Written by 
Reviewed by PUE calculator
To calculate PUE, divide the total energy a data centre uses by the energy used by its IT equipment. A data centre that draws 1,500 MWh in a year, of which 1,000 MWh reaches the servers, storage and network, has a PUE of 1.5. A PUE of 1.0 would mean every kWh goes to the servers, with nothing spent on cooling, power conversion or lighting.
Try your own figures, both in MWh for the same year.
- Cloud providers
- Industry average
On the calculator's scale, the shaded band is the fleet-wide PUE that AWS (1.14), Microsoft Azure (1.17) and Google Cloud (1.09) publish, and the line is the industry average of 1.54 from the Uptime Institute's 2025 survey. Individual regions can sit outside the band.
This guide covers the PUE formula, a worked example, how to calculate DCiE, what a good PUE number looks like, the figures AWS, Azure and Google Cloud publish, and how Greenpixie uses PUE to estimate the electricity and emissions behind cloud usage.
Key takeaways
- PUE = total facility energy ÷ IT equipment energy. Lower is better, and 1.0 is the theoretical minimum.
- Measure it over a full year: cooling load changes with the seasons, so a single reading can flatter or punish a site.
- The industry average was 1.54 in 2025. The large cloud providers publish figures between about 1.09 and 1.17.
- PUE measures facility overhead only. It says nothing about how efficient the servers are, how much water is used or how clean the grid is.
What is power usage effectiveness (PUE)?
Power usage effectiveness (PUE) is a ratio that shows how much of a data centre's energy reaches its servers, storage and networking, and how much goes on overhead such as cooling, power conversion and lighting. It was introduced by The Green Grid and is standardised as ISO/IEC 30134-2.
Understanding PUE is important for data centers and their customers because it is the most widely used measure of data center energy efficiency, and it is simple: two meter readings and one division. That simplicity is also its limit, which we cover below.
The power usage effectiveness formula
PUE = total facility energy ÷ IT equipment energy
- Total facility energy: Everything the data centre draws from the grid and any on-site generation: IT equipment plus cooling, chillers, fans, uninterruptible power supplies (UPS), power distribution losses and lighting.
- IT equipment energy: The energy used by servers, storage and networking equipment. Depending on how detailed the metering is, it is measured at the UPS output, at the power distribution units that feed the racks, or at the servers themselves.
You can also write it as PUE = 1 + (overhead energy ÷ IT energy). A PUE of 1.2 means 0.2 kWh of overhead for every kWh of IT load.
Use energy usage (kWh or MWh over a period), not power consumption at a single moment (kW). Annual energy usage is the standard basis, because it captures hot and cold seasons and changes in load.
How to calculate PUE: worked example
- Measure the site's total energy consumption for the year: Take it from the utility meters at the site boundary. Say the site used 4,380 MWh.
- Measure IT energy use for the same year: Take it from the UPS outputs or the rack-level power meters. Say the IT load used 3,500 MWh.
- Divide: 4,380 ÷ 3,500 = 1.25. The PUE is 1.25.
- Work out the overhead: 4,380 − 3,500 = 880 MWh went on cooling, power conversion and lighting, which is 0.25 kWh for every kWh of IT load.
For an accurate PUE, keep the measurement points the same from year to year. Moving the IT meter from the UPS output to the rack can change PUE without anything in the building changing.
To try your own figures, use the PUE calculator at the top of this page.
How to calculate DCiE
Data center infrastructure efficiency (DCiE) is PUE turned upside down and shown as a percentage:
DCiE = (IT equipment energy ÷ total facility energy) × 100%, which is the same as 100% ÷ PUE.
In the example above, DCiE = 3,500 ÷ 4,380 × 100% = 80%. Four-fifths of the energy reaches the IT load. PUE has largely replaced DCiE in reporting, but some older dashboards and tender documents still ask for it.
| PUE | DCiE | Overhead per kWh of IT load |
|---|---|---|
| 2.0 | 50% | 1.0 kWh |
| 1.5 | 67% | 0.5 kWh |
| 1.25 | 80% | 0.25 kWh |
| 1.1 | 91% | 0.1 kWh |
Partial PUE (pPUE) applies the same idea to one part of a site, such as a single data hall or a containerised module: the total energy used inside that boundary divided by the IT energy within it. Operators use pPUE to find which zone is pulling the site's overall PUE up.
What is a good PUE number?
A good PUE depends on the type and age of the data centre and its climate. As a guide:
- Industry average: A weighted average of 1.54 in 2025, according to the Uptime Institute's annual survey, the sixth year in a row with almost no change. Sites of 20 megawatts and above averaged 1.44.
- Efficient enterprise or colocation sites: Around 1.3 or below. Only 15% of Uptime's survey respondents reported a PUE of 1.3 or better in 2025.
- Hyperscale cloud providers: 1.09 to 1.17 across their published fleets (see below).
Hot, humid climates need more cooling, which raises power consumption, so the same design will run at a higher PUE in Singapore than in Stockholm. Newer sites with liquid cooling and higher temperature set points tend to do better.
PUE for AWS, Azure and Google Cloud data centers
Each provider publishes PUE for its own estate, but on a different basis:
| Provider | Latest published PUE | Basis |
|---|---|---|
| Google (all Google data centres) | 1.09 fleet-wide, trailing twelve months | By campus, trailing twelve months, all seasons and sources of overhead |
| AWS | 1.14 global average, 2025 (1.15 in 2024) | Calendar year, published for most regions |
| Microsoft Azure | 1.17 global, FY25 (Americas 1.16, EMEA 1.16, Asia Pacific 1.28) | Fiscal year (July to June), datacenters Microsoft fully owns and controls, operational 12 months or more |
Because the measurement boundary, the period and whether leased capacity is counted all differ, these figures compare best within one provider rather than across them. AWS's Stockholm region, for example, reported 1.09 in 2025, against 1.14 for AWS as a whole (AWS).
