Summary

  • The Commission’s first reporting period counted 770 data centres, about 36% of the estimated sites subject to reporting; missing coverage and incomplete indicators limit what an EU-wide average can establish.
  • The Commission has opened consultation on minimum performance standards through 14 December 2026 and targets a Q2 2027 legislative proposal. Recommended PUE and WUE baselines in its report are policy inputs, not binding thresholds.

A floor needs a map

Europe is trying to answer two questions that look similar on a label but demand different evidence. The first is how to make data-centre energy and water use more visible. The second is what minimum performance a facility should be legally required to meet. On 21 September, the European Commission advanced a common rating scheme and opened a separate consultation on minimum performance standards. It invited responses by 14 December 2026 and said it aims to present a legislative proposal in the second quarter of 2027. A label can describe reported performance; a floor can change which facilities invest, retrofit, expand or qualify.

Treating the two as one step would overstate what has already been decided.

The distinction matters because the first EU reporting period is a partial map. Under the Energy Efficiency Directive, reporting is required for data centres with installed IT power demand of at least 500 kW. For the first period, May to September 2024, the Commission’s report says 770 sites reported, roughly 36% of its estimated total of facilities that should report. Six Member States submitted no data-centre reports; five submitted fewer than three. The dataset was anonymised, so the public cannot trace each observation to a named facility or independently reconstruct local coverage.

The Commission calls the first sample a useful start and says more reporting is expected. That is fair. Early reporting can reveal inconsistent definitions, locate weak measurement practice and identify technologies worth examining. But a sample covering about one-third of the estimated liable fleet cannot quietly become the whole sector. A national average built from a handful of facilities can be an accurate average of those submissions and a poor description of the country’s operating fleet. Missing records are not proof of poor performance; neither are they evidence that the absent facilities are efficient.

One percentage hides two different gaps

The report’s quality review deemed 70.1% of the reported data reliable. This is not a finding that 70.1% of facilities are reliable, nor a certification rate. It describes reported data points that passed plausibility and consistency checks. The number of observations also changes from one indicator to another because facilities did not submit the entire set of key performance indicators. Coverage and data quality are therefore separate questions: a submitted value can be consistent while many facilities remain unseen, and a complete facility return can still contain a value that needs validation.

That difference should shape the design of a minimum standard. A regulator needs to know not only how many reported sites fall above a threshold, but what fraction of the eligible population reported, which indicators each site supplied, how missing values were treated, and how representative the group is by size, age and location. Without those denominators, a threshold comparison risks making confidence look stronger than the underlying evidence.

The proposed metrics also answer different questions. Power Usage Effectiveness (PUE) divides total facility energy by IT-equipment energy; a higher value means more energy is used for cooling and other non-IT systems per unit of IT load. Water Usage Effectiveness (WUE) relates water input to IT energy. The Commission reports a 2024 average PUE of 1.36. It also says roughly half of reporting centres exceed a recommended PUE baseline of 1.5 and/or a recommended WUE baseline of 0.4. “Exceed” here means a higher, less favourable value.

These baselines appear in the Commission’s analysis as recommended reference points; they are not adopted legal limits.

The water comparison is especially sensitive to what is counted. The report’s weighted average WUE of 0.58 comes from 458 facilities with reliable, non-zero water values. Sites relying on dry cooling were excluded from that calculation. The figure can describe the reporting subset and still fail to represent water pressure across the full fleet or local catchments. Cooling design, climate, facility size and water availability all affect the interpretation.

A low EU-wide WUE cannot establish that a site is harmless in a stressed basin; a higher value does not by itself tell a regulator whether a different cooling design was technically or locally feasible.

Heat reuse makes the same point in another form. Only 67 sites reported some waste-heat reuse in the first period. The Commission estimated that about 1.8% of generated heat was reused, while warning that reporting gaps and values that could not be fully validated make that estimate potentially unrepresentative. Reuse depends on nearby demand, a transport network, investment and whether supply matches the hours when a district-heating customer needs heat. A rule that rewards reuse without those conditions could score geography and local infrastructure as if they were solely operator choices.

The evidence must follow control

The report acknowledges a further complication for colocation facilities: the operator may not have complete information about customer-controlled IT equipment. If a performance obligation depends on a tenant’s workload or energy measurement, assigning the entire duty to the building operator can place the legal risk on a party that cannot observe or change the relevant input. The answer is not to exempt colocation by default. It is to define a workable data-sharing route, identify which party controls each measurement and specify what evidence is acceptable when customer data cannot be obtained.

The Commission’s own policy sequence points toward that question. The reporting database and Delegated Regulation (EU) 2024/1364 provide the data collection layer. The new common rating scheme is intended to turn reported information into a public label. Minimum standards would add a compliance threshold. Each step increases the consequence of error: an imperfect input can first create an imperfect comparison, then a misleading benchmark, and finally an investment or enforcement decision.

Heng Lu’s Note 71, “The Policy Mirror,” offers an editorial lens here, not evidence about EU law: read the instrument to see what an institution believes it can measure and govern, then test that belief against the operating record. Applied carefully, the question is not whether the Commission has authority to pursue energy efficiency. It is whether a given metric is observable, comparable and attributable to the actor expected to improve it.

The Commission’s study page says the preparatory study began at the end of January 2026 and runs for about 24 months, while the Commission separately targets a Q2 2027 proposal. That timeline creates a question for the consultation: which elements can be decided from the first two reporting cycles, and which need later validation? It does not prove the proposal will be delayed or that the timetable is unworkable. It does make the quality of interim evidence a material policy choice.

What a defensible standard would disclose

The Commission says possible standards could use PUE, WUE, REF or a combination, and should account for facility size, age, climate and location. A defensible framework would take that seriously. It would publish each metric’s reporting denominator and sample size; distinguish non-reporting from measured underperformance; show measurement periods and system boundaries; and contextualise water indicators with local stress and cooling design. It would state how new, retrofitted and older facilities are treated, how colocation customers contribute data, and how an operator can correct an erroneous return or challenge an attribution.

The rule should also be reviewable. Thresholds built from an incomplete first sample should carry a scheduled reassessment against later reporting rounds, with published validation methods. A common EU rule may reduce fragmentation among national requirements, as the Commission argues. Yet harmonisation is valuable only if the underlying comparison is meaningful. Otherwise the EU standard could make paperwork uniform while leaving performance measurement uneven.

For now, the evidence supports a narrower conclusion than either advocacy camp might want. Data centres consumed 68 TWh of electricity in the EU in 2024, and the Commission cites a projection of 114 TWh by 2030. That trajectory makes efficiency and infrastructure planning consequential. The first reporting dataset is enough to justify improving visibility and preparing options; it is not enough to treat every reported average as a complete account of the sector. The next test is whether the proposal makes the map more complete before it turns the map into a legal floor.

Sources