Skip to content

Drawing No. EH–TH–052 // Data Center Engineering

Data Center Water Usage & WUE Calculator

Reviewed August 2026

Estimate data-center water consumption, cooling-water requirements and WUE from IT load, PUE and heat-rejection architecture. Compare wet towers, hybrid cooling, evaporative assist and dry/liquid-cooled designs; calculate freshwater requirement, evaporation, blowdown, drift, discharge and optional upstream electricity-related water.

Boundary matters: “Closed-loop liquid cooling” describes the internal coolant loop, not necessarily the final heat sink. A closed loop can reject heat through a wet cooling tower or a dry cooler. This tool therefore separates internal cooling architecture from the evaporative heat-rejection share and reports site input, freshwater and consumptive water separately.

What this model calculates

The screening model turns IT capacity and PUE into total facility power and heat rejection. The wet fraction is routed through an evaporative cooling-water balance; dry rejection uses no routine evaporative makeup. Treatment losses, reclaimed-water share and non-cooling site water are then added to close the facility water balance.

01 // IT & PUERated IT × utilization → IT energy → total facility power.
02 // Heat rejectionUseful heat export is removed; remaining heat splits into wet and dry paths.
03 // Water balanceEvaporation + blowdown + drift → tower makeup → raw water.
04 // WUE & boundarySite input, freshwater and consumption are reported separately per IT kWh.

How cooling architecture changes data center water consumption

quick comparison
Cooling architectureTypical site-water tendencyEngineering reason
Wet cooling towerHighHeat rejection intentionally evaporates water; blowdown and drift add makeup demand.
Hybrid / adiabaticMedium / climate dependentDry rejection handles part of the annual load; water is used during hotter or higher-load periods.
Direct liquid + wet towerHighA closed server coolant loop can still send heat to an evaporative final heat sink.
Direct liquid + dry coolerVery lowHeat is rejected sensibly to air, so routine evaporative cooling-water makeup can approach zero.

These are qualitative architecture tendencies, not universal design values. Climate, controls, PUE, water chemistry, economizer operation and the reporting boundary can materially change measured water use.

1. IT load & facility energy

power basis

For a steady facility, nearly all electrical input ultimately becomes heat. PUE therefore provides a convenient screening estimate of total heat that must be rejected, before useful heat recovery.

2. Cooling architecture

wet vs dry heat rejection

For hybrid or climate-dependent systems, the wet-share input can be interpreted as an equivalent annual-average share. For design-point tower sizing, use the actual design wet load instead.

3. Water source & treatment

freshwater, reclaimed water, losses

Closed loop does not define the water footprint

heat must still leave the facility
IT RACKSelectricity → heatCLOSED COOLANT LOOPwater / dielectric / refrigerantWET HEAT REJECTIONDRY HEAT REJECTIONWet tower consumes water by evaporation.Dry cooler rejects heat to air without routineevaporative cooling-water makeup.The same internal liquid loop can feed either.

Reported-WUE shortcut

convert a published WUE into water volume

Reference example: AWS reports a 2025 global data-center WUE of 0.12 L/kWh, using water withdrawn per IT load. Treat this as a published fleet metric, not a universal design value.
Equivalent average flow
Equivalent daily water
Equivalent annual water
Boundary warning
Match definitions before comparing

Background & method

foldable technical notes

WUE boundaries and diagnostic water-intensity ratios

DOE FEMP expresses site-based WUE as annual site water usage divided by annual IT equipment energy, normally in L/kWh. ISO/IEC 30134-9:2022 formalizes WUE as a data-center KPI and defines measurement/reporting categories. The published 2022 ISO edition remains current while a second edition is under development; always check the reporting category and boundary before comparing WUE values.

This tool therefore labels its primary result site-input intensity (WUE-style): all modeled external site water divided by IT energy. It also provides two explicitly non-standard diagnostics: freshwater-input intensity excludes the modeled reclaimed/alternative-water share, while consumptive-loss intensity counts modeled evaporation, drift and the consumptive portion of other site water. These ratios are useful for engineering comparison but should not be presented as standard WUE without matching the applicable measurement boundary.

site-input intensity = external site water input ÷ IT energy
Cooling-tower equations

For the wet heat-rejection path, the model uses a simplified heat balance and dissolved-solids balance consistent with the existing EngineerHub power-plant water model.

circulating flow = Qwet ÷ (ρ · cp · ΔT)
evaporation = f_latent · Qwet ÷ (ρ · hfg)
blowdown = evaporation ÷ (CoC − 1) − drift
tower makeup = evaporation + blowdown + drift
raw feed = tower makeup ÷ pretreatment recovery

Constant screening properties are ρ = 1000 kg/m³, cp = 4.186 kJ/kg·K and hfg ≈ 2400 kJ/kg. They are adequate for preliminary comparisons, not detailed cooling-tower selection.

