WUE Averages Run Fivefold Apart Until You Define the Water

Sunbird’s DCIM glossary puts the average data center WUE at 1.8 L/kWh. Lawrence Berkeley National Laboratory’s latest U.S. data center energy report puts the national average site WUE just over 0.36 L/kWh, roughly one fifth of that. The Green Grid white paper that introduced water usage effectiveness also prints 1.8 L/kWh, but as the U.S. aggregate water intensity of thermoelectric power generation, which describes power plants rather than data centers. Before any cooling method can be ranked on water, the two documents that define the metric have to be read side by side, because they do not count the same water.

The Green Grid and ISO/IEC 30134-9 put different water over the same kilowatt-hour

The Green Grid’s White Paper #35, published in 2011, divides annual site water usage by IT equipment energy and tells operators to reuse the IT energy value already measured for PUE. The water it counts includes humidification plus cooling tower evaporation, blowdown and drift. WUE therefore inherits the same boundary that PUE draws around the building, with water in the numerator instead of energy.

ISO/IEC 30134-9:2022 keeps the IT energy denominator but defines annual water usage as total water input minus total water returned out of the data center boundary. In the formula printed in the standard’s public preview, blowdown piped off site leaves through that returned-water term. White Paper #35 counts the same blowdown as used.

The ISO text also states water in m³ and energy in MWh. One m³/MWh is 1,000 L divided by 1,000 kWh, which equals 1 L/kWh, so values carry between the two conventions unchanged.

A third reading circulates in secondary sources. The Environmental and Energy Study Institute’s 2025 explainer describes WUE as water divided by total facility energy and cites an average of 1.9 L/kWh. On a total-energy denominator the reported value drops by a factor equal to the facility’s PUE, so one plant can publish two numbers.

A data center water figure that does not state its numerator and its denominator cannot be compared with another one.

A cooling tower caps evaporation at 1.47 L per kWh of heat

Physics sets how much water evaporation needs per unit of heat. NIST’s Chemistry WebBook lists CODATA enthalpies of formation at 298.15 K of −285.830 kJ/mol for liquid water and −241.826 kJ/mol for water vapor.

The difference, 44.004 kJ/mol, is the heat of vaporization. Divided by the molar mass NIST gives, 18.0153 g/mol, it comes to 2,443 kJ per kilogram evaporated.

One kilowatt-hour is 3,600 kJ. Evaporation per kWh of heat rejected is 3,600 kJ ÷ 2,443 kJ/kg = 1.47 kg, which is 1.47 L taking one kilogram of water as one liter.

That result is an upper bound on evaporation, since it assumes every joule leaves as vapor while a real tower also passes some heat to the air as sensible heat. It is not a cap on site WUE, because a chiller plant’s tower also rejects compressor, pump and fan heat on top of the IT load.

Blowdown comes on top. The Federal Energy Management Program’s cooling tower fact sheet gives blowdown = evaporation ÷ (cycles of concentration − 1).

Assuming three cycles, blowdown is half the evaporation, 0.74 L, and makeup water is 2.21 L per kWh of heat.

White Paper #35 counts all of that makeup as site water. The ISO formula, with the blowdown discharged across the boundary, keeps only the evaporated 1.47 L.

The ratio between the two readings is cycles ÷ (cycles − 1), which comes to 1.5 at three cycles. The same tower, on the same day, reports two site WUE values before any difference in cooling hardware enters.

Lawrence Berkeley National Laboratory’s simulations put water-cooled chillers without economizers at the top of the site WUE range, driven by cooling tower water. Its liquid cooling case, meaning coolant carried to the IT equipment instead of cooled air, pairs a waterside economizer with a water-cooled chiller, and the report says that design still consumes substantial water like any evaporative system. Warmer facility water lowers WUE by cutting chiller hours and adiabatic assist, not by changing what a tower does with each kilowatt-hour of heat.

That makes liquid cooling a water decision only through the heat rejection plant it enables. A loop that follows the cold plate path to facility water and ends at a cooling tower evaporates the same water per kWh of heat that reaches the tower; the savings come from the hours it lets a dry cooler or an economizer carry the load instead. The same report projects the U.S. average site WUE rising to between 0.45 and 0.48 L/kWh by 2028 and names the water consumption of liquid-cooled systems among the drivers.

Against the evaporation ceiling, the national average of just over 0.36 L/kWh is about a quarter. That ratio only works if much of the modeled IT heat leaves through economizers, dry coolers or waterless chillers for most of the year.

Table 1. Published and calculated WUE-type figures and what each countsL/kWh

* Upper bounds per kWh of heat reaching the tower, not per kWh of IT energy.

