European Energy Efficiency Directive: new obligations for data centers that are necessary but…not (yet) sufficient!

Written by Boavizta on 11.04.2025.

The European Energy Efficiency Directive (EED) introduced in 2012 different measures aiming at achieving the European Union’s targets for energy efficiency and climate change mitigation. For example, it has introduced mandatory energy audits and energy management systems for certain stakeholders, some requirements for district heating and cooling systems (DHC), ‘energy saving’ obligations for energy suppliers, and measures to better inform final consumers.

After the introduction of partial amendments in 2018 (see details here), a completely revised version came into force in October 2023 after almost 2 years of negotiation (see details here). This new version introduces new obligations for certain owners and managers of data centers, while the Delegated Act adopted on 14 March 2024 provides additional reporting of sustainability and performance indicators (see details in Appendix).

By requiring the reporting of different indicators and data, the Delegated Act to the Energy Efficiency Directive (EED) should allow the European Commission to have access to more detailed information on the energy consumption of data centers, their potential benefits for better grids management (demand-response, peak-load shaving, frequency regulation…) and for waste heat recovery. However, some clarifications and improvements should be made to make this additional reporting exercise more robust and useful for the effective implementation of more efficient and frugal practices. First and foremost, the reporting methods should be more in line with the physical and temporal reality of the systemic and local phenomenon at stake (direct and induced), and should be based on more granular and transparent data. The practices for energy purchase contracting and guarantees of origin should be strengthened and ‘sanitized’ to avoid misleading or even counterproductive effects. Even though this directive has historically been focused on energy issues, it would be necessary for it to rely on a multi-criterion and life cycle approach to avoid transfer of negative environmental impacts. To do this, the requirement to use inventory data based on robust methods and reliable, transparent, representative of changes in practices and audited by qualified independent third parties would be paramount. Also, in view of the significant uncertainties and variations on some key parameters, sensitivity analyses would be necessary for fully informed long-term decision-making. In order to maximise its effect while reducing the administrative burden for the actors concerned, a better articulation and harmonisation with other European and national regulatory frameworks seems essential. A better articulation and harmonization with other European and national regulatory frameworks would also be necessary to maximise its impacts while reducing the associated administrative burden for concerned stakeholders. Last but not least, beyond reporting, coercive measures should be rapidly implemented to better regulate (with an extra-territorial scope) the non-essential and detrimental uses such as online services that enable fraudulent video generation from generative AI.

Key highlights (1)

Although the annex to the delegated act requests the reporting of information relating to the functions provided to the electricity grid in terms of stability, reliability and resilience of the electricity network (peak shaving, frequency regulation…), the tracking of the consumed electricity and drawn power at a finer time scale (at least hourly) would be necessary for fully informed decision and relevant actions. Furthermore, a granular analysis of the workloads’ power drawn distribution according to their latency-sensitivity would be also useful to identify those that can be balanced to reduce the subscribed maximal power capacity(2) and maximize the renewable energy consumption while better optimizing hardware resources and overall costs, especially when stationary batteries are used to store surplus of renewable electricity (3).
Even if the indicators such as total energy consumption and total energy consumption of information technology equipment (EIT) should provide better visibility on the energy used to power IT equipment compared to the total energy consumed by all data centers, they will not allow to identify and improve all the actions that should be implemented to improve the efficiency and frugality, including in terms of IT architecture & layout, and sizing of electricity conversion and distribution infrastructures (see figure 1).
Even if the act has the merit of introducing finer indicators taking into account some data centers’ components such as Central Processing Units (CPUs), it should go even further by considering the specific impact of graphics processing units (GPUs) and specialized integrated circuits (ASICs) such as tensor processing units (TPUs) whose use is skyrocketing for the training and inference of models, especially those used for deep learning and generative AI. Indeed, the CSERV indicator cited in the Delegated Act is based on the SERT(4) “active performance” analyses that consider only CPUs, storage and memory worklets (in its latest version dating from 2022). It is also regrettable that they provide exemptions for HPC servers and servers with Auxiliary Processing Accelerator while these types of servers now consume a lot of electricity because of their high processing capacity (and high cooling requirements).
By distinguishing the Guarantees of Origin (GO) linked ot not to Power Purchase Agreements (PPAs), the Delegated Act should (partly) unveil some dubious practices that minimize their impact by purchasing contracts that have counterproductive effects on the climate(5)(6)(7). Nevertheless, the act should have introduced robust criteria to strengthen the definition of GO and PPA especially by considering environmental attributes that allow to factor in the induced impacts over the system’s life cycle and from data more in line with the temporal and physical reality(8). Indeed, any power plant caused environmental impacts(9) and the environmental attributes of the electricity that transits through the electrical grids fluctuate all the time according to the activation of the means of production connected to them and activated (or deactivated). Beyond the necessity to revise the merit order approach(10) to better factor in the negative externalities (financial and environmental) of the current electrical system(11) it would be necessary to introduce a robust certification system that guarantees with transparency the environmental impacts caused by the electricity supplied to a data center (and ideally consumed(12)) at least at an hourly rate, and ideally at the nodal level (13)(14)(15)(16).

