Methodology

What we include

The map covers Greater London and the M25 fringe: very large campuses and other significant data centre proposals from the Slough–Heathrow availability zone through Park Royal to Docklands and smaller inner schemes where they change the local planning picture. It is not limited to the biggest cloud operators alone.

Where the boundary sits

The region is the M25 orbital plus its functional fringe: anywhere inside the M25, or on a recognised metro cluster that shares London’s grid and planning gravity even where it sits just outside the ring — the Slough Trading Estate and Slough–Heathrow zone, the Hemel Hempstead and Harlow (M11) corridors, and the northern SoS-called-in schemes marketed into London. This is “functionally-London load”, not a local-authority line. A handful of sites sit materially beyond the M25 and are included only because the operator brands them as London-metro capacity (the Digital Realty Gatwick / Crawley / Woking cluster); each is flagged beyond M25 on its dossier so the ring’s scope stays legible.

When we map a site

A pure megawatt floor would hide locally-material schemes, so a site earns a pin if it meets any one of these, not a single size threshold:

• stated or planning capacity of roughly 10 MW or more; or
• a resolvable live planning application on a London borough (or agreed fringe) portal; or
• a new-build or major-redevelopment campus with a real land footprint; or
• a distinctive local impact even when smaller (heat reuse, contested land, active objection).

We do not map generic enterprise server rooms, small colocation cages with no planning footprint, or press-only schemes with no planning trail — those sit on the Research watchlist until something resolves. This is not a census of every rack in London; it is the set of schemes that change the planning, land, or grid picture. Missing something material? Tell us.

Campuses vs buildings

Where a campus is disclosed both as a whole and as individual halls (Equinix Slough, Kao Harlow, Telehouse Docklands), the