What is PUE and what does it tell you about data centre efficiency?

Power usage effectiveness (PUE) is the ratio of the total energy a data centre uses to the energy used by its IT equipment, measured over a year:

PUE = total facility energy ÷ IT equipment energy

A PUE of 1.0 would mean every kilowatt-hour went to IT equipment. In reality the facility also spends energy on cooling, power conversion and distribution losses, lighting and other support systems. A PUE of 1.5 means that for every 1 kWh delivered to IT, the supporting infrastructure uses a further 0.5 kWh.

The metric is useful, widely understood and now formally standardised, and it is also frequently misused. PUE tells you how efficiently the infrastructure supports the IT load. It does not measure how efficiently the IT itself works, how much useful computing it produces or how sustainable the facility is.

Where the definition comes from

PUE was developed by The Green Grid, an industry consortium, and has since been standardised internationally. The current standard is ISO/IEC 30134-2:2026, the second edition, published in January 2026, which replaced the 2016 edition and its 2018 amendment. ISO describes the revision as adding guidance for mixed-use buildings (a mixed-use variant, mPUE), updating measurement requirements and clarifying how to treat on-site generation and energy that goes unmeasured. Because the standard defines measurement categories, a PUE figure should always come with a statement of how and where it was measured.

The US Department of Energy's 2024 best-practice guide uses the same annual-energy definition and offers orientation figures: an average data centre has a PUE of around 1.6, some very efficient facilities achieve below 1.1, and it cites the Uptime Institute's 2022 survey average of 1.55 for large data centres. These figures give context, and none of them is a target for any particular site, since climate, density, redundancy and load all influence what is achievable.

Measured over a year

PUE should be calculated from annual energy (kWh); instantaneous power gives a misleading picture. Cooling energy rises and falls with the weather, and IT load with demand, so a PUE measured on a cool night at high IT load will look far better than the year-round reality. A design PUE for a facility not yet built is a target, and only a year of metering will show whether it is met.

When a supplier quotes a PUE, then, the follow-up questions write themselves. Is it annual, or a design or instantaneous figure? Where is IT energy measured: at the UPS output, the power distribution units or the IT equipment itself? What sits inside the boundary, such as offices, shared plant or on-site generation? And which measurement category of ISO/IEC 30134-2 does it follow?

The blind spots

PUE ignores the efficiency of the IT. Consolidating underused servers or replacing old hardware reduces total energy but can make PUE worse, because the overhead does not fall in proportion. It cannot see useful output either: a facility running idle servers efficiently can post an excellent figure.

Nor is it a sustainability score. PUE does not reflect the carbon intensity of the electricity, water consumption or heat reuse. The DOE guide places it within a family of related metrics, including energy reuse effectiveness (ERE), water usage effectiveness (WUE) and carbon usage effectiveness (CUE), that cover what PUE leaves out. Finally, PUE is sensitive to load: a new facility at low utilisation often has a poor PUE for no better reason than that its fixed overheads are spread over little IT load.

When PUE gets worse but energy falls

Total energy falls by 250 MWh, about 16%, yet PUE rises from 1.60 to 1.69, because the overhead fell less than the IT load did. Judged on PUE alone, a good decision looks like a bad one. The energy saving also has to be translated separately into cost, using the tariff, and into emissions, using the carbon intensity of the electricity.

Using PUE well

PUE is at its most useful when tracked over time for the same facility, with the same boundary and method, alongside total energy and IT load. Used that way it is good at exposing infrastructure waste such as overcooling, poor airflow management or inefficient UPS operation at low load. It should always travel with other measures: total energy, IT utilisation, water use, carbon intensity and heat reuse. Comparisons between facilities need particular care, because differences in climate, density, redundancy and measurement method make them misleading unless the methods match.

Reporting obligations

In the EU, the Energy Efficiency Directive requires data centres with an installed IT power demand of at least 500 kW to monitor and report their energy performance every year, and Delegated Regulation (EU) 2024/1364 specifies the information and key performance indicators to be reported to a European database. On 21 September 2026 the European Commission proposed a common rating scheme for data centres, intended to make their energy and water use more transparent. Operators should confirm current obligations and national implementation with the relevant authority.

Reading a rising trend

An operator's monthly PUE rises over a summer and stays high into autumn. Weather explains the summer rise, but not autumn. Comparing trends of IT load, cooling plant power and inlet temperatures shows that a chiller sequencing change made during maintenance has left an extra chiller running at low load. Restoring the sequence brings PUE back to its expected trend. PUE flagged the issue; the diagnosis needed the underlying meters.