Summary
- Article 4's enacted heading is “Data centre sustainability indicators” and its entire operative text is one sentence sending the indicators and their methodology to Annex III. It contains no database, no deadline and no publication duty. The database is Article 5; the operator clock is Article 3(1).
- Article 5(2) requires the information and indicators to be made public “in an aggregated manner, at Member State and Union level”. Article 5(5) requires the Commission and Member States to “keep confidential all information and key performance indicators for individual data centres”. The per-facility return is not a public artefact and not a redaction call.
- The publication duty in Article 5(2) states no interval anywhere in the provisions read, while the operator deadline in Article 3(1) is dated to the day. And Annex IV's three-data-centre minimum applies only to points (vi)–(ix) — narrower than the industry reading that already retells it as a general rule.
On 18 November 2026 I read the half of Commission Delegated Regulation (EU) 2024/1364 that my 17 November briefing could not reach. That earlier piece worked the entrance test and the indicators, and said in terms that Articles 4, 5 and 6 were unread, so the scheme's output could be reported as unestablished but never as unpublished. Here is what the text says.
Article 4 first, because Article 4 was the article I had expected to carry the database. Its enacted heading is “Data centre sustainability indicators”. One sentence follows, and then the article ends: “The data centre sustainability indicators and the methodology to calculate them are set out in Annex III” (Official Journal, English). The same sentence appears in the Official Journal HTML, in the Official Journal PDF and in the consolidated text dated 17 May 2024 (OJ PDF; consolidated text). No database. No deadline. No publication duty. Readers who go to Article 4 looking for the Commission's machinery will find a signpost to an annex.
The machinery sits in three places, and the first one is a clock that binds operators rather than institutions. Article 3(1): “By 15 September 2024, then by 15 May 2025, and every year thereafter, reporting data centre operators shall communicate to the European database the information and key performance indicators set out in Annex I and Annex II regarding the data centre they operate” (Official Journal, English). The dates are exact, and they fall on the operator. Entry into force came on the twentieth day following publication in the Official Journal, which by my own count from the 17 May 2024 date is 6 June 2024. The first interval, 15 September 2024 to 15 May 2025, is 242 days; the next, to 15 May 2026, is 365, with no leap day in between.
The second place is an interface. Article 5(1) requires the European database to apply “a common user interface as well as a common application programming interface ensuring that all reporting data centres are able to communicate, in the same way, the information and key performance indicators referred to in Annexes I and II” (Official Journal, English). Note what the paragraph does not do: it does not say who runs the database, and the operative articles I read never use the verb to establish. That attribution sits in recital 10, which records that “the Commission is to establish, a European database on data centres, in accordance with Article 12(3) of Directive (EU) 2023/1791” (Official Journal, English). A recital, not an operating obligation.
The third place is the perimeter, and this is the paragraph worth the reader's time. Article 5(2): “The information, and key performance indicators, communicated to the European database, and the data centre sustainability indicators, in accordance with Annex III, shall be made public in an aggregated manner, at Member State and Union level, in accordance with Annex IV.” Then two access paragraphs — Member States get everything reported by data centres in their territory, the Commission gets everything reported, full stop. Then Article 5(5): “The Commission and Member States concerned shall keep confidential all information and key performance indicators for individual data centres that are communicated to the database pursuant to Article 3” (Official Journal, English).
Read those two paragraphs together and the competitive question is settled before any official touches a file. The public unit is an aggregate at two levels. The individual unit is what the same article calls confidential, in the imperative, addressed to the two institutions that hold it. There is no redaction judgement here, no business-secrecy balancing test drafted into the text, and no operator election to publish. An operator's name and its power usage effectiveness never appear together in a public document because the regulation never made them a public document.
The exposure that colocation providers argue about was decided at the drafting table.
Article 5(6) then lets the aggregate travel: the collected aggregated data “can be reused for European statistics in line with the principles defined in Regulation (EC) No 223/2009”. The register of the scheme's output extends past the scheme.
Two seams make this more than a settled question.
The first is the missing cadence. Article 3(1) dates the operator's return to the day. Article 5(2) imposes a duty to publish and states no interval at all. Nothing in Articles 1 to 5, in recitals 8 to 11 or in Annexes III and IV, as I read them, fixes how often the aggregate must appear — not annual, not quarterly, not once. Article 6, the last of the act's six articles, is entry into force and nothing else (Official Journal, English). So this scheme can be fully compliant on the operator side and still leave its public output undated. That is an argument from silence inside the provisions I inspected, and it stays one: I did not read every recital, and I did not visit whatever the database publishes today.
