WUE: 1.8-1.9 L/kWh for evaporative-cooled data centers (U.S. DOE / Data Center Knowledge). Air-cooled chiller plants run 0-0.3 L/kWh with no direct evaporative loss.
• Hyperscalers beat the average by 6-9x on WUE: Microsoft reported 0.27 L/kWh fleet-wide in 2025; AWS reports a global average of 0.19 L/kWh.
• Absolute consumption is still rising with AI load. Google’s disclosed data center water consumption grew from 4.3 billion gallons (2021) to 7.7 billion gallons (2024) as AI-driven electricity demand rose 27% year over year.
• Indirect water — from power generation — typically exceeds on-site cooling water. UC Riverside/UT Arlington research puts grid-electricity water intensity at roughly 10x the direct, on-site cooling draw for a typical US data center.
• Closed-loop liquid cooling removes most direct water use. Microsoft quantifies savings above 33 million gallons per year per converted facility; the trade-off is the back-end heat-rejection design (dry cooler vs. cooling tower) still matters for total site water.
• Regulatory and community pushback is now a siting variable, not a footnote. Active moratoriums or building pauses have hit Georgia counties, the Netherlands, and other jurisdictions through 2026, frequently citing water alongside power and land.
• Site selection should model water risk explicitly. Water-stressed basins (parts of the US Southwest and Southeast, Aragón in Spain, several fast-growing Latin American cities) now carry real permitting and reputational risk regardless of the WUE number an operator can hit on paper.

For live facility data including cooling type disclosures, see our data center catalog and market statistics.

How WUE is measured and what it doesn’t capture

The Green Grid defines WUE the same way it defines PUE (Power Usage Effectiveness): a ratio, calculated as total site water consumption in liters divided by total IT equipment energy in kWh, over an identical measurement period. It captures three main direct water draws:

• Cooling tower makeup water — replacing water lost to evaporation and blowdown in evaporative and hybrid systems.
• Humidification — maintaining target relative humidity in the white space, a smaller but non-zero draw in dry climates.
• Water treatment losses — backwash and reject water from on-site treatment systems feeding the cooling loop.

Site WUE (sometimes called WUE-site) excludes source energy WUE, which adds the water consumed generating the facility’s electricity at the power plant. That indirect figure depends entirely on the local grid’s generation mix — thermoelectric plants (coal, gas, nuclear) are heavy water consumers for their own cooling, while wind, solar PV, and hydro-light grids carry a much smaller water footprint per kWh. UC Riverside and UT Arlington researchers estimated the indirect component at roughly 10x the direct, on-site figure for a typical US data center drawing from a thermoelectric-heavy grid — meaning a facility with an excellent 0.2 L/kWh site WUE can still have a source-energy WUE above 2 L/kWh once grid water is counted.

WUE benchmarks by cooling technology

Cooling technology	Typical WUE (L/kWh, site only)	Water mechanism
Open-loop evaporative cooling towers	1.8-1.9 (industry average)	Continuous evaporation + blowdown for heat rejection
Hybrid (economizer + evaporative assist)	0.5-1.2	Evaporative cooling used only above an outdoor-temperature threshold
Air-cooled chiller plant (dry cooler, no evaporation)	0-0.3	No direct evaporative loss; higher electricity use for compressor-driven cooling
Direct-to-chip liquid cooling (closed loop)	Near 0 at the chip loop	Recirculating coolant; water use shifts to the facility-level heat rejection method
Best-in-class hyperscale fleet average	0.19-0.30	Mix of air-cooled and optimized evaporative sites, reported by AWS and Microsoft

Figures compiled from U.S. Department of Energy federal data center guidance and hyperscaler sustainability disclosures; ranges vary by climate and specific plant design.

The trade-off that this table doesn’t show directly: lower WUE frequently means higher PUE, because dry, air-cooled heat rejection uses more electricity (running compressors and larger fan arrays) than evaporative cooling, which uses the latent heat of vaporization to reject heat cheaply. Operators in water-stressed but power-abundant regions increasingly accept a PUE penalty of a few tenths of a point to avoid evaporative water draw; the reverse trade is common in power-constrained but water-abundant markets. See our power density guide for how rising rack densities are pushing this calculus toward liquid cooling regardless of the water angle.

Disclosed hyperscaler water consumption

Company	Data center water consumption	Year / source
Google	7.7 billion gallons	2024, Google Environmental Report 2025
Google	4.3 billion gallons	2021, same report series
Microsoft	~1.7 billion gallons (estimated from disclosed WUE and energy use)	2023 disclosure basis
Meta	~870 million gallons	2023 disclosure basis
Microsoft fleet-wide WUE	0.27 L/kWh	2025, Microsoft Environmental Sustainability Report
AWS global average WUE	0.19 L/kWh	Company disclosure

Google is the largest disclosed water consumer among hyperscalers, by more than 4x versus the next reported figure, reflecting both its data center footprint and a lower share of dry-cooled sites relative to Microsoft’s newer fleet. All three companies have public water targets: Microsoft aims for zero-water evaporation on new AI-optimized designs with rollouts from 2027, Google targets 120% replenishment of freshwater consumption by 2030, and AWS has committed to be “water positive” by 2030.

