There is a number that says more about a data centre than any brochure: how much of the energy entering the building actually reaches the IT systems, and how much is lost along the way — cooling, electrical conversions, lighting. That number is called PUE, and it is worth understanding properly even if your “data centre” is a server room with three racks.
What PUE is, in one line
PUE (Power Usage Effectiveness) is the metric proposed in 2007 by The Green Grid, the international consortium for the efficiency of IT infrastructure. The definition fits in a single division:
PUE = total energy drawn by the facility / energy drawn by the IT systemsThe ideal value is 1.0: all the energy goes to IT, zero losses. A PUE of 2.0 means that for every watt consumed by servers, storage and network, another whole watt goes on “overhead” — above all cooling. The closer the number gets to 1, the more efficient the facility.
How much the world consumes
The annual surveys by the Uptime Institute — the industry reference for these studies — tell a story in two acts: a big improvement in the metric's early years, as the basic practices spread, and then a global average stuck at around 1.5–1.6 for many years now. Put plainly: the “average” data centre still wastes roughly a third of the energy it consumes.
And that is the average of the data centres that measure themselves. Small, unattended server rooms — a domestic air conditioner, no airflow containment, no dedicated meter — typically sit above 2: for every euro of electricity doing useful work, another euro evaporates.
Measuring it while spending (almost) nothing
The good news: to get started you do not need a DCIM system. You need two numbers, taken over the same period:
- the total energy of the room or the building — from the meter or the bill, if the room has a supply you can single out;
- the energy of the IT systems — the UPS output reading is often the most convenient point, or a clamp meter on the panels feeding the racks.
Monthly sampling is already enough to see the seasonality (summer cooling makes itself felt) and to give management a number. The “grown-up” version — continuous measurement, alarms, trends — comes later, with instrumentation integrated into monitoring: first comes the habit of measuring, then the tool.
The first measures are the most profitable
It is the same logic the Data Center Maturity Model puts in black and white: the jumps from the lower levels are largely made with low-cost measures, and those are the ones that pay back the most.
- Airflow management: separating hot and cold air (aisles, blanking panels in the racks, sealing cable openings) — often the single measure with the best cost/benefit ratio;
- More honest set-points: many rooms are kept colder than necessary; raising the operating temperature within the ranges specified by manufacturers reduces the cooling workload;
- Free cooling: using outside air when conditions allow, instead of producing cold mechanically;
- Consolidate and switch off: virtualisation, deduplication, decommissioning “zombie” systems — every IT watt eliminated saves further watts in cascade across the overhead.
Where PUE misleads
PUE measures the efficiency of the facility, not of the IT. This is where its paradox arises: if you consolidate your servers and halve IT consumption, the denominator drops and the PUE… gets worse, even though you are consuming less in absolute terms. An excellent PUE can coexist with hundreds of servers running at 10%.
That is why PUE should be read alongside its sister metrics — CUE for the carbon footprint, WUE for the water footprint, ERF for heat reuse — and within a broader grid that also looks at the real utilisation of compute, storage and network. That is exactly what the DCMM does: PUE is the thermometer, not the diagnosis.
The proof at a non-hyperscaler scale
That these principles also work outside the giants is something we have verified on ourselves. The green data centre we deliver our services from is designed around efficiency starting with the building envelope: a low-transmittance timber structure, low density, high-temperature operation (~28 °C), free cooling with convective flows alone in winter — and power 100% from renewable sources, documented with Guarantees of Origin and Scope 2 emissions of zero (the certificates are public and downloadable).
The point is not the record: it is that measuring, designing on data and improving step by step works at any scale — from the three-rack room upwards. The full story of our approach is on the commitment to the environment page; if you want to put a number on your infrastructure, let's start there.
Sources and references: The Green Grid (PUE definition, 2007; Data Center Maturity Model, 2011); Uptime Institute, Global Data Center Survey (annual surveys of the industry's average PUE).