Why “liquid cooled” can still mean high water use

Direct-to-chip liquid cooling controls how heat leaves the server. The facility must still reject that heat outdoors. If the facility loop rejects heat through an evaporative tower, routine water consumption remains possible; if it rejects heat through air-cooled equipment, routine evaporative cooling water can approach zero. Microsoft’s newer AI data-center designs explicitly pair closed-loop liquid cooling with zero-water-evaporation heat rejection, while other liquid-cooled designs may use towers.

Water stress and WUI

Equal water volumes do not imply equal environmental impact in different locations. The Green Grid released Water Usage Impact (WUI) v1 as a metric that relates data-center water consumption to local water-stress level. This page does not reproduce that scoring method; use The Green Grid’s official WUI calculator for water-stress scoring.

What is not modeled
  • hour-by-hour weather, economizer envelopes or wet-bulb limits;
  • chiller curves, fan/pump power and detailed PUE decomposition;
  • blowdown chemistry limits by source-water constituent;
  • commissioning/fill water, construction water, humidification transients or emergency storage;
  • site permit limits, municipal allocation, drought curtailment or water-right priority;
  • embodied water in hardware manufacture.

Data center water-use FAQ

common questions about WUE, consumption and cooling

How much water does a data center use?

There is no single universal value. Site water use depends on IT load, PUE, climate, cooling architecture, hours of operation, cycles of concentration, water-treatment losses and whether the facility uses reclaimed water. In this calculator, wet evaporative rejection can require substantial makeup water, while fully dry heat rejection can reduce routine cooling-water demand to nearly zero.

What is data center WUE?

Water Usage Effectiveness (WUE) normalizes data-center water use to IT energy, typically in litres per kilowatt-hour. Published WUE values are only comparable when their measurement boundary is comparable: for example, whether the numerator represents site water withdrawal, water consumption, or another reporting category.

How much water does a 100 MW data center use?

It depends strongly on cooling design. In the linked EngineerHub worked example, 100 MW of average IT load at PUE 1.15 uses about 1.71 million m³/year of external site water under the illustrative full-wet-rejection assumptions, about 0.60 million m³/year with a 35% equivalent wet share, and only ancillary site water in the fully dry-rejection scenario. These are transparent screening scenarios, not industry averages.

Does liquid cooling always use less water?

No. Direct-to-chip liquid cooling describes how heat is collected from IT equipment. The heat must still be rejected outdoors. A liquid-cooled data center connected to a wet cooling tower can still consume significant water, while the same internal liquid loop connected to a dry cooler can operate with little or no routine evaporative cooling-water demand.

What is the difference between water withdrawal and water consumption?

Withdrawal is water taken into the facility or system. Consumption is the portion not returned to the original water body or supply on the relevant accounting basis, commonly because it evaporates or is otherwise incorporated or lost. A facility can therefore have a large withdrawal but a smaller consumptive loss, or vice versa depending on its cooling and discharge configuration.

Technical references

metric definitions and current industry context

[1] U.S. DOE FEMP — Cooling Water Efficiency Opportunities for Federal Data Centers. Defines PUE and site-based WUE and describes chiller/condenser-water/cooling-tower heat rejection. DOE FEMP.
[2] ISO/IEC 30134-9:2022 — Data centres KPI, Part 9: Water usage effectiveness. International standard defining WUE measurement, calculation and reporting. ISO.
[3] The Green Grid — Water Usage Impact (WUI) v1. WUI adds water-stress context to data-center water consumption. The Green Grid.
[4] AWS Sustainability. AWS reports global data-center PUE 1.14 and WUE 0.12 L/kWh in 2025; AWS describes its WUE as water withdrawn per kWh of IT load. AWS.
[5] Microsoft — datacenter water intensity and cooling methods, June 2026. Describes cooling towers, hybrid fluid coolers, direct air, air-cooled chillers and closed-loop direct-to-chip designs with zero water evaporation. Microsoft.
[6] Google Data Centers — Operating sustainably. Describes the energy-water tradeoff in data-center cooling and site-specific balancing of water and energy. Google.

Preset wet-share values are illustrative EngineerHub screening assumptions. To make architecture comparisons cleaner, all revised presets use the same neutral PUE of 1.15 and 100% tower-makeup treatment recovery; neither value is asserted as an industry average or vendor design datum. Change PUE, treatment recovery and all water-balance inputs to project-specific values before using the model in engineering or environmental studies.

Related EngineerHub resources

continue the water and heat-rejection analysis