Indirect water: the grid electricity component

Every kilowatt-hour a data center draws from the grid carries an embedded water cost at the power plant generating it — thermoelectric plants withdraw and consume water for their own steam-cycle cooling. This is why a facility’s total water footprint can look very different depending on where it sits:

• Grid dominated by natural gas or coal steam plants: high indirect water intensity, often several liters per kWh at the generation source.
• Grid dominated by wind, solar PV, or hydro: indirect water intensity drops sharply — solar PV and wind have minimal operational water needs compared to thermoelectric generation.
• Nuclear-heavy grids: water-intensive per kWh at the plant (large cooling water withdrawals), though most of that water is returned to its source rather than consumed. See our SMR and nuclear power guide for how new nuclear procurement for AI data centers intersects with this.

This is the practical reason renewable power procurement and water-efficient design are frequently bundled in hyperscaler sustainability plans: cutting a facility’s electricity intensity or shifting its power mix reduces indirect water consumption even without touching the cooling plant.

Regulatory and community pressure by region

Water has become a formal siting constraint, not just an environmental talking point, in several markets as of 2026:

Region	Status	Driver
Georgia (US), several counties	Building moratoriums, including Augusta’s extended pause	Reports of dry residential wells near existing campuses; water-deficit projections by 2030
Netherlands (national)	Repeated pauses on new hyperscale approvals since 2022	Land, grid, and water constraints combined; local election results reinforced the pause
Aragón, Spain	Active community campaigns against new approvals	Drought exposure in the Ebro basin; “Tu Nube Seca Mi Río” campaign
Uruguay	Google campus proceeded only after dropping water-based cooling	2022 court disclosure showed proposed 2 million gallons/day draw from a strained supply
US, 20+ states	Moratorium or restriction bills under active discussion in 2026	Combined water, power, and property-tax concerns; more than $130 billion in projects reportedly delayed or abandoned in Q1 2026 alone

The common pattern: disclosure gaps — water use hidden behind NDAs between developers and local utilities — have been a bigger flashpoint than the raw consumption numbers themselves. Operators publishing facility-level WUE and sourcing plans upfront have generally faced less local opposition than those relying on confidential agreements later reported by local media or freedom-of-information requests.

Southeast Asia and emerging-market water context

Coloprice’s core coverage region — Southeast Asia, plus the CIS and neighboring markets from Q3 2026 — has its own water dynamics distinct from the US and EU cases above:

• Singapore ended its 2019-2022 building moratorium with sustainability conditions (minimum PUE thresholds under the DC-CFA scheme) rather than a water-specific cap, but the underlying land and cooling-capacity scarcity pushes new approvals toward higher-efficiency, often less water-intensive designs. See our Singapore market guide.
• Johor and Bangkok have materially more available land and water than Singapore, which is part of their appeal for hyperscale campuses, but both sit in monsoon-dependent river basins where dry-season stress is a known planning risk. Coverage in our Johor guide and Bangkok vs. Johor vs. Singapore comparison tracks capacity growth in both markets.
• New capacity in water-stressed basins elsewhere (parts of India, for instance) faces the same disclosure and community-relations dynamics seen in Georgia and Uruguay; see our India market guide.

What this means for buyers and site selectors

1. Ask for site-specific WUE, not corporate averages. A hyperscaler’s fleet-wide 0.2-0.3 L/kWh figure can mask individual facilities still running open-loop evaporative towers at 1.8+ L/kWh; request the number for the specific site under consideration.
2. Model total water risk, not just WUE. A low-WUE, air-cooled facility on a water-stressed but coal-heavy grid can carry a larger total (direct + indirect) water footprint than a moderate-WUE evaporative site on a hydro- or renewables-heavy grid.
3. Treat water disclosure as a due-diligence item alongside power. Our due diligence checklist covers power and cooling audits; add a water-sourcing and permitting review for any greenfield or expansion site in a contested region.
4. Expect cooling-technology mix to keep shifting toward closed-loop liquid systems as rack densities rise for AI workloads — a trend that reduces water risk as a side effect of solving the power-density problem, not as its primary goal.
5. Track local moratorium and permitting activity before committing to a region. Georgia, the Netherlands, and parts of Spain show that water-driven opposition can add months to years of permitting delay even where power and land are otherwise available — check /data-centers/ for operator and market context before shortlisting a site.