Data centre waste heat: above 1 MW it has to be reused. How much leaves the boundary?

The data centre you are designing is above one megawatt, and the specification says waste heat recovery. Your supplier did not object, and had no reason to: in the worked case published in the appendix to the Recommendation that sets the indicators, the same plant scores 1.24 on one metric and 2.26 per cent on another. Both figures are correct, and both describe the same machine. The number that will end up in the European report is neither of the two, and it is measured at a point your plant drawing has not settled yet.

Above a megawatt, reuse stops being good practice

Article 26(6) of Directive (EU) 2023/1791 leaves no room: “Member States shall ensure that data centres with a total rated energy input exceeding 1 MW utilise the waste heat or other waste heat recovery applications unless they can show that it is not technically or economically feasible in accordance with the assessment referred to in paragraph 7.” The threshold is the total rated energy input, not the 500 kW of installed information technology power demand that triggers the annual report.

Article 26(7)(d) gives shape to the contrary demonstration: an installation level cost-benefit analysis in accordance with Annex XI, applying “where the following installations are newly planned or substantially refurbished”. The second subparagraph reaches straight into the mechanical design: “The analysis referred to in the first subparagraph, point (d), shall consider cooling system solutions that allow removing or capturing the waste heat at useful temperature level with minimal ancillary energy inputs.” The transposition deadline for Article 26(4) to (14) was 11 October 2025; Directive (EU) 2024/1788, which amends Directive (EU) 2023/1791 as from 5 August 2026, leaves Article 26 untouched.

Four indicators, four different denominators

Recommendation ITU-T L.1328 (09/2025), Specification for waste heat reuse in telecommunication rooms and data centres, edition 1.0, approved on 22 September 2025 by Study Group 5 and freely downloadable, defines four indicators besides PUE in twenty-two pages. No two of them share a denominator.

The energy reuse factor (ERF) comes from ISO/IEC 30134-6:2021 — a paid standard, of which the Recommendation cites only title and edition — and is the energy reused outside the data centre over total consumption, with an explicit warning attached: “Any equipment outside of the data centre boundary for increasing the temperature delivered, such as heat pumps, shall not be included in the calculation.” The recovery factor (RF) is the heat emitted by the cooling systems over the heat generated by the IT equipment: the potential, not the result. The waste heat reuse rate (WHRr) is the heat actually reused over the reusable heat. The overall energy efficiency of reuse (WHRee) is the heat reused over the electricity consumed by the recovery system: in practice, the coefficient of performance of the heat pump.

Ten points, and one that appears in no formula

Clause 11 sets ten measurement points, M1 to M10, and a horizon: “the time intervals for measuring electricity and heat used in the parameter calculations are set as 12 months” — twelve months, not a week of trials. Heat is expressed in gigajoules and energy in kilowatt-hours: as with the PUE measurement points, the wrong unit shifts the ratio by an order of magnitude.

Then there is M8: “Return heat from waste heat recovery systems to cooling systems [GJ]”, the heat the recovery system sends back into cooling. Table 1 requires it to be measured; it never appears in any formula in clause 12. And it is exactly the quantity that Commission Delegated Regulation (EU) 2024/1364 requires to be taken out: “If part of the waste heat is reused for cooling the data centre, that part must be subtracted from the reused waste heat, that is to say, subtracting the share of flow rate of cooling fluid used in the data centre”. The same annex also fixes the point: “Reused energy shall be measured at the boundary of the data centre at the point where the energy provided is handed off to be used by the other party.”

There is more: clause 12 calculates ERF as M7 divided by the sum of M1 and M2, where M7 is the heat leaving the cooling systems, not the heat handed over to anyone. The Regulation calculates the same ERF as E_REUSE over E_DC, measured at the boundary and net of the return. Same acronym, different numerators.

Twelve megawatts, and 2.26 per cent

Appendix I — which does not form an integral part of the Recommendation — describes a data centre of 25 000 m², 3 000 racks and 12 000 kW of IT load, in a continental climate. In 2023 it consumed 120 800 000 kW·h, of which 93 000 000 were drawn by the IT equipment; the recovered heat is chilled water at 15/21 °C, which two heat pumps lift to 55/50 °C to warm 23 000 m² of offices: 7 650 GJ over the year.

Hence the three figures: RF 2.26 per cent, WHRr 1.24, WHRee 5.25. The reuse rate exceeds unity because the heat handed over is the sum of the heat recovered and the electricity of the pump, 7 650 GJ plus 1 800 GJ — energy that ERF does not count when the pump sits outside the boundary. Two details show up only once the document is open: the text gives M9 as 9 000 GJ while the calculation just below uses 9 450 GJ, and 7 650 GJ become 2 100 000 kW·h where the exact conversion gives 2 125 000.

What to write into the specification

Which indicator, by name and by formula: RF, WHRr, WHRee or ERF, with numerator and denominator written out. A performance figure without a denominator is not a threshold.
The data centre boundary, drawn on a layout: it is the line that decides how much heat actually leaves; without it, E_REUSE cannot be verified by anyone.
The return to cooling, metered separately: the meter on M8 has to be planned even though no formula in the Recommendation uses it, because the Regulation subtracts it.
Units and period: gigajoules, kilowatt-hours, twelve consecutive months; average waste heat temperature taken at the heat recovery exchanger.
The condition of the plant before the measurement: “pipelines, valves, pumps and other equipment is intact, to ensure that there is no air leakage, water leakage and other phenomena” — no air or water leaks anywhere. It is a line for the acceptance record, as with thermal commissioning.

What we do not know

We have not verified on a primary source whether and by which act Italy has transposed Article 26: the deadline of 11 October 2025 has passed, and Legislative Decree 136 of 3 July 2026, which does cite Directive (EU) 2023/1791, transposes the gas market directive instead. Of ISO/IEC 30134-6:2021 and EN 50600-4-6 we report only the references cited by the freely available documents: they are paid standards and we have not read their text. The case in Appendix I is a non-binding example set in a cold climate.

The point

Heat recovery is not bought on a performance figure: it is bought on an indicator, a boundary and a measurement point. Skip them at the design stage and you end up with a plant that works and a report that does not hold. This is the work we bring into a compliance specification: the indicator named, the boundary drawn on the layout, the meters on M7, M8, M9 and M10 planned from the start, and the verification carried out on acceptance, with the test report handed over together with the plant. Flow rates, supply and return temperatures, pump consumption and IT load then stop being separate sheets and become a single map of the site, on which an AI flags the gap between the heat recovered and the heat actually handed over — inside the client perimeter, on-premise or in a dedicated cloud with data centres in Italy, together with CSIDIA, the other company in the group. For a data centre, our own crews, no subcontracting.

Are you designing a heat recovery scheme above one megawatt, or do you have to report the reused heat of a plant already running? Talk to one of our engineers: the site visit is free, and the measurement point is settled before the pipework is ordered.