** Power-generation figure, not a data center WUE; shown for the shared value only.

Source: The Green Grid, “WP#35 Water Usage Effectiveness (WUE)” (2011); LBNL, “2024 United States Data Center Energy Usage Report” (Dec 2024); EESI, “Data Centers and Water Consumption” (Jun 2025); Sunbird DCIM glossary (undated); NIST Chemistry WebBook; DOE FEMP cooling tower fact sheet (Feb 2011)

Neither national average rests on metered site water

LBNL’s averages are outputs of thermodynamic simulations run against climate data from 965 weather stations and adjusted toward industry-reported values. The report assumes systems run as designed and says actual PUEs and WUEs will likely be worse. The glossary’s 1.8 L/kWh carries no source on its page.

The ISO preview ends before the standard’s water categories and its annex on energy water intensity, so how the full text treats blowdown quality and reused water is not confirmed here. The EWIF values used below come from White Paper #35, which drew them from a 2003 NREL report. None of these metrics weighs local water stress, which the ISO preview assigns to a separate reporting category along with land consumption.

The site boundary moves water to the power plant instead of removing it

White Paper #35 adds a second metric for decisions that reach past the fence: WUEsource = (annual site water ÷ IT energy) + EWIF × PUE, where EWIF is the energy water intensity factor, the water consumed per kWh generated. Its suggested U.S. aggregate EWIF is the same 1.8 L/kWh that glossaries quote as an average WUE.

Take two facilities on one grid. Plant A rejects heat by evaporation at site WUE w; plant B uses no site water and runs a higher PUE. Their WUEsource values match when EWIF × (PUE of B − PUE of A) = w, so the PUE penalty plant B can carry before it consumes more total water is w ÷ EWIF.

At the evaporation ceiling, w is 1.47 L/kWh, and on the U.S. aggregate grid the break-even is 1.47 ÷ 1.8 = 0.82.

The white paper’s state table moves that line a long way. Illinois, listed at 3.97 L/kWh, brings it down to 0.37.

California, listed at 0.19 L/kWh, pushes it to 7.7, a penalty no real plant approaches.

Real towers run well under the ceiling. At LBNL’s national average site WUE, the break-even on the aggregate grid is 0.36 ÷ 1.8 = 0.20.

On an Illinois-type grid, the same site WUE gives 0.36 ÷ 3.97 = 0.09.

These are break-even values, not typical ones, and they hold only while plant B’s site water stays at zero. A waterless design that costs a tenth of a PUE point is a clear total-water gain on an average U.S. grid and close to a wash on a grid as water-intensive as the one the white paper lists for Illinois, while site WUE alone scores it as a win in both places. Which plant suits a given site is a design decision that needs qualified engineering review and current local grid data, since the published EWIF tables are two decades old.

WUE is one ratio with more than one numerator and two denominators in circulation, so the glossary average and LBNL’s modeled average are not two measurements of the same thing. A cooling tower cannot evaporate more than 1.47 L per kWh of heat, blowdown at three cycles adds half again under The Green Grid’s count and nothing under the ISO formula, and the modeled national average sits near a quarter of that evaporative ceiling. Once power-plant water enters through WUEsource, a waterless plant’s advantage depends on the grid’s water intensity and the PUE it costs, and on a water-intensive grid a penalty under a tenth of a PUE point can erase it.

References

• WP#35 Water Usage Effectiveness (WUE): A Green Grid Data Center Sustainability Metric / The Green Grid / 2011 / https://www.thegreengrid.org/system/files/store/WUE_v1.pdf
• ISO/IEC 30134-9:2022, Part 9: Water usage effectiveness (WUE), public preview / ISO/IEC, preview distributed by iTeh Standards / March 2022 / https://cdn.standards.iteh.ai/samples/77692/4dbd2731397e4663bc8b05ce14c05814/ISO-IEC-30134-9-2022.pdf
• 2024 United States Data Center Energy Usage Report (LBNL-2001637) / Lawrence Berkeley National Laboratory / December 2024 / https://eta-publications.lbl.gov/sites/default/files/2024-12/lbnl-2024-united-states-data-center-energy-usage-report_1.pdf
• Cooling Towers: Understanding Key Components of Cooling Towers and How to Improve Water Efficiency / U.S. Department of Energy, Federal Energy Management Program / February 2011 / https://www.energy.gov/sites/prod/files/2013/10/f3/waterfs_coolingtowers.pdf
• Water, NIST Chemistry WebBook, SRD 69 / National Institute of Standards and Technology / undated / https://webbook.nist.gov/cgi/cbook.cgi?ID=C7732185&Mask=1F
• Data Centers and Water Consumption / Environmental and Energy Study Institute / June 25, 2025 / https://www.eesi.org/articles/view/data-centers-and-water-consumption
• What Is Water Usage Effectiveness (WUE)? Data Center Glossary / Sunbird DCIM / undated / https://www.sunbirddcim.com/glossary/water-usage-effectiveness-wue