Note 1: Google has introduced in 2021 ‘Time-based Energy Attribute Certificates’ (T-EACs) to hourly match the ‘carbon-free’ electricity supplied in grids with the electrical consumption of its data centers at the regional level(17)(18). Although this initiative demonstrates the benefits and the technical feasibility of a time-stamped and regionalized approach, it could be improved by taking into account the impacts induced at the systemic and nodal levels over the life cycle, including the embodied emissions such as proposed by the Carbon Explorer framework ⁴.

Note 2: The Emissions First Partnership (EFP) approach supported by Amazon and Meta in particular, is much more disputable because it is based on the counterfactual analysis of the presumed avoided emissions. Moreover, the short-term approach of the method do not take into account the long term effects that can decarbonize the electrical grids over their life cycle(19). The approach based on the hourly marginal emissions rates supported by the US non profit organization WattTime(20) and the company REsurety faces the same limits even if a very few details are given on the method, data source and hypotheses taken to elaborate the rates that have been released in opensource(21) except that they would be compatible with the current rules of the GHG Protocol for the evaluation of the ‘operating margin’ emissions(22)…

Note 3: the GHG Protocol guide for quantifying GHG emission reductions from projects connected to the electricity network(23) (2007 version) is based on ‘simplified methods’ and recommends the use of more complex models to examine in an integrated manner the effect of a new project on the electricity grid. In addition, the GHG Protocol’s guidances for the reporting of GHG emissions of purchased electricity (also called ‘scope 2’)(24) according to the “location-based” approach must be based on the “average energy generation emission factors for defined geographic locations, including local, subnational OR national boundaries” ; whereas the ‘market-based’ approach should factor in the GHG emissions emitted by the ‘generators’(25) from which the reporter contractually purchases electricity. So it does not systematically consider the physical and economical reality of the phenomenon involved in the electricity supply such as the temporal & geographical variations of the GHG content of the electricity mix (incl. within a country or a region). We hope that the ongoing revision process will tackle these issues.

Comparison of guidelines provided by the GHG Protocol for the reporting of emissions associated to scope 2 (market-based vs location-based methods)

By requiring the reporting of quantities of wasted heat and its temperature, the delegated act should have allowed them to assess potential heat recovery. Nevertheless, the reporting of annual average values do not allow to qualify these recovery potentials with accuracy. Furthermore, as the delegated act allows for the deduction of the amount of heat recovered on site for cooling the data centre, it would be useful to ask for details about the energy performances and environmental impacts caused by the processes used for this recovery.

Note 1: the German Federal Energy Efficiency Act (EnEfG), which came into force in November 2023, has already introduced new obligations for companies that manage data centers, especially in terms of waste heat recovery. For example, all companies with total final energy consumption of more than 2.5 GWh had until 1 January 2025 to report waste heat quantities on an online platform. An information sheet has been released to describe the different obligations and fines to be imposed on companies that do not comply with their obligations.

Note 2 : in France, the n°2021-1485 act obliges datacenters to recover the waste heat they generate, in particular through district heating and cooling networks, OR comply with a ‘indicator respectent un “numerical indicator determined by decree on a multi-annual horizon in terms of efficiency in the use of power”…
Even though the 2019 Eco-design directive (cited by the delegated act for the assessment of IT equipment capacity) and the Eco-design for sustainable products regulation of 2021 are supposed to set additional requirements for the assessment of the impact of servers on their life cycle (beyond energy aspects and water consumption during operations), these regulations should be more ambitious to take into account key environmental impacts along the life cycle (depletion of resources, land use and loss of biodiversity, GHG emissions…). To this end, the obligation to use common public inventory databases based on robust methods, and reliable and transparent data that reflect the current practices and that are audited by independent third parties would be paramount. Furthermore, in view of the significant variations and uncertainties affecting certain key parameters (inventory data, RES production potential, network losses, O&M…), sensitivity analyses should be requested for well informed decision-making.
Furthermore, even though the 2023 EED requires member states to apply sanctions that are “effective, proportionate and dissuasive” In the case of infringements, it is questionable whether all the Member States will implement enforcement measures and engage the necessary financial and operational resources (and willingness) to carry out controls and check all data reported by the data centers’ managers and owners.
Finally, beyond the need to evolve certain indicators and ensure better harmonization between different regulations (to ease their adoption), it would be necessary to evolve the European regulation on artificial intelligence (AI) to better evaluate and regulate the impacts caused by the use of artificial intelligences with an extraterritorial scope. An incentive to greater transparency and rigor from the actors of the entire value chain are also paramount to make the assessment of impacts more robust and to allow comparisons between stakeholders.

