The EU's Energy Efficiency Directive (EED) gathers data on data center energy use and efficiency across the bloc, some of which is made publicly available. The release of 2024 data has re-emphasized concerns about the quality of this data.

Since the first dataset was published in 2024 (for 2023), there have been significant concerns about reliability, the reporting process and the prescribed reporting formats. The second round of reporting in 2025 (for 2024) was compromised by concerns over confidentiality, highlighting the need to address unresolved EED reporting deficiencies. The Netherlands was the first EU member state to publish its data and did so more transparently than any other EU member state, exposing persistent problems in data quality and the effectiveness of the EED.

The deadline for submitting data for 2025 (May 31, 2026) has passed, but the dataset has yet to be published. The data for 2023 and 2024, as presented on the data analytics platform Qlik Sense, reveals that the EED has missed the majority of the enterprise data centers and around half the colocation facilities in member states that have reported so far (see EED app, Qlik Sense). Moreover, its data center types and size categories limit the insights the dataset could provide, even if reporting was complete.

There are three clear issues: the volume of data reported, the limitations of the EU's classification scheme and the quality of the data itself.

Coverage

Data is being added to the visualization on an ongoing basis. Since early July 2026, data for Austria, Belgium, Bulgaria, Czechia and Estonia has been added, along with additional facility entries for countries that are already reporting. At the time of writing, the visualization shows 2024 data for 22 member states and 2023 data for 20 states.

The five missing member states in 2024 are Ireland, Latvia, Romania, Slovenia, and Slovakia. This is a substantial group that Uptime Institute estimates to account for around 10% of the total number of data centers located in the EU, including the city of Dublin (Ireland), which is one of Europe's top five data center hubs.

Some countries' data has been delayed because they have not transposed the EED into national law. In Ireland, for example, data has been collected but cannot be published until transposition is complete. All EU member states (except the Czech Republic) currently face infringement procedures for failing to completely transpose the EED before the deadline of October 2025.

The detailed data published for the Netherlands is incomplete. The visualization of data from the other 21 nations is less detailed and is evidently also incomplete.

The currently published EED dataset for 22 states reports 829 data centers, with total IT power of 4.9 GW. Both these figures are much lower than the actual total. Compared with two European data sources, the EED data appears to capture only a fraction of the total number of data centers (within the countries it covers), with particularly limited representation of enterprise facilities.

Datacentermap.com. This is a publicly available commercial website, indicating that there are 2,395 data centers in the 22 member states currently listed in EED data (and a total of 2,647 across the 27 EU member states). These figures include colocation, hyperscale, cloud and edge/network data centers with 50 kW or more of IT power. These are mostly public facilities; the website does not attempt to include enterprise data centers, although a small number may be represented. Even excluding enterprise data centers, the website has identified nearly three times as many facilities in the same territory as are reported in the EED data center database.

The European Data Centre Association (EUDCA). EUDCA's annual report, The state of European data centres, includes enterprise facilities and reports a substantially higher total than both Datacentermap.com and the EED. This confirms expectations that Europe hosts a significant number of enterprise facilities. The EUDCA reports 10,477 data centers with 50 kW or more of IT power in the European area (the EU 27 plus Norway, Iceland, Switzerland and the UK), which is a 2024 figure, reported in 2026. EUDCA used a different methodology in 2025, and reported a larger number of data centers for 2024; In 2024, they did not report data center numbers.

EUDCA breaks the data down between enterprise, colocation, hyperscale and scale data centers (wholesale colocation or powered-shell data centers), and provides total numbers for the EU area in 2024 and capacity in 2025 (see Table 1). Although the figures are from different years, they suggest a market divided into four comparably sized groups. A large number of enterprise data centers appear to approximately match the capacity of a small number of hyperscale facilities, and a larger number of scale facilities are somewhat outmatched by the capacity of the hyperscale sector.

The EUDCA figures include data centers as small as 50 kW. No hyperscale or scale data centers, and very few colocation data centers will be smaller than 500 kW, but many of the 9,000 enterprise facilities may fall below this threshold. EUDCA reported total enterprise data center power to be 3,100 MW in 2025, which suggests an approximate average of 300 kW per data center — but the true average may be slightly lower because 2025 capacity is divided by 2024 numbers. Assuming all enterprise data centers are smaller than 1 MW and follow a right-skewed distribution, a conservative assumption would be that 25% are above 500 kW. This implies a total of 2,250 enterprise data centers and 3,798 data centers in total in the European area (27 member states plus Iceland, Norway, Sweden and the UK).

Table 1 Total number and capacity share of data centers across EU (EUDCA report 2025)

The EUDCA report does not break down data center numbers by state. It breaks down IT capacity figures by state, but only for colocation and hyperscale facilities in 2024 (7.4 GW and 5 GW, respectively). Colocation capacity, which is likely to be more representative than hyperscale capacity, can provide approximate scaling factors.

EUDCA's figures also include four non-EU nations (Iceland, Norway, Sweden and the UK), which have a combined colocation capacity of 3 GW. This gives a grand total of 10.4 GW for the whole region, and implies that data centers across the region are approximately 1MW on average.

The EUDCA data reports a combined colocation capacity of 6,743 MW for the 22 member states listed in EED data so far. Using an approximate scaling factor of 6,743/10,400 (0.64), it can be estimated that the EED should have collected data for around 2,430 data centers for those 22 states, nearly triple the number it has actually tracked.

Despite Datacentermap.com's explicit omission of enterprise data centers, the EUDCA and the Datacentermap.com estimates are remarkably close.

The EED's data center categories do not align exactly with EUDCA's, but they can be condensed to enterprise, colocation (all data center types that include colocation services) and cloud (cohosting). EUDCA's hyperscale and scale categories are also combined and the same scale factor applied (see Table 2).

