EU Data Centre PUE 2025: What the First Figures Show

Key takeaways

• The Commission’s public dashboard shows a weighted PUE of 1.27 for 773 data centres in 11 countries (2025 reporting period, last updated 25 September 2026). It averages the sites that reported. It is not a count of all EU data centres.
• Size matters most in the table. Very large sites (over 10 MW of IT power) average 1.19. Small sites (500 to 1,000 kW) average 1.60, worse than the very small group at 1.52.
• Germany supplies 454 of the 773 sites (58.7%) but about 33% of the reported energy. Germany and the Netherlands together report about 61% of it.
• The 2025 figures do not continue a 2024 series. The 2024 dashboard view has 829 sites in 22 countries, and the Commission’s own report of 21 September 2026 quotes 1.36 for a different data set.
• To benchmark a site, use your own PUE for the same period and compare it only with your own size category. Read the gap as a rough position. The dashboard publishes averages, not spreads.

The first number most operators will see is 1.27. That is the weighted average power usage effectiveness (PUE) on the European Commission’s public data-centre dashboard for the 2025 reporting period. PUE is total data-centre energy divided by the energy used by the IT equipment alone. A PUE of 1.27 means 27 kWh of cooling, power conversion and other overhead for every 100 kWh the servers use.

The figure comes from reports under Article 12 of Directive (EU) 2023/1791, which applies to data centres with at least 500 kW of installed IT power. Delegated Regulation (EU) 2024/1364 sets the indicators. This article reads the dashboard tables, shows two charts, and says what a site operator can and cannot conclude from them. Our EED reporting guide covers the reporting duty itself.

What the 2025 dashboard covers

The 2025 view lists 773 data centres in 11 countries. The dashboard’s country filter offers 23 country codes, and the EU has 27 member states, so most countries are missing. The overview shows 20.7 TWh of energy use, 4.4 GW of installed IT power and 8,752,919 m3 of water input. The nine countries with published energy figures add up to 20.68 TWh and 4.37 GW, which matches those totals. If those 20.7 TWh cover a full year, the average draw is about 2.4 GW (20.68 TWh divided by 8,760 hours).

The dashboard’s About sheet says it shows aggregated data from the data centres that fulfilled their reporting obligation. The numbers describe those sites. They do not describe the EU stock.

The other three indicators are ratios in the same style. Water usage effectiveness (WUE) is water input divided by IT energy in megawatt-hours, so the headline 0.51 means 0.51 m3 per MWh, the same as 0.51 litres per kWh. The renewable energy factor (REF) is the renewable share of total energy, 0.84 in the headline. The energy reuse factor (ERF) is the share of energy sent out as reused heat, 0.0236 in the headline, or 2.4%. REF counts renewable energy bought through guarantees of origin and power purchase agreements as well as on-site generation, so 84% does not measure what physically flows into the building.

Size category: the larger the site, the lower the PUE, with one exception

The five categories follow the installed IT power bands in Annex IV of the regulation. The dashboard table gives these 2025 averages: very small 1.52 (143 sites), small 1.60 (178), medium 1.44 (185), large 1.40 (177) and very large 1.19 (90). From small to very large, weighted PUE falls in every step.

The exception is the very small group. Its average, 1.52, is lower than the small group’s 1.60. The table has no spread, so it cannot say whether a few poor sites lift the small average or whether most small sites are poor. The very small group covers 100 to 500 kW, which is below the 500 kW reporting threshold, and the dashboard does not say why those 143 sites report.

Annex IV says the average is weighted by total energy consumption. A site that uses 100 GWh counts ten times as much as a site that uses 10 GWh. That is why the headline sits close to the very large figure. Each very large site has more than 10 MW of IT power, so the 90 sites hold more than 900 MW. The other four categories can hold at most 71.5 + 178 + 370 + 1,770 = 2,389.5 MW, using each band’s upper limit. Out of 4,371 MW, the very large sites therefore hold at least 1,981 MW, or about 45%. This bound covers installed IT power, not energy. Energy per unit of IT power varies by country, from about 3.0 TWh per GW in Luxembourg to about 6.5 TWh per GW in Italy, so the energy share of the very large group can differ from its power share. Ninety sites, 12% of the list, set the headline.