Appendix- Details of recent measures that target data centers

Companies with an average annual energy consumption of more than 85 TJ in the past three years, are obliged to implement an energy management system (EMS) certified by an independent body in accordance with relevant European or international standards (no later than 11 October 2027);
Companies with an average annual energy consumption of more than 10 TJ in the last three years (and not subject to the obligation to implement EMS), must achieve an energy audit (by 11 October 2026 at the latest). This audit should be carried out in an independent and cost-effective manner by qualified or accredited experts or implemented and supervised by independent authorities under national law.
Sites with an installed IT power demand of at least 500 kW, must publish information described in Annex VII.

Note: except for information subject to Union law and national law protecting trade and business secrets and confidentiality, and information relating to defense and civil protection.

Information that was expected to be made publicly available before the Delegated Act:
The name of the data center, the name of the owner and operators of the data center, the date on which the data center commenced operations and the municipality in which the data center is established.
The data center floor area, installed capacity, annual volume of data input and output, and the volume of data stored and processed within the data center.
The “performance” of the data center during the last full calendar year including its energy consumption, power use, temperature settings, fatal heat usage, water consumption and the use of renewable energies, based, where applicable, on the standard CEN/CENELEC EN 50600-4 “Information Technology — Data Centre Facilities and Infrastructures” (information to be replaced by the sustainability and key performance indicators introduced in March 2024 by the delegated act -see details below).

Note: Article 32 of the 2023 EED requires Member States to determine a regime for sanctions applicable to violations of national provisions adopted and take the necessary measures to ensure the implementation of these sanctions (by 11 October 2025). These sanctions must be effective, proportionate and dissuasive.

New indicators that are expected to be reported by the Delegated Act:

As announced, the Delegated Act adopted on 14 March 2024 requires the reporting of sustainability indicators and performance indicators:

Annex I details the information to be reported in the European database on data centers;
Annex II, which specifies the key performance indicators to be followed and the associated measurement methodologies;
Annex III describes the sustainability indicators and associated calculation methods.

General information to be reported on October 11th, 2025 the latest (see Annex I of the Delegated Act of March 2024):

N°		Description
1	a	Data center name used to identify and describe the reporting data center
	b	Owner and operator of the data center including the name and contact details
	c	Location i.e. Local Administrative Unit Code (LAU code) of the location of the reporting data center (building or site) in accordance with the most recent LAU tables of Eurostat
	d	The type of data center: enterprise data center’, ‘colocation data center’ or ‘co-hosting data center’, combined with one of the values ‘structure’ or ‘group of structures’ If a colocation data center also offers co-hosting services or if a co-hosting data center also offers colocation services, this shall be indicated.
	e	Year and month of entry into operation
2	a	Electrical infrastructure redundancy level at high voltage level / at low voltage level (line-up) / at rack level
2	b	Cooling infrastructure redundancy level at room level / at rack level. For the redundancy levels, if “N” represents the baseline number of components or functions to satisfy the normal conditions, redundancy shall be expressed compared to that baseline “N”, for example as “N+1,” “N+2,” “2N”, etc. Facility redundancy can apply to an entire site (backup site), systems or components. Information technology redundancy can apply to hardware and software

Key performance indicators to be reported (cf. Annex II of the Delegated Act of March 2024):