How Greenpixie uses PUE in cloud emissions
Cloud customers can't meter the data centres they use, so PUE is how facility overhead gets into a cloud carbon footprint. In Greenpixie's methodology, IT electricity is estimated for each line of cloud usage from the hardware behind each SKU, multiplied by the region's PUE to give electricity usage, then multiplied by the grid carbon intensity for the region, by hour where the data allows:
The PUE for each region is the provider's own published figure, carried unchanged. AWS regions take the value AWS publishes for them. Azure regions take the value for their geographic grouping. A Google Cloud region takes its campus's trailing twelve-month value, or the average where it matches several campuses. Where a provider publishes nothing that covers a region, Greenpixie assigns a value from other published figures and records which location it came from, and why (Greenpixie grid methodology).
"Cloud providers don't give customers a meter reading, so PUE is how we bring the data centre's overhead into the numbers. We take the IT electricity behind each line of cloud usage, add the facility's inefficiency using the provider's published PUE, then multiply by the local grid carbon intensity at the time the work ran. Leave PUE out and both the electricity and the carbon come out too low."
Take a workload that uses 100 kWh of IT electricity:
| Region PUE | IT electricity | Total electricity | Data centre overhead |
|---|---|---|---|
| 1.10 | 100 kWh | 110 kWh | 10 kWh |
| 1.25 | 100 kWh | 125 kWh | 25 kWh |
That is about 14% more electricity for the same work, purely from data centre overhead, before grid carbon intensity is applied. Because Greenpixie applies each region's PUE to every line of usage, that difference shows up in the electricity, carbon and water figures for each line of a customer's cloud bill.
Greenpixie's methodology has been independently verified in accordance with ISO 14064-3, prepared with reference to ISO 14064-1, and designed to be in line with the GHG Protocol Corporate Standard. You can compare the electricity, carbon and water of the same cloud usage across regions in the Greenpixie app.
What PUE doesn't tell you
- IT efficiency: PUE only measures overhead. A data centre full of idle or outdated servers can still have a low PUE. Energy efficiency inside the servers, from more efficient hardware and better use of resources, is a separate job.
- Carbon: Two sites with the same PUE can have very different emissions, depending on the carbon intensity of the grid behind them.
- Water: Evaporative cooling can lower PUE while using more water. Water usage effectiveness (WUE) covers that side.
- Comparability: Different boundaries, periods and measurement points mean a published PUE is a guide, not a precise ranking.
How data centers improve PUE
- Raise cooling set points within equipment limits, and use free-air or economiser cooling where the climate allows.
- Contain hot and cold aisles so chilled air isn't wasted mixing with hot exhaust.
- Move high-density racks to liquid cooling.
- Use efficient UPS systems and reduce power conversion steps.
- Right-size cooling and power systems to the actual IT load to improve operational efficiency: a half-empty hall still carries fixed overhead, which pushes PUE up.
For cloud users, the levers are different: choosing efficient regions and cutting the IT load itself by removing idle resources. Consolidating servers cuts total energy, even though it can nudge a site's PUE up, which is one reason to track total energy alongside PUE. Our article on data centre energy consumption covers the wider picture.
Frequently asked questions
How do you calculate PUE in data centers?
Divide the data centre's total energy consumption by its IT energy use over the same period, usually a year. 1,500 MWh total and 1,000 MWh IT gives a PUE of 1.5.
How do you calculate PUE and DCiE?
PUE = total facility energy ÷ IT energy. DCiE = IT energy ÷ total facility energy × 100%, which is 100% ÷ PUE.
What is a good PUE for a data center?
Below the 2025 industry average of 1.54 is better than most. Around 1.3 or below is efficient for an enterprise or colocation site, and the large cloud providers report 1.09 to 1.17.
Can PUE be less than 1?
No. The energy used by IT is part of the site's total energy use, so the ratio can't fall below 1.0. Heat reuse, where waste heat is exported to a heating network, doesn't change PUE; other metrics cover it.
What is the PUE of AWS, Azure and Google Cloud?
AWS reported a global average of 1.14 for 2025, Microsoft 1.17 for FY25 across its own datacenters and Google 1.09 across its whole fleet on a trailing twelve-month basis. Each uses its own measurement boundary and period.
Does PUE include water or carbon?
No. PUE only covers energy overhead. Water is measured with WUE and carbon depends on the grid's carbon intensity.
From PUE to your cloud bill
See the energy, carbon and water behind your cloud usage
GPX Cloud adds them to every line of your AWS, Azure and Google Cloud billing data, with each region's data centre overhead included.
Sources
- The Green Grid, PUE: A Comprehensive Examination of the Metric
- Uptime Institute, Global Data Center Survey 2025
- Google, Data center efficiency (accessed 9 November 2026)
- AWS, Sustainability (accessed 9 November 2026)
- AWS, Sustainability fact sheet: Sweden
- Microsoft, Measuring energy and water efficiency for Microsoft datacenters (accessed 9 November 2026)
- Greenpixie Cloud methodology: Assessment and calculation
- Greenpixie Grid and regions methodology: Assessment and calculation
- Greenpixie Grid and regions methodology: Outputs and interpretation
