In short
- PUE = total facility energy ÷ IT equipment energy, both in kWh over the same 12 months.
- 1.0 is the theoretical floor. Everything above it is overhead: cooling, power losses, lighting.
- The global average is about 1.54 (Uptime Institute, 2025); hyperscale fleets report 1.09 to 1.15.
- Defined by The Green Grid in 2007, standardised as ISO/IEC 30134-2 in 2016.
- Germany now requires 1.2 for new data centers; the EU collects PUE from every site of 500 kW IT or more.
The formula
A PUE of 1.30 means that for every 100 kWh delivered to servers, storage and network, the building uses another 30 kWh to cool them, convert power and keep the lights on.
The inverse, DCiE (data center infrastructure efficiency), expresses the same thing as a percentage: DCiE = 1 ÷ PUE × 100 %. A PUE of 1.25 is a DCiE of 80 %. The industry has largely dropped DCiE, but older reports still use it.
A worked example: 2 MW of IT at 70 % load
A hall with 2 MW of installed IT, running at an average 70 % over the year, uses 2 MW × 0.70 × 8,760 h = 12,264 MWh of IT energy. The meters on the cooling plant, the power chain and the building add the rest:
| Energy use over 12 months | MWh | Share of IT |
|---|---|---|
| IT equipment (servers, storage, network) | 12,264 | 100 % |
| Cooling (chillers, CRAH fans, pumps, towers) | 2,453 | 20 % |
| Power chain losses (UPS, transformers, PDUs) | 981 | 8 % |
| Lighting, security, building services | 245 | 2 % |
| Total facility energy | 15,943 | PUE 1.30 |
Cut cooling from 20 % to 12 % of IT energy, for example with containment and free cooling, and the total falls by 981 MWh to 14,962 MWh: PUE 1.22. At $120 per MWh that is about $118,000 a year on a single 2 MW hall.
What counts as IT, and what counts as facility
| IT equipment energy | Facility overhead |
|---|---|
| Servers, including their internal fans | Chillers, cooling towers, dry coolers, pumps |
| Storage arrays and backup systems | CRAH / CRAC units and fan walls |
| Network switches, routers, firewalls | UPS, transformer and distribution losses |
| Monitoring and management equipment in the racks | Lighting, security, fire systems, BMS |
Where the IT energy is metered matters as much as what is metered. That is what the measurement categories describe.
Measurement categories 0 to 3
| Category | Basis | IT energy measured at | Use |
|---|---|---|---|
| Cat 0 | Peak demand (kW) | UPS output | Entry level, no longer recommended |
| Cat 1 | Annual energy | UPS output | Common minimum |
| Cat 2 | Annual energy | PDU output | Better: excludes PDU losses from IT |
| Cat 3 | Annual energy | Server input | Most accurate |
Always state the category with the number. A Category 1 PUE counts PDU losses as IT energy and so looks better than the same site measured at Category 3.
Five ways a PUE figure misleads
1. Low utilisation inflates it
Cooling and electrical overheads have a fixed share. A new hall at 30 % load can show a PUE far above its design value. Compare sites at the same load point.
2. Climate sets the floor
Free-cooling hours vary by site. A Nordic campus gets far more economiser hours than the same design in Singapore or the Gulf.
3. A lower PUE can hide higher IT energy
Raising supply-air temperature cuts chiller energy but can raise server fan power, which counts as IT. Check total facility energy, not PUE alone.
4. Partial PUE is not PUE
A pPUE for one hall or one subsystem leaves out shared plant. Ask which boundary a quoted figure uses.
5. Design PUE is not annual PUE
Design PUE is usually calculated at full load on a design day. Annual PUE at real load is what regulators and lenders look at. Singapore quotes full load, Germany an annual average.
PUE and its sister metrics
| Metric | Measures | Unit |
|---|---|---|
| WUE | Water usage effectiveness | litres per kWh IT |
| CUE | Carbon usage effectiveness | kg CO₂ per kWh IT |
| ERF | Energy reuse factor | reused ÷ total energy |
| REF | Renewable energy factor | renewable ÷ total energy |
EU data centers of 500 kW IT or more now report all four alongside PUE. Heat reuse does not lower PUE; it shows up in ERF. See allinHEATREUSE for the heat side.
Questions about PUE
What is a good PUE?
For a new build in a temperate climate, 1.2 to 1.3 as an annual average is good and 1.2 is now the legal limit for new sites in Germany. Hyperscale fleets report 1.09 to 1.15. The global average is about 1.54.
Can PUE be below 1.0?
No. Total facility energy always includes the IT energy, so the ratio cannot fall below 1.0. Exporting waste heat does not change PUE; it is measured with ERF.
Is a lower PUE always better?
Not on its own. A site can lower PUE by shifting energy into the IT side, for example hotter air that makes server fans work harder. Look at total energy and at WUE as well, since evaporative cooling trades water for power.
How often should PUE be measured?
Meter continuously, review monthly and report the annual figure. Monthly values show the effect of the seasons; the annual value is the one to compare and to report.
What is the difference between PUE and DCiE?
They are the same information. DCiE is the inverse of PUE as a percentage: PUE 1.25 equals DCiE 80 %.
Sources
- ISO/IEC 30134-2:2016, Key performance indicators: Power usage effectiveness (PUE).
- The Green Grid, PUE: A comprehensive examination of the metric (White Paper 49).
- Uptime Institute, Global Data Center Survey 2025.
- Commission Delegated Regulation (EU) 2024/1364; German Energy Efficiency Act (EnEfG), § 11.