Power Usage Effectiveness (PUE)
Also known as: PUE, data centre efficiency, power usage effectiveness, data center PUE
Power Usage Effectiveness (PUE) is the standard metric for data centre energy efficiency.
It is calculated as the ratio of total facility energy consumption (including cooling, lighting, and power distribution) to the energy consumed by IT equipment (servers, storage, networking). A perfect PUE of 1.0 would mean all energy goes directly to computing. In practice, the best hyperscale data centres achieve PUE around 1.1; typical enterprise data centres are 1.5-2.0. Lower PUE is better.
Power Usage Effectiveness (PUE), explained at three levels
At home
PUE tells you how much wasted energy a data centre has. A PUE of 1.0 would be perfect: every watt of electricity goes to computers. A PUE of 2.0 means that for every watt used by computers, another watt is wasted on cooling, lighting, and power distribution. Most good modern data centres are between 1.1 and 1.4. Older, less efficient data centres might be 2.0 or higher: essentially using twice as much electricity as the computers themselves actually need.
At work
PUE matters to businesses because it affects the carbon and cost efficiency of their cloud computing and co-location services. When comparing data centre providers, a lower PUE (all else equal) means less wasted energy and a smaller carbon footprint per unit of compute. Cloud providers typically publish their average PUE; co-location providers should be able to provide facility-specific PUE data. However, PUE alone does not tell the whole story: a very efficient data centre in a coal-intensive grid may have a higher carbon footprint than a less efficient data centre running on renewables.
At home
PUE is calculated as: PUE = Total Facility Power / IT Equipment Power. If a data centre uses 2.5 MW total and the IT equipment uses 2.0 MW, PUE = 2.5/2.0 = 1.25. The remaining 0.5 MW is overhead: cooling (the largest overhead in most facilities), UPS losses, transformers, and lighting. Reducing PUE means reducing this overhead without reducing the amount of IT equipment powered.
At work
The main drivers of PUE improvement are cooling system efficiency and server utilization. Cooling improvements include using outside air cooling (free cooling) when ambient temperature is low enough, moving from air cooling to liquid cooling for high-density GPU clusters, and using evaporative cooling instead of mechanical chillers. Higher server utilization improves PUE indirectly: a data centre where servers are idle most of the time has low IT load while cooling overhead remains high, increasing PUE. Virtualization and cloud computing improved PUE in the industry partly by increasing server utilization.
At home
PUE has limitations as a metric. It does not capture the carbon intensity of the electricity consumed: two data centres with the same PUE can have vastly different carbon footprints depending on their electricity source. It does not capture water consumption (evaporative cooling trades higher water use for lower electricity use). It does not capture server utilization: a very efficient, very idle data centre has good PUE but poor resource efficiency. The Green Grid has proposed additional metrics (WUE for water, CUE for carbon) but PUE remains the dominant reported metric.
At work
From a practitioner standpoint, when evaluating cloud providers or co-location facilities from an environmental perspective, request annual average PUE data for the specific facility (not just corporate average), the carbon intensity of the electricity supply, and the renewable energy sourcing approach. A facility with PUE 1.15 running on 100% renewable electricity is environmentally superior to a facility with PUE 1.1 running on coal-intensive grid electricity, despite the higher PUE: the carbon metric matters more than the efficiency metric.
Real-world examples
At home
Nadia works in facilities management and was asked to report her company's co-location data centre's environmental impact. She asked the co-location provider for their annual PUE, found it was 1.6 (meaning 60% overhead energy), and compared it to the industry average of 1.5 and Google's reported 1.1. She used this comparison to make the case for renegotiating to a more efficient facility.
At work
Google publishes quarterly PUE data for each of its data centres globally. Their trailing twelve-month global average PUE was 1.10 in 2023, meaning about 91% of all energy consumed went directly to computing, with only about 9% going to overhead. This transparency benchmark has influenced the broader industry to improve reporting and to work toward sub-1.2 PUE as standard for new hyperscale construction.
Common confusions
- A lower PUE does not automatically mean a lower carbon footprint. PUE measures efficiency of energy use within the data centre; it does not measure the carbon intensity of the electricity. A data centre with PUE 1.1 powered by coal has a higher carbon footprint per compute unit than a data centre with PUE 1.5 powered by solar. When assessing environmental impact, always consider both PUE and carbon intensity of the electricity supply.
- PUE is an average metric and can be misleading. Annual average PUE varies with seasons (cooling overhead is lower in winter), with IT load (lower load means higher PUE because cooling overhead remains roughly constant), and with measurement methodology. Compare PUE figures from different providers only when calculated using the same Green Grid methodology and the same measurement period.
- PUE does not account for what the computing is being used for. A data centre processing medical records, financial transactions, and video streaming all have the same PUE if their cooling efficiency is the same, but the social and economic value of the compute differs enormously. PUE is an efficiency metric for the physical facility, not a measure of the value created by the computing.
For analysts
In BA artefacts, PUE appears as a green IT procurement criterion and a sustainability reporting metric. When specifying co-location or managed hosting requirements, include: 'The facility's annual average PUE must not exceed [X], verified by third-party attestation, with quarterly reporting to the sustainability team.' For cloud services, request that the provider supply facility-level PUE data for the regions used, to support Scope 2 emissions calculations in the organization's sustainability report.
History & origin
PUE was developed by The Green Grid consortium in 2006 as a standardised metric for data centre efficiency, replacing inconsistent earlier measures. It was adopted by the US EPA and became the de facto industry standard for reporting data centre energy efficiency. Google was an early champion, publishing its quarterly PUE figures from 2008 as part of its environmental transparency efforts: helping establish industry best practice. The metric has limitations (it does not capture carbon intensity, water use, or server utilization) but remains the most widely reported data centre efficiency metric.