Summary

  • Delegated Regulation (EU) 2024/1364 puts its 500 kW doorway on a quantity defined in Article 2(14) and counts "data centres" defined in Annex A of the energy-statistics Regulation, a definition the counting instrument never carries.
  • Article 3(2) let a facility omit eleven of the eighteen Annex II energy and sustainability points in the first period, and all four published ratios are built from those omissible quantities.
  • Annex IV prints the number of reporting facilities and the four averages; it does not print, indicator by indicator, how many facilities supplied the number behind each average.

The doorway is exact

Start with the part that is written to the digit. Article 1 applies to operators of data centres with an installed information technology power demand of at least 500 kW (the delegated regulation). Article 2(14) then says what that quantity is: the sum of the nominal power demand, in kilowatts, of the networks, servers and storage equipment installed in the data centre computer room floor area.

That is neither a facility's electricity bill nor a design load on the transformer. It is the sticker on the kit standing in the hall. The Commission's FAQ, in the September 2026 version, supplies the perimeter rule that goes with it: the threshold applies to installed capacity rather than design or potential capacity, IT power is calculated per site, and data centres at different addresses are separate entities even inside one local administrative unit (FAQ v1.9).

The counted object is filed elsewhere

Article 2 defines enterprise, colocation and co-hosting data centres, their operators, floor areas, installed IT power demand and rated IT load. It does not define a data centre. The word the whole scheme turns on lives in Annex A, point 2.6.3.1.16, of Regulation (EC) No 1099/2008 on energy statistics, as replaced by Commission Regulation (EU) 2022/132:

"A data centre is defined as a structure or a group of structures used to house, connect and operate computer systems/servers and associated equipment for data storage, processing and/or distribution, as well as related activities." (2022/132, Annex A)

The FAQ points to exactly that provision; that is the joint between the two instruments. Read the borrowed definition beside the per-site rule and an ordinary operator question appears with no published answer: a campus of halls at one address is a "group of structures" and therefore one reporting entity, while the same halls across a road are, on the FAQ's reading, two.

The dossier is a register wearing statistics clothes

Annex I collects the data centre's name; its owner and operator, with contact details; its location as a Local Administrative Unit code; its type; and the year and month it entered operation — followed by electrical redundancy at high voltage, low-voltage line-up and rack level, and cooling redundancy at room and rack level (consolidated text, Annex I). Whatever else the European database becomes, at intake it is a named, located, dated register of facilities above half a megawatt.

Eleven of eighteen, by permission

Annex II asks for 24 key performance indicators in three blocks. Eighteen of them are the energy and sustainability block, lettered (a) to (r). Article 3(2) then does something a reader should sit with: for the first reporting period it permits omission of points 1(d), 1(e), 1(h) to 1(l) and 1(o) to 1(r), for technical reasons, with an explanation.

That list is eleven points long: (d), (e), (h), (i), (j), (k), (l), (o), (p), (q), (r). My count, taken from the two provisions.

The ratios are assembled from the omissible set

Annex III defines four sustainability indicators. Set them beside the Annex II lettering and the shape of the first published table becomes visible:

The ratios are assembled from the omissible set
Ratio Definition Annex II inputs Omissible in the first period
PUE E_DC / E_IT 1(d) total energy consumption, 1(e) IT equipment energy both
WUE W_IN / E_IT 1(h) total water input, 1(e) both
ERF E_REUSE / E_DC 1(j) waste heat reused, 1(d) both
REF E_RES-TOT / E_DC 1(o) total renewable energy consumption, 1(d) both

Every numerator and every denominator of all four ratios sits inside Article 3(2)'s permission. That mapping is mine — I have joined Annex III to Annex II point 1 — and no document I read states which facilities used the derogation, in what numbers, or why. What the text does establish is narrower and enough: the first published averages were computed over quantities a facility was entitled to withhold, and the instrument does not say how many withheld anything.

