PUE has been the data centre industry's headline efficiency metric for over a decade. The reason is partly historical and partly practical. PUE is a single number, easy to compare across sites, and it answers a question that mattered a great deal when most efficiency gains came from cooling design and building infrastructure. It still has a place. It is just no longer enough.
The structural problem with PUE is what it does not measure. A facility with a PUE of 1.1 looks efficient on paper. If the servers inside it are over-provisioned, running at low utilisation, on hardware two or three generations behind the current curve, the building's PUE will not flatter that outcome. The wasted kilowatts are upstream of the meter; the metric is happy regardless.
The result is a measurement gap that has grown wider every year as building efficiency has improved and IT efficiency has not.
What proper measurement actually looks like
A useful sustainability picture for a data centre covers at least five distinct dimensions:
- Building infrastructure efficiency (the original PUE remit)
- The efficiency of the IT load itself, which is what PUE quietly omits
- Clean energy procurement, both contracted and operational
- Water use, particularly in cooling
- Material flows: the embodied carbon and recovery rates of the hardware on site
The first one is well measured. The other four are not, by most operators, in any honest way. That asymmetry is now the more interesting story in data centre sustainability.
And the standards that name them
The international standards body has, in fact, codified a set of metrics that cover this ground. The ISO 30134 series of data centre KPIs includes:
- PUE: Power Usage Effectiveness (the building's overhead)
- WUE: Water Usage Effectiveness
- ERF: Energy Reuse Factor (what is recovered from waste heat)
- ITEEsv: IT Equipment Energy Efficiency for servers
- ITEUsv: IT Equipment Utilisation for servers
- REF: Renewable Energy Factor
- CUE: Carbon Usage Effectiveness
PUE is the one everyone reports. Most data centre disclosure stops there, occasionally extends to WUE, and leaves ITEEsv, ITEUsv, REF, ERF and CUE unaddressed. The argument for closing that disclosure gap is increasingly being made by the standards body that wrote the metrics. The metrics exist. The reporting around them is what is missing.
Why IT efficiency matters most
Of those five, the area with the largest available gain is IT efficiency itself. The reason is structural: servers are the single largest energy consumer inside any data centre, and the variation in efficiency between server generations (and between configurations within a generation) is significant. A measured rather than modelled view of that variation reveals refresh and consolidation opportunities that conventional reporting misses entirely.
That measurement is what the Interact dataset was built to provide. Operations per Watt per server, across every major vendor and generation, derived from measured behaviour rather than vendor spec sheets.
What this means for procurement
For a buyer of data centre services, the practical implication is to ask different questions. A provider's PUE answer is the start of the conversation, not the end. Useful follow-ups: how are your servers performing per Watt of work, what is the embodied carbon of the hardware in your estate, where are you on water, where are you on heat recovery, and what does your refresh policy actually look like.
The providers that have done the work will have answers. The ones that have not will give you a PUE figure and move on. That asymmetry is the new disclosure gap, and it is the one worth paying attention to.