N°		Description
Energy & sustainability indicators	a	“Installed information technology power demand” (“PDIT” in kW) NB: where the installed information technology power demand has changed during the reporting period, a weighted average shall be used.
	b	Data center total floor area (“SDC” in m²).
	c	Data center computer room floor area (“SCR” in m²).
	d	Total energy consumption (“EDC” in kWh) by using the methodology in the CEN/CENELEC EN 50600-4-2 standard or equivalent. The amount of EDC coming from backup generators (EDC-BG in kWh) shall be measured separately. NB: the measurement points shall be set at the primary & secondary supply of energy & at every additional supply
	e	Total energy consumption of information technology equipment (“EIT” in kWh) shall be measured in accordance with the category 1 methodology for the calculation of the PUE set out in the CEN/CENELEC EN 50600-4-2 standard or eq.
	f	Electrical grid functions is the information on whether any functions that support the stability, reliability, and resilience of the electrical grid are provided by the data center, such as peak demand shifting or firm frequency response (FFR)
	g	Average battery capacity (“CBtG” in kW) is the average capacity of the DC batteries that were offered to the grid via a “relevant market” or “contracts for elec. grid functions
	h	Total water input (“WIN” in m³) as defined by CEN/CENELEC EN 50600-4-9 standard WUE Category 2, or if not possible, the methodology set out in Cat. 1 or eq std.
	i	Total potable water input (“WIN-POT” in m³)
	j	Waste heat reused (“EREUSE” in kWh) following CEN/CENELEC EN 50600-4-6 std or eq. NB: if part of the waste heat is reused for cooling the DC that part must be subtracted from the reused waste heat
	k	Average waste heat temperature (“TWH” in C°) shall be measured as the temperature of the fluid used to cool the ICT equipment in the DC computer room averaged over the year, and across every measurement points.
	l	Average setpoint information technology equipment intake air temperature (“TIN” in C°) measured as the average setpoint temperature in all DC rooms, set as a setpoint command to the cooling system used for the ICT equipment averaged over the year
	m	Types of refrigerants used in the cooling and air conditioning equipment
	n	Cooling degree days (“CDD” in degree-days) for the location of the reporting DC during the last calendar year, by using the methodology used by Eurostat and the JRC or equ. and with a base temperature of 21 C° NB: open access sources shall be used to determine the CDD

The total energy consumption of information technology equipment “E_IT” shall be measured in accordance with the Category 1 methodology for calculating PUE (Power Usage Effectiveness) set out in the CEN/CENELEC EN 50600-4-2 standard or its equivalent. The annual power consumption should be combined with each uninterruptible power system (UPS) connected to the data center computer equipment.

For data centers that do not have an inverter, the E_IT’s power consumption can be measured at the power distribution unit (PDU) connected to the data center computer equipment, either in accordance with the Category 2 methodology for calculating PUE as set out in CEN/CENELEC EN 50600-4-2, or at a measurement point specified by data centers.

Figure 3 Positioning of the point for measuring the temperature of the fatal heat

N°		Description
1 Energy & sustainability indicators	o	Total renewable energy consumption (“ERES-TOT” in kWh) determined according the methodology set out in the CEN/CENELEC EN 50600-4-3 std or eq.
	p	Total renewable energy consumption from Guarantees of Origin (“ERES-GOO” in kWh) NB: GO cannot be counted for more than one DC or be created from PPA or on-site renewables
	q	Total renewable energy consumption from Power Purchasing Agreements (“ERES- PPA” in kWh) NB: any GO created as a result of such PPA must be included in ERES-PPA. Otherwise, the concerned amount of energy shall be subtracted from the measured ERES-PPA.
	r	Total renewable energy consumption from on-site renewables (“ERES-OS” in kW) NB: any GO created as a result of on-site Ren sources must be owned and retired by the reporting DC. Otherwise, the amount of energy in question shall be subtracted from the measured ERES-OS.

Indicators related to IT equipment capacity to be reported (see Annex II):

N°

Description

ICT capacity indicators

ICT capacity for servers (“CSERV”) = the sum of the SERT active state performance or eq for all servers as declared in the manufacturer information in accordance with Regulation 2019/424

The active state performance value for the configured server or group of servers in a DC computer room shall be either interpolated from the declared active state performance value for a configuration declared under the Regulation (EU) 2019/424, or provided by a server manufacturer, or provided by a table of values for CPU part numbers created from a large SERT dataset, or estimated from a large dataset of measured values where a “recognized” calculation method exists.

ICT capacity for storage equipment (“CSTOR” in petabytes)

shall be the storage capacity, namely the sum of the raw (addressable) capacity of all SSD and HDD storage devices installed in all the storage equipment as declared by the storage device manufacturer.

NB: colocation data centre operators may calculate CSTOR by extrapolating the value that corresponds to at least 90% of the installed information technology power demand of all new storage equipment installed in the reporting data centre.

The IT equipment capacity indicators will need to be measured for servers and data storage equipment as defined in the Regulation (EU) 2019/4244 which defines the ecodesign requirements for servers and data storage products (not Regulation 2021/341 amending part of Directive 2019/4244…).

