The PUE formula

The Green Grid defines PUE as total data-center energy divided by IT equipment energy. A PUE of 1.0 would mean that every unit of energy entering the facility is used by IT equipment, with no energy consumed by cooling, power conversion, lighting or other supporting systems.

In practice, a facility always has some supporting-energy requirement. The value of PUE is therefore not that it produces a universal score, but that it gives operators a consistent way to track how infrastructure overhead changes over time.

A simple calculation example

If a data center uses 1,500 kWh of total facility energy during a measurement period and the IT equipment uses 1,000 kWh, the PUE for that period is 1.5.

The remaining 500 kWh represents facility overhead such as cooling, fans, pumps, UPS losses, transformers, lighting and other non-IT loads included within the measurement boundary.

What can improve or worsen PUE

Cooling efficiency, outside-air conditions, electrical conversion losses, equipment loading and the design of the power and mechanical systems can all affect PUE.

A lightly loaded facility can sometimes show a worse PUE because fixed infrastructure overhead is being spread across a smaller IT load. This is one reason PUE should be interpreted together with operating context rather than viewed as a standalone measure of quality.

Why comparisons need context

Two facilities can have different PUE values for legitimate reasons, including climate, redundancy requirements, operating load, building design and the measurement boundary used in the calculation.

For that reason, PUE is especially useful for tracking the same facility over time and identifying efficiency trends after changes to cooling, power distribution or operating practices.

PUE and AI infrastructure

High-density AI deployments are changing the relationship between IT load and supporting infrastructure. Higher rack densities can increase cooling and electrical-distribution demands, while more efficient cooling architectures can reduce some overheads.

Operators evaluating AI capacity should therefore use PUE alongside other metrics such as rack density, cooling-system performance, available power capacity and workload utilization.