Methodology

Last updated: 13 November 2026

We track operating data centres and planned projects across Europe, and every number traces to a published source or a documented estimate. Each live site gets one capacity estimate: a published IT capacity where one exists, otherwise a conversion from floor area. On top of that, we model the enterprise server rooms that publish nothing – a large volume of very small facilities. Of our Europe-wide estimate of 13 615 MW, 55% is published figures, 25% is converted from floor area and 20% is modelled enterprise capacity.

1. Sources

Every figure is stored with the document it came from, the date we read it and its exact wording. Sources are operator websites and reports, investor filings, building and environmental permits, grid queues, regulators’ registers, and trade press reporting on these. We never use commercial data centre trackers. Where sources disagree, both are kept and the more reliable one is published.

2. Status

A site counts as operational only when a source says it is live, opened or in service. Sites move through announced, land secured, permitted, grid secured, under construction, partially operational and operational, each step backed by evidence. Capacity totals cover live sites only. Many campuses never reach the capacity they advertise, which is a common reason behind the differences you might notice when comparing with other data trackers.

3. Types of capacity

A grid connection is often two to three times the IT load behind it, and operators use the two interchangeably. We put a lot of effort in separating these capacities and being clear on which one is reported for a given site:

TypeMeaningCounted in aggregates
IT capacity (design, operational, available)power for the IT equipment: at full build-out, running today, or on the marketyes
Grid connectionthe site’s electrical connectionyes, as a fallback, and flagged
Requested connectiona grid application not yet grantedno
Unqualifieda bare “MW” figure with no indication of what it measuresno

We avoid converting between types or adding different types together. A bare “MW” is not guessed into a type unless there’s evidence to back it up. If UPS capacity is provided, it counts as IT capacity. Backup generators are not counted as capacity. A group-wide total (“1.5 GW across Europe”) is not disaggregated across sites. Energy consumption figures (MWh) are not used to estimate capacity, because most sites run well below what they have installed.

4. One number per live site

Each live site contributes one figure, from the first of these it can reach:

  1. on a site with several buildings, the buildings in service, each with its own figure, rather than a campus figure that includes buildings not yet built;
  2. a published IT capacity. For a finished building we prefer its full build-out figure, because the capacity an operator has contracted out of a finished building understates what it holds. For a building still being fitted out we take what is running now;
  3. a published grid connection;
  4. floor area × a measured density (section 5).

A capacity shown for a whole site is either the whole site’s own figure or the total of every building in service.

Two consequences: 1 334 MW of the live total is grid connection, which overstates IT load (Ireland most of all), and 168 live sites have no usable figure, so the real total is higher.

5. Floor area to capacity

Where only floor area is published, we apply the median kW per m² of sites that publish both. The figure and the area must cover the same thing: a whole finished site, or a single building in service, never a site-wide target set against the halls standing today. We keep total floor area, white space (the data halls) and ground footprint strictly apart.

Applied tokW/m²Measured from
White space1.87 kW/m²220 colocation sites
Floor area published by an operator0.95 kW/m²129 colocation sites
Floor area from a building register0.50 kW/m²82 sites
Hyperscale floor area1.47 kW/m²22 hyperscale buildings
Footprint × storeys from a national register0.50 kW/m²as register floor area

Registers count whole buildings, including offices and plant rooms, so register floor carries about half the load of an operator’s figure. Where an operator’s own financial filings state its IT load per m² of floor, we use that ratio for that operator’s floor instead of the median. Hyperscale halls get their own density, but the sample is small and widely spread (0.8–2.9 kW/m²). A footprint is converted only where a register gives the storey count. Densities are recalculated at every update, and every page states how much of its figure was converted.

6. Enterprise capacity (modelled)

Many data centers are server rooms and in-house facilities run by banks, manufacturers, hospitals, universities, public bodies and managed-hosting providers. They are about 85% of Europe’s total site number, mostly under 500 kW, and they publish almost nothing. Rather than show them as zero, we model them, following the example of EUDCA. This is the only figure we publish that does not rest on a primary source. It has no sites or locations and is always labelled as a model.

Scope. Single-tenant facilities of 50 kW or more, plus managed-hosting providers. This matches the industry association’s own definition.

Method.

  1. Anchor. Germany’s energy-efficiency register (EnEfG § 13) is Europe’s only statutory census of data centers. For 2025 it lists 300.1 MW of enterprise and 74.6 MW of hosting capacity. We classified every site owner by economic sector to get a capacity per unit of sector size. Four sectors with no German site above the threshold get a share by employment rather than zero.
  2. Scale. These intensities are applied to each country’s employment by sector. The research sector uses university and government R&D spending instead.
  3. In-house tilt. Each country is adjusted by how much of its larger firms’ IT infrastructure is maintained in-house rather than outsourced, relative to Germany (Eurostat ICT survey, 2015–2018).
  4. Level. Europe’s total is the association’s 3 101 MW (EUDCA / Pb7 Research, 2026). Three countries are held at their own national measurements, and the remainder is split across the rest by steps 1–3:
countryenterprise capacitysource
Germany750 MWthe register plus unregistered capacity (Borderstep Institute study for the federal economics ministry)
Switzerland110 MWSwiss Federal Office of Energy, 2026 data center study
Netherlands121 MWDutch Data Center Association, 2026

What we add. A country total includes only the model’s figure minus the enterprise and hosting sites we already hold: 2 660 MW across Europe. Adding the full figure would count those sites twice.

Checks. The model reproduces the German register exactly. With its own measurement removed, the Netherlands lands within 6% of it. The country shares also match an independent per-country distribution to within a typical 13%.

Limits. Nothing below 50 kW is counted, and countries measured more completely suggest the European total is low: a Dutch count from 10 kW puts roughly half of enterprise capacity below 50 kW and the in-house survey dates from 2018.

7. Regulators’ totals

Where a regulator publishes a national total, such as Germany’s 1 594.1 MW across 485 data centres for 2025, we show it separately and never subtract it from ours. The two count different sites: Germany’s register covers only sites above 300 kW that have filed, while ours covers every site we hold evidence for.

8. Locations

Coordinates are never produced by AI. They come from public registers, then national open address services, then OpenStreetMap’s geocoder, then manual placement on national open aerial imagery. No commercial map service is used. Sites without a reliable location stay in the totals but are not shown on the map.

9. Known issues

  • The totals are lower bounds: sites without figures are excluded, and the enterprise capacity is probably underestimated.
  • Grid connections inflate some countries.
  • Floor-area conversions are medians, and individual sites can differ widely.
  • Capacity is not consumption: sites rarely run at full load, and new ones take years to fill.

Corrections

If a figure is wrong or out of date, point us to a public source and we will update it, keeping a record of the change: data@ornaq.ai