Summary

  • Article 1 of Delegated Regulation (EU) 2024/1364 reaches operators whose "installed information technology power demand" is "at least 500 kW" — a nominal rating over networks, servers and storage inside the computer-room floor area, declared by the party being reported on.
  • The PUE and WUE figures that circulate as Europe's data-centre scorecard are not indicators in the return at all: Annex III derives each from two reported inputs, both supplied by the same operator, and nothing in the text I read requires an independent verification, audit or certification.

A door you measure yourself

The act opens with a threshold and a unit. Article 1 applies to data-centre operators whose "installed information technology power demand" is "at least 500 kW" — those are the regulation's own words, in the consolidated English text on EUR-Lex. Half a megawatt, and the campus behind it can be two hundred times larger: 500 kW is 0.5 MW, and a 100 MW site is 200 times the number that decided whether the site had to file at all.

What the number counts is narrower than what the site draws. As the text I have reads Article 2(14), installed IT power demand is the sum of the nominal power demand, in kilowatts, of the networks, servers and storage installed in the computer-room floor area. Nominal, not metered. Installed, not consumed. Power, not energy. The cooling plant, the switchgear, the generators and the building sit outside the sum, which is the right engineering choice for an IT-load test and also the reason the figure cannot be compared with a connection agreement.

Then comes the allowance that turns a measurement into a declaration. Annex II point 1(a), on my reading, permits a weighted average where demand changes during the reporting period — and permits rated IT power demand where installed demand cannot be determined. Rated means nameplate. An operator that cannot establish installed demand can open the door with a figure printed on equipment rather than observed at a meter.

Three dates, and the first two already spent

Article 3(1) sets the calendar in one sentence: "By 15 September 2024, then by 15 May 2025, and every year thereafter". From the first date to the second is 242 days; from the second to 15 May 2026 is 365. All three fall before this article's dateline.

The period reported is the calendar year that has just ended, and a centre that operated for less than a year reports only the period it operated. A hall energised in October therefore describes three months against a year that closes on 31 December: a fair rule, and the first place where two returns can be compared without being about the same thing.

What the return actually holds

The reading I have of Annex II lists twenty-four indicators in three families.

Energy and sustainability: installed IT power demand; total data-centre floor area and computer-room floor area; total data-centre energy, with backup-generator energy measured separately; IT-equipment energy; whether the centre provides services supporting grid stability, reliability or resilience; average battery capacity offered to the grid; water entering the centre boundary and potable water input; waste heat reused outside the boundary and its average temperature; refrigerant types; cooling degree days on a 21 °C base; and renewable energy in four lines — a total, purchased and retired Guarantees of Origin, power-purchase agreements, and on-site generation.

ICT capacity and traffic: server capacity on a SERT basis or equivalent; storage capacity as the raw addressable capacity of installed drives, with capacity taken as of 31 December; average provisioned incoming and outgoing bandwidth; and total incoming and outgoing data traffic in exabytes, for which the operator may rely on its own data, its customers', a telecommunications operator's or a service provider's.

Twenty-four numbers, none of them a ratio — and one procedural sentence that matters more than most of them. Records of measurement points and devices must be kept for at least ten years.

The famous ratio is not in the return

PUE is not an Annex II indicator. Neither is WUE. Annex III derives four ratios from reported inputs: PUE as E_DC over E_IT; WUE as W_IN over E_IT; ERF as E_REUSE over E_DC; REF as E_RES-TOT over E_DC.

Read that as an auditor would. Each ratio is two operator-supplied numbers with an operator-drawn boundary between them, and the ratio itself has no row in the return from which anyone could check it. The definitions belong to the instrument; the inputs belong to the operator; the quotient is arithmetic.

There is also a unit seam. The reported IT-energy indicator is in kWh. The WUE denominator is specified with E_IT expressed in MWh. Reconciling them needs a factor of one thousand, and the reading I have does not say where that conversion is applied. A filer closes that seam without thinking. A reader cannot close it at all.

The renewable lines have a similar shape. Annex II separates total renewable energy from Guarantees of Origin, from power-purchase agreements and from on-site generation, and REF divides the total by total data-centre energy. A retired certificate and a roof full of panels enter the same numerator by different doors, and the numerator is a total the operator assembles.

Whose IT energy is it?

The duty sits on the reporting data-centre operator. On my reading, the act defines that operator by type — enterprise, colocation, co-hosting — and describes such operators as the persons managing the centre and, depending on type, its building, customer-facing space, power, cooling, networks, servers, storage, software or services. The word duty-holder does not appear in that reading at all.

In a colocation hall the boundary gets interesting fast. The operator runs the cooling and the power chain; much of the IT energy inside E_IT belongs to customers who buy space. The reading I have does not say how the ratios treat tenant equipment, and I will not invent a rule for it. What a reader should carry away is that two operators of identically sized halls could draw their reporting boundaries differently and both be describing their own centre accurately.

Ten years of records, and no auditor

Nothing in the reading I have requires an independent verification, an audit or a certification of a reported figure. What remains is the ten-year retention of measurement points and devices: the operator's own file, kept long enough to outlast several generations of hardware and most tenancies.

That is not nothing. It is the one part of this regime a counterparty could ever ask to see. It is also, by construction, a record the operator makes about itself — the same party that chose the figure chooses what evidences it.

The grid indicator points the other way

Two indicators touch the electricity system, and both describe what the centre gives rather than what it needs: whether the centre provides services supporting grid stability, reliability or resilience, and the average battery capacity it offers to the grid. Nothing in the entrance test requires a connection agreement, an allocated capacity or a firmness commitment. A site can stand inside this instrument on 500 kW of nameplate in the computer room while still arguing with a transmission operator about whether it will ever be energised.

The half I could not read, and what it would decide

I read the delegated act, not the directive that authorises it. Article 12 of Directive (EU) 2023/1791 was not obtained for this piece, so nothing here quotes or characterises the empowering provision, its transposition deadline, or any start date written there. Articles 4, 5 and 6 of the regulation — the database and publication clauses — were not read either, so this article says nothing about what the Commission or Member States publish, in what form, at what granularity, or how individual returns are handled. Whether the aggregate that eventually appears names facilities or only countries is settled in clauses I have not seen.

What I can say about the text I hold: EUR-Lex shows the act as having been changed, with a consolidated version dated 17 May 2024; the authentic Official Journal PDF was not opened; and the Annex reading I have states that it condensed the detailed CEN/CENELEC methodology clauses, which means the measurement rules a filer must actually follow live in documents this piece never touched. No volume of reported data, no operator count, no facility count and no aggregate efficiency figure appears here, because none was read; and no named operator is accused of anything.

What the first filer has bought

A mandatory return with self-declared inputs is not a scandal. It is a baseline. The early filings establish the numbers that later benchmarks, comparisons and — if they ever arrive — verification layers get built on. Whoever files a rated figure in year one has defined the denominator of an argument for the following decade of records.

So treat a facility-level PUE attributed to this regime as a claim with a name on it. Ask which inputs were measured and which were rated. Ask where the operator drew the boundary around its computer-room floor. Ask for the measurement-point record, which is supposed to exist for ten years and is the only thing here that could ever contradict the number on the brochure.

For the operators themselves the choice is front-loaded: measured or rated, weighted or point-in-time, tenant energy inside or outside the ratio. Each option is defensible on the text I hold. Together they decide what the first comparable European data-centre dataset will mean — and not one of them is visible in the scalar the market will quote.

The Commission's own entry sits here: European Commission.