Table 2 EED reported facilities versus scaled EUDCA facilities across 22 EU states

Given the likely mismatches within and between categories and the approximate scaling of the EUDCA numbers, it appears that the EED Is receiving reporting data from most of the hyperscale facilities and possibly half of the colocation data centers. The EED is likely collecting data from one-quarter of enterprise facilities, although this fraction is uncertain due to EUDCA's inclusion of 50 kW facilities.

Classification

Size categories

One notable issue with the EED data lies in its size categories, which are defined in a way that reduces the value of the data. Only one size category covers data centers larger than 10 MW. This range includes modest colocation sites and hyperscale facilities and provides no visibility into size distribution, and introduces uncertainty because no upper limit is specified.

Uptime Intelligence estimated IT power demand for each category by multiplying the number of facilities in each category by the midpoint of that category's IT load range (see Table 3). The midpoint for the very large category was set at 50 MW because this gives a total power (4.8 GW) that is close to the EED's actual total (4.9 GW).

Table 3 Power demand and PUE by data center size category (EED data)

One useful observation emerges from the EED data: large and very large data centers (larger than 2 MW) use 90% of the data center IT power. Smaller data centers consume comparatively less power, meaning that efforts to make them more efficient will deliver only minimal returns (which would still be true even if many more smaller data centers were included in the EED's dataset).

Definitions

The EU's data center definitions are also problematic. The EED classifies data center entries into three broad categories:

• Enterprise. Owned and operated by a single entity for the IT needs of that entity.
• Colocation. Customers install their own equipment in the operator's site.
• Co-hosting. Customers run their workloads on the operator's equipment.

These categories are based on the European standard EN 50600, which Uptime Institute has analyzed (Sustainability: regulations, reporting and facilities management Q&A) and commented on (EED reporting and indicators: a wake-up call). Co-hosting is intended to capture cloud facilities, but Uptime Institute has previously commented that the definition does not accurately reflect the operation of cloud facilities (EED recast: new reporting proposals raise confidentiality concerns).

The definitions are likely to obscure hyperscale data centers. Huge facilities running social media applications, such as Meta's Facebook or WhatsApp, or Internet services, such as Google Search, often only run one or a few applications for their owner/operator. These sites will, or should, be classified as enterprise facilities under the EED definitions, and will be grouped with a large number of smaller, traditional, in-house back office IT facilities run by commercial organizations operating in sectors such as banking, health and retail.

Other examples include Amazon: AWS would clearly be defined as co-hosting, despite its largest customer being Amazon's retail business. This application could also qualify as an enterprise application, unless Amazon and AWS identify as two separate entities. Microsoft and Google offer co-hosting services (Microsoft Azure and Google Cloud) alongside services they run for customers (such as Microsoft Office 365, Microsoft Teams and Google Search, Gmail, YouTube), which are clearly enterprise applications.

Colocation and co-hosting definitions overlap because retail colocation involves a cloud operator (defined as a co-hoster under the EED), installing and operating its own equipment in a colocation facility.

In addition, EN 50600 subdivides each category into a "single structure" if a data center occupies a single building or a "group of structures" if a data center occupies more than one building. The group of structures subcategory is common in EED responses. One of the 58 cohosting-only facilities, around 10% of the colocation-only facilities, and approximately 16% of the enterprise facilities in the dataset are recorded as a group of structures. The proportion is highest (38%) for colocation facilities offering cohosting services.

The EED also allows certain categories to be combined:

• Colocation data center with co-hosting services (presumably a colocation site where the colocation operator also offers cloud services).
• Co-hosting data center with colocation services (presumably a data center primarily used for cloud services, which also offers colocation space).

Without clearer definitions, the potential for overlaps is inevitable, increasing the risk of inconsistent classification and inaccurate reporting.

Data quality

The EED dataset reveals several apparent errors and inconsistencies.

Detailed examination is difficult, but the Netherlands data provides information on individual data centers and means that the aggregated totals in the EED visualization can be checked against the actual figures in the raw data. This comparison could reveal how the EED's types and size categories affected the data collection, but Uptime Intelligence has not attempted a full analysis of the Dutch dataset.

Excess reporting

The EED data includes 126 very small facilities with less than 500 kW of IT power, even though they are not required to report under the directive. The majority of these facilities are in Germany (71), France (21), and the Netherlands (11), with a small number located in Belgium, Denmark, Spain, Greece, Hungary and Luxembourg.

Germany and the Netherlands both operate a national data center reporting scheme that extends to facilities as small as 100 kW, which may account for many of the sub-500 kW facilities appearing in the dataset. Other listings may reflect voluntary reporting, which is permitted under the EED framework.

Inaccuracies

Uptime Institute has examined the detailed per-site data from the Netherlands and identified many instances of incomplete or incorrect reporting (see Incomplete data threatens effectiveness of EED). The rest of the dataset is also unlikely to have been completely validated.

This suggests that published figures for total power, PUE or other metrics at category or sub-category level are likely inaccurate. Some views of the dataset produce obvious errors: individual member states appear, in some cases, to have weighted average PUE values as low as 0.15 or as high as 18.14.

In addition, the visualizer appears to be introducing errors when aggregating data. For example, Finland's national average WUE appears to be a clearly non-feasible 9 million l/kWh (above 2 l/kWh is concerning), even though each of the data center categories in Finland taken separately has a creditably low WUE (0.04 l/kWh or lower).

Energy reuse factor is also poorly represented in the visualization. Finland's 10 colocation data centers report a weighted average energy reuse factor of 48%. This figure would be exceptional for one facility — achieving this performance across 10 facilities as an average is unlikely, without further explanation.