Data traffic indicators to be reported (see Annex II):

The delegated act allows data center operators to measure these indicators from any source or combination of «sufficiently reliable available» data sources, including data measured directly by the operator, data reported by data center customers, or data provided by telecommunications operators and service providers:

N°		Description
3 Data traffic indicators	a	Incoming traffic bandwidth (“BIN” in gigabytes per second) shall be measured as the total provisioned bandwidth for incoming traffic to the data centre compute.
	b	Outgoing traffic bandwidth (“BOUT” in Gbs) shall be measured as the total provisioned bandwidth for outgoing traffic from the data centre computer room, aggregated for all the connectivity capacity, and averaged over the year
	c	Incoming data traffic (“TIN”, in exabytes) shall be measured as the total incoming data to the data centre computer room, aggregated over the course of the reporting year, irrespective of the number of the data centre’s connections.
	d	Outgoing data traffic (“TOUT”, in exabytes) shall be measured as the total outgoing data from the data centre computer room, aggregated over the course of the reporting year, irrespective of the number of the data centre’s connections.

Information to be made public

Several pieces of information will be made public but in an aggregated manner (cf. table below). Moreover, Article 12 of Directive 2023/1791 specifies that information subject to EU law and national law protecting commercial and trade secrets and confidentiality shall not be made public.

Note: performance indicators will be treated as confidential information.

At the level of each Member State AND at European Union level
i	Number of reporting data centers
ii	Distribution of reporting data centers by size category*
iii	Total installed information technology power demand
iv	Total energy consumption
v	Total water consumption
vi**	Average PUE, Average PUE, average PUE per type of data centre, and average PUE per size category*
vii**	Average WUE, average WUE per type of data centre, and average WUE per size category*
viii**	Average ERF, average ERF per type of data centre, and average ERF per size category*
ix**	Average REF, average REF per type of data centre, and average REF per size category*

Annex IV of the delegated act defines 5 categories of data centers to be used for data aggregation:

Very small data center 100-500 kW
Small data center 500-1000 kW
Medium size data center
Grand data center 2 – 10 MW
Very large data center > 10 MW

** The so-called sustainability indicators will have to be aggregated in a weighted manner using energy consumption as a weighting factor.

Notes

1 :
Non-exhaustive list

2 :

Varun Sakalkar et al., Data Center Power Oversubscription with a Medium Voltage Power Plane and Priority-Aware Capping, 2022

3 :
Bilge Acun et al., Carbon Explorer: A Holistic Approach for Designing Carbon Aware Datacenters, v3 2023

4 :

SPEC, The SERT® Suite Design Document 2.0.x, 2022

5 :

Anders Bjørn et al, Renewable energy certificates threaten the integrity of corporate science-based targets, 2022

6 :
MIT Technology Review, Google, Amazon and the problem with Big Tech’s climate claims, July 2024

7 :
Isabel O’Brien for the Guardian, Data center emissions probably 662% higher than big tech claims. Can it keep up the ruse?, September 2024

8 :
including the electricity losses and impacts related to conversion, storage, transmission, distribution of electricity and grids congestion management.

9 :
from the extraction of minerals necessary for its manufacture and operation to its end of life

10 :
ie. en donnant la priorité aux moyens de production dont les coûts marginaux à courts termes sont les plus bas ce qui peut notamment créer des effets contre-productifs en période de pointe si des moyens de production d’énergie fortement carbonés doivent être temporairement activés et aussi engendrer des coûts de transmission et de gestion des congestions supplémentaires.

11 :
Qu’est-ce que la logique de « merit order » et des coûts marginaux de production électrique ?

12 :
In case of self-production and self-consumption for instance or local sharing of energy as allowed by the EU Directive 2024/1711.

13 :
Wilson Ricks and Jesse D. Jenkins, The Influence of Demand-Side Data Granularity on the Efficacy of 24/7 Carbon-Free Electricity Procurement, 2024

14 :
Iegor Riepin, Tom Brown, System-level impacts of 24/7 carbon-free electricity procurement in Europe, 2022

15 :
Eicke and Schittekatte, Fighting the wrong battle? A critical assessment of arguments against nodal electricity prices in the European debate, 2022

16 :
Verhaeghe et al, Système de prix nodaux : expérience américaine et perspectives pour l’Europe, 2018

17 :
https://cloud.google.com/blog/topics/sustainability/t-eacs-offer-new-approach-to-certifying-clean-energy?hl=en

18 :
Qingyu Xu et al, System-level impacts of voluntary carbon-free electricity procurement strategies, 2024

19 :
New Climate Institute, Briefing: 24/7 renewable electricity matching is a far more credible approach for the GHG Protocol and the SBTi than the Emissions First Partnership proposal, Octobre 2024

20 :
https://watttime.org/wp-content/uploads/2023/12/WattTime-AccountingForImpact-202209-vFinal2.pdf

21 :
https://gridemissionsdata.io/

22 :