The second is the thin-category rule, which is the sharpest hook in the act and the most widely mis-told. Annex IV point (a): “For points (vi)–(ix), presentation of aggregated data per type of data centre and per size category will be possible only if the respective category contains data from at least three data centres” (Official Journal, English; the same sentence in the OJ PDF). Three data centres in a category, or the breakdown does not appear. The scope matters: the sentence attaches to points (vi)–(ix), and as drafted it does not reach the size-band head-count distribution, so a Member State can still show that it holds four facilities in a band without breaching that sentence. Industry commentary already flattens it — the European Data Centre Association writes that “should a category aggregate less than 3 reporting data centres, then no information shall be presented” (EUDCA) — a paraphrase that drops the points (vi)–(ix) limitation. Quote the operative sentence, not the summary of it.
Inside the same point (a) sits the weighting, and it deserves the same care: “the aggregation of the sustainability indicators shall be performed with a weighted metric aggregation, using the total energy consumption as the weighting factor” (Official Journal, English). That is my reading of the arithmetic, not a Commission statement: a published average weighted by total energy consumption follows the largest consumers. A small, efficient operator contributes almost nothing to the number that will describe its country.
The size bands themselves cross a unit seam inside one table — very small at 100–500 kW, small at 500–1 000 kW, then medium at 1–2 MW, large at 2–10 MW and very large above 10 MW (Annex IV transcription; third-party text, used only to corroborate the OJ band wording). Two of the five boundaries are written in kilowatts and three in megawatts.
Three comparators, each with its own delta. Singapore put a number on the applicant at the door — PUE no higher than 1.25 at 100 per cent IT load, with no delivery deadline, no reporting return and no penalty in the factsheet I read in November; that is a gate, and this is a statistic. Equinix publishes a per-cabinet revenue figure with an unprinted denominator; here the per-facility figure is not published at all, and the aggregate replaces it — the opposite perimeter, so the comparison is about disclosure design rather than about numbers.
ERCOT's Batch Zero classified projects into classes with aggregates and no public roster; the return that ERCOT left unpublished is the return this regulation makes confidential in terms, which is the distinction between an unpublished record and a protected one.
Nine boundaries travel with the piece. Nothing here states any interval for publication, and the absence is limited to the provisions I inspected; the article does not claim the aggregate is never published, is overdue, or is annual. As returned, the Union-level limb of Annex IV repeats the weighted-aggregation sentence but not the three-data-centre sentence, which appears only under the Member State-level point (a) — I report that as observed text from a truncated, table-formatted extraction, not as a finding that Union breakdowns carry no small-count rule.
The Annex IV text came through an extracted PDF stream with table layout, not a visual read of the typeset page, and one aggregator's note that Article 3 was amended in November 2026 is unverified and asserted nowhere. Recitals beyond 8 to 11, and Annexes I to III, were truncated or absent from what I read; Articles 1 and 2 were not read in full. Whether the Commission operates rather than establishes the database is not answered by the operative articles, and the text read contains no operating obligation.
No operator, Member State or facility is named, no count of data centres and no aggregate PUE, WUE, ERF or REF value is published here, because none was in the documents I read. The database's public interface was not visited. Article 12(3) of Directive (EU) 2023/1791 was not read, so the empowering provision is known only through recital 10's cross-reference. No implementing act, guidance or CEN/CENELEC methodology document was read, so Annex III's methodologies are named and never described. And the three renderings were compared on the sentences quoted above only.
The European Commission's entry in BTW's directory is here.
So: the article I spent a fortnight assuming carried the database carries one cross-reference. The duty I assumed would be argued over in a redaction meeting is a duty of confidence written into the same article as the publication requirement. And the scheme's public output has a defined unit, a defined weighting and no defined date. Anyone treating a European data centre efficiency average as a market signal should know which of those three came from the text and which one is simply missing.
Member Briefing
Read the full trend briefing
Members read the full briefing with the source notes behind every finding.
Strategic Circle
A daily briefing in your language, built from changes we have verified.
Join for USD 20/monthCancel anytime.
Members also get the full research library, a daily briefing on the networks they follow and the private forum.






