🏛️ Issues and Advocacy Training
Courses › Data Centers: Types, Purposes and How They Work

Lesson 7 of 9 · about 16 minutes

Efficiency Metrics: PUE, WUE and CUE

You'll be able to

  • Define and calculate PUE, WUE and CUE
  • State typical and best-in-class PUE values with sources
  • Explain what each metric leaves out
  • Check a developer's efficiency claim with simple arithmetic

Three ratios show up in every data center sustainability report and most developer presentations: PUE for power, WUE for water and CUE for carbon. Each is simple arithmetic. Each also measures less than it seems to, and the gap between what a metric says and what people assume it says is where most of the useful questions live.

PUE: power usage effectiveness

Power usage effectiveness divides all the energy a facility uses by the energy that reaches the IT equipment:

PUE = total facility energy / IT equipment energy

A PUE of 1.0 would mean every kilowatt-hour goes to computing and none to cooling, lighting or power losses. Nobody gets there. A PUE of 2.0 means one unit of overhead for every unit of IT. The metric is standardized internationally as ISO/IEC 30134-2, first published in 2016 and revised in 2026. Always use energy over a full year, not a single moment, because cooling needs swing with the seasons.

Worked example: PUE from a utility bill

The invented Ridgeline Data Center reports these yearly figures:

  • Total energy from the utility meter: 87,600 MWh
  • Energy measured at the IT power strips: 70,080 MWh

PUE = 87,600 / 70,080 = 1.25.

Overhead = 87,600 minus 70,080 = 17,520 MWh. If the site cut PUE to 1.15 with the same IT load, total energy would be 70,080 times 1.15 = 80,592 MWh, saving 7,008 MWh a year. At $0.07 per kWh, that is 7,008,000 kWh times $0.07 = about $490,000 a year.

Typical PUE values

Facility or groupPUESource
Industry average, 20072.50Uptime Institute survey history
Industry average, 20141.65Uptime Institute
Industry average, 20251.54Uptime Institute 2025 survey
Google fleet, trailing twelve months to Q2 20261.09Google data center efficiency page
NREL ESIF data center (Golden, Colorado)1.036 annualizedNational Renewable Energy Laboratory

The industry average has barely moved since 2019. Uptime attributes that to older facilities, which are a big share of the stock, and to hot or humid climates where efficient cooling is harder. New hyperscale and AI builds commonly target 1.1 to 1.3. When a developer promises 1.2, that is believable for a new building. When they promise 1.05 in a hot climate, ask how.

Where PUE misleads

  • It says nothing about how much total energy is used. A 1 GW campus at PUE 1.1 uses vastly more electricity than a 5 MW site at 1.6.
  • It rewards adding load. Fixed overhead spread over more IT energy lowers PUE even as total energy rises.
  • It ignores what happens inside the server. Server fans count as IT energy. Moving cooling work from building fans to server fans can make PUE look better without saving anything. Liquid cooling shrinks server fans, which can make PUE look slightly worse even when the whole system saves energy.
  • It ignores water. Evaporative cooling lowers PUE by using water. A site can buy a good PUE with a high WUE.
  • It ignores whether the computing is useful. An idle server at PUE 1.1 is still wasted power.

WUE: water usage effectiveness

WUE = site water use (liters) / IT equipment energy (kWh)

Standardized as ISO/IEC 30134-9:2022, WUE usually counts water consumed on site for cooling and humidification. It does not include water used at power plants, sometimes called source WUE, unless the report says so. Typical values vary enormously with climate and cooling type: closed-loop sites can be near zero, while evaporative sites in hot, dry places can exceed 1 to 2 L/kWh in summer. Microsoft reported a fleet average of 0.30 L/kWh in fiscal 2024, down from 0.49 in 2021. Uptime's 2025 survey found only 47 percent of operators even collect water data, so many facilities cannot report a WUE at all.

Worked example: WUE from a water bill

Ridgeline's water utility bills show 41,000,000 liters for the year. IT energy was 70,080 MWh, which is 70,080,000 kWh.

WUE = 41,000,000 / 70,080,000 = 0.585 L/kWh, about 0.59.

In gallons: 41,000,000 / 3.785 = about 10.8 million gallons a year, or about 29,700 gallons a day on average. Summer days may be three or four times that.

CUE: carbon usage effectiveness

CUE = total CO2 from facility energy (kg) / IT equipment energy (kWh)

Standardized as ISO/IEC 30134-8:2022. If all energy comes from one grid, a shortcut works: CUE = grid emission factor times PUE. The Energy Information Administration put the 2023 U.S. average at about 0.81 pounds of CO2 per kWh of generation, which is about 0.367 kg.

Worked example: CUE on three grids

Ridgeline has a PUE of 1.25. Here is its CUE on three hypothetical grid mixes:

GridEmission factor (kg CO2/kWh)CUE = factor times 1.25Yearly CO2 for 70,080 MWh IT
U.S. average 20230.3670.4670,080,000 times 0.46 = about 32,200 metric tons
Mostly gas-fired (0.96 lb = 0.435 kg)0.4350.54about 38,100 metric tons
Mostly hydro and wind (illustrative 0.05)0.050.06about 4,400 metric tons

Same building, same PUE, nearly ninefold difference in emissions. Where a data center sits on the grid matters more for carbon than how efficient its cooling is.

CUE has its own traps. Companies often report "market-based" emissions that subtract renewable energy certificates or power purchase agreements, which can be far lower than "location-based" emissions from the actual local grid. Both are legitimate accounting methods under the Greenhouse Gas Protocol and they answer different questions. When a report gives one number, ask which method it used.

Common mix-ups

  • Claim: A low PUE means a green data center. What the evidence shows: PUE measures overhead only. Total energy, water use and the carbon intensity of the grid all matter separately.
  • Claim: A company that buys renewable credits for 100 percent of its use runs on clean power every hour. What the evidence shows: Annual matching means total purchases equal total use over a year. At night or on still days the site may draw from fossil plants. Some companies now report hourly matching, which is a stricter test.

Reporting angle

  • Ask for design PUE and WUE, and once the site runs, actual annual figures. Ask whether the WUE counts only cooling water or all site water.
  • Ask whether emissions figures are location-based or market-based.
  • Read the hyperscaler's annual sustainability or environmental report for site or regional figures; some list them by data center location.
  • Check the arithmetic. Developers' slides sometimes list a PUE and a total power figure that do not match their stated IT load.

Research angle

Check yourself

1. A facility uses 130 GWh a year in total, of which 100 GWh goes to IT equipment. What is its PUE?

2. What did Uptime Institute's 2025 survey report as the industry average annual PUE?

3. WUE is measured in what units?

4. A data center buys all its power from a grid averaging 0.81 pounds of CO2 per kWh and has a PUE of 1.4. About what is its CUE in kg CO2 per kWh of IT energy?

5. Which change would lower PUE while doing nothing to reduce total energy use?

6. Which metric would you use to compare cooling water use between two facilities of different size?

Terms in this lesson

PUE
Power usage effectiveness: total facility energy divided by IT equipment energy. 1.0 is the theoretical floor.
WUE
Water usage effectiveness: site water use in liters divided by IT equipment energy in kWh.
CUE
Carbon usage effectiveness: CO2 emissions from the facility's energy divided by IT equipment energy, in kg CO2 per kWh.
Overhead energy
Energy used for cooling, power conversion losses, lighting and everything else that is not the IT equipment.
Trailing twelve months (TTM)
An average over the last 12 months, used to smooth out seasonal swings.

Full glossary

Sources