This article gives no unweighted mean, because the table does not show the energy behind each category. The simple average of the five category values is not the headline and should not be quoted as one.

Country mix: Germany is most of the sites, not most of the energy

Germany reports 454 of the 773 sites, or 58.7%. It is tempting to read the 1.27 as a German number. The energy column says otherwise. Germany reports 6.86 TWh, about 33% of the 20.68 TWh with published values, at a weighted PUE of 1.36. The Netherlands reports 80 sites but 5.68 TWh (27%) at 1.22. Together the two countries report about 61% of the energy.

Average site size differs widely. A German site in the table uses about 15 GWh. A Dutch site uses about 71 GWh. A Belgian site, from 1.99 TWh across 19 sites, uses about 104 GWh. Country PUE runs from 1.14 in Belgium to 1.44 in Austria, with Denmark at 1.17, Sweden at 1.23, Italy at 1.43 and Germany at 1.36. Weighting those rounded country values by energy gives about 1.28, which agrees with the dashboard’s 1.27 within rounding. Without Germany, the same sum gives about 1.23. Both are recalculations from rounded inputs, so treat them as approximate.

Cyprus (2 sites) and Malta (1 site) show “n/a” for every indicator. The dashboard gives no note on why. Separately, Annex IV allows type and size-category averages only where at least three data centres contribute, which fits the type row with 2 sites that also shows “n/a”.

Type table: read the labels before the numbers

The type table has ten rows. Each label combines a business model (colocation, enterprise or co-hosting) with a building layout (a single structure or a group of structures), sometimes with extra services. The largest rows are colocation in a single structure (282 sites, PUE 1.35), enterprise in a single structure (257 sites, 1.49) and co-hosting in a single structure (59 sites, 1.22). Enterprise groups of structures show 1.13 from 35 sites. Merging rows would need the energy behind each, which the dashboard does not show, so the rows stay separate here. Use the table to find your own row, not to rank business models.

What the figures do not show

• No like-for-like year. The 2024 view lists 829 sites in 22 countries. Its very small group averages 1.72 and its very large group 1.20. The drop in the very small figure to 1.52 in 2025 cannot be separated from the change in which countries report.
• A different 1.36. The Commission’s report of 21 September 2026 counts 770 sites for the first reporting period (May to September 2024) and quotes an EU average PUE of 1.36. It uses the 681 sites with PUE values it judged reliable. The dashboard shows 829 sites for 2024, and it does not explain the gap to 770.
• Averages only. Every cell is one weighted average. No minimum, maximum or median appears, so the table cannot tell you how many sites beat 1.27.
• Reported data. Operators submit the values. The dashboard does not describe checks on them.
• Figures can change. The last update is 25 September 2026, and later submissions or corrections may move the averages.

How to compare your own site

Use the definitions the regulation uses. Follow these steps.

1. Find your category from installed IT power: up to 500 kW very small, 500 to 1,000 kW small, 1 to 2 MW medium, 2 to 10 MW large, above 10 MW very large.
2. Calculate PUE for the same reporting period as the dashboard view. Total energy (E_DC) is measured at the input of the data-centre system, before the supply transfer switchgear, and includes fuels and other energy used for cooling. IT energy (E_IT) is measured at the uninterruptible power supplies that feed the IT equipment, or at the power distribution units where a site has no such supply.
3. Put your number next to the category average, not the EU headline. A 3 MW site is in the large category, where the average is 1.40.
4. Read the gap as a position, not a rank. Take a hypothetical site with 13.14 GWh of total energy and 9.0 GWh of IT energy. Its PUE is 1.46, which is 0.06 above the large-category average. The average is energy-weighted, so the biggest sites in that category shape it, and the typical site may sit somewhere else.

If your PUE sits well above the category figure, airflow management is one place to look. Our explainer on cold and hot aisle containment covers the trade-offs. The Commission is also consulting on possible minimum performance standards until 14 December, and our consultation guide explains what to submit. Our regulation tracker shows where each EU and national rule stands.

Annex IV lists the public information: counts, totals and averages. A spread of PUE inside each size category is not on that list. A benchmark built from the dashboard therefore stops at the average. A site that wants to know how it ranks needs a level of data the regulation does not yet publish.