Power Usage Effectiveness is the ratio of total facility power draw to IT equipment power draw. A PUE of 1.0 would mean every watt entering the building powers a server; nothing is consumed by cooling, lighting, or losses. No facility achieves 1.0. The most efficient hyperscale designs land between 1.05 and 1.15. Modern enterprise colocation typically lands between 1.25 and 1.45. Older multi-tenant facilities still operating commercially in Europe land between 1.5 and 1.8.
The number matters because it is one of the few facility-level metrics that compounds directly into your operating cost. It also matters because European energy prices have moved in a direction that makes 0.2 of PUE worth a meaningful sum on a meaningful contract.
What 0.2 of PUE costs you
Take a 100 kW IT load on a five-year contract, paying €0.18 per kWh including all surcharges and grid costs. That IT load alone consumes roughly 4.4 GWh over the five years.
- At PUE 1.4, total facility consumption is ~6.1 GWh. Energy cost: roughly €1.1M.
- At PUE 1.2, total facility consumption is ~5.3 GWh. Energy cost: roughly €0.95M.
A 0.2 PUE delta is around €150,000 over five years on a 100 kW deployment, before any other delta in the contract. On a 1 MW deployment in the same configuration, the same delta is €1.5M. PUE is not a soft factor.
What “guaranteed PUE” actually means
Most providers will quote a design PUE, not an operational PUE. The two diverge for predictable reasons: PUE varies seasonally (a Dutch facility runs lower in winter), it varies with utilisation (an under-loaded facility runs worse), and it varies with how the operator runs the cooling tighter or looser. We push providers to commit to operational PUE under contract — typically a rolling annual average — and to specify the measurement methodology.
A tight measurement methodology has a few features:
- IT load measured at the rack PDU, not at the data hall feed.
- Total facility load measured at utility intake or equivalent.
- Methodology documented at category 3 in The Green Grid’s PUE classification.
- Reporting cadence of at least monthly.
- Independent attestation if the contract is large enough to warrant it.
Without these, the PUE number is a marketing signal, not a procurement input.
Where PUE breaks down
PUE is not the only metric that matters and it is gameable in well-known ways. Three caveats we apply:
It rewards low cooling load, not low waste. A facility that returns waste heat to a district heating network has done something genuinely useful and PUE does not credit it. The Energy Reuse Effectiveness (ERE) metric exists for this and is becoming more common in EU-funded builds.
It penalises high-density configurations even when they are net-better. A 50 kW liquid-cooled rack producing the same compute as five 10 kW air-cooled racks may have a slightly worse measured PUE in some accounting setups, despite delivering the same work for less energy. We compare apples to apples by normalising to compute output where the buyer cares about that.
It does not capture the carbon profile. A facility on 100% certified renewables at PUE 1.4 has a smaller carbon footprint than a coal-fed facility at PUE 1.1. PUE is an efficiency metric, not a carbon metric. CFE (Carbon-Free Energy) percentage and contracted PPA structure are separate questions and we ask them separately.
How we use PUE in shortlisting
For every active shortlist, we ask three questions:
- What is the design PUE, the most-recent twelve-month operational PUE, and the worst-month operational PUE?
- Will the operator commit to a rolling-12-month operational PUE under contract, with a measurement methodology written into the SLA?
- Does the operator publish ERE and CFE alongside PUE, or only PUE?
Facilities that decline to commit to operational PUE in writing are flagged in the shortlist. They are not automatically excluded — sometimes the price more than compensates — but the buyer sees the trade-off explicitly.
PUE is not a future standard. It is a current procurement filter. Treating it as anything else is leaving money on the table.