A second thing is worth keeping straight. Per facility, ERF multiplied by PUE gives E_REUSE divided by E_IT, and REF multiplied by PUE gives E_RES-TOT divided by E_IT: four ratios resting on two denominators — total facility energy and IT energy — with the facility's own PUE converting between them. That identity is my arithmetic, and it does not transfer to the published table, whose four averages are computed across facilities and cannot be recombined into one another.

What the exit prints

Article 5(2) makes the reported information and indicators public "in an aggregated manner, at Member State and Union level", in accordance with Annex IV (Article 5 and Annex IV). Annex IV's public list includes the number of reporting data centres, the distribution by size category, total installed IT power, total energy consumption, total water consumption, and average PUE, WUE, ERF and REF. Category breakdowns are permitted only where a category holds at least three data centres. Article 5(5) requires the Commission and the Member States to keep confidential all information and indicators for individual data centres, while Articles 5(3) and 5(4) hand Member States everything for their own territory and the Commission everything reported.

The table can therefore tell you how many. It cannot tell you who, and it does not tell you, indicator by indicator, how many facilities supplied the number each average rests on. The first of those is a design choice written into the article. The second is a field the schema does not have.

The coverage line divides a count by an estimate

The Commission's October 2026 report on the first period states that "770 data centres reported their data in the first reporting period, representing around 36% of the estimated total of data centres in the EU falling under the reporting requirement" (report on the energy efficiency of data centres). The same report prints the averages — 1.36 for PUE, 0.58 for WUE — and records that only 67 data centres reported any waste-heat reuse, about 1.8 per cent of total heat generated.

The denominator behind the headline comes from a slide: "The project team estimated that there are 2161 data centres in EU under the reporting obligation", alongside about 14,000 GWh of consumption, roughly 0.5 per cent of EU electricity (DG ENER presentation to the CA EED plenary). 770 divided by 2,161 is 35.6 per cent, which is what "around 36%" rounds to. The numerator is a count of reports. The denominator is a contractor's estimate, and no document I read states how it was built.

The report also prints the gaps: "Six Member States have reported zero data centres, and five Member States have reported fewer than three data centres." Run the subtraction — mine, not the report's — and of 27 Member States at most 21 reported anything at all, five of which reported fewer than three facilities. At most sixteen Member States can therefore hold a population in which a category breakdown survives the three-facility rule. Sixteen is a ceiling, not a count of what appears in the published tables.

Then there is the band. Annex IV's size ladder opens with "very small, 100-500 kW" — a category below the only threshold in the instrument, inside a regulation whose own title calls this the first phase of a common Union rating scheme. BTW reported that ladder on 18 November 2026. It is here because it belongs to the same table, not as a new discovery.

What I read, and what I did not

The dashboard is a URL and nothing more in this piece. The Commission identifies a public dashboard carrying the aggregated data; I did not open the interactive sheet, and no figure, filter, suppression behaviour or last-update date is quoted from it (the dashboard as identified). A URL is not a value.

The first analysis of the cycle is a technical-assistance deliverable: the First Technical Report of the study supporting Article 12(5) of Directive (EU) 2023/1791, produced by EY, AIT and Borderstep for the Commission, recorded by the Publications Office (publication record). I read the October 2026 Commission report as search-returned excerpts of its PDF rather than line by line, and Annex IV arrived in substantial but possibly truncated form.

One question stays open, and it is the one that matters to anybody quoting these numbers: whether each average was computed over the facilities that supplied that indicator, or over all facilities that reported. Nothing I read settles it, and the per-indicator count that would settle it is not published.

What a reader should ask for

Three disclosures would make this table auditable without naming a single facility: the number of respondents behind each indicator; the share of the estimated 2,161 that the reporting population represents, stated as a count against a base rather than a percentage of an unnamed one; and the base of each average — indicator suppliers, or all reporters.

Until those arrive, treat 1.36 as a provisional figure with an unknown denominator, not as a European benchmark. That is not a scandal; it is how an average of optional numbers behaves. The first period was the one period in which the inputs were optional by right. The next aggregate is the test of whether the base gets printed, because the permission in Article 3(2) is written for the first reporting period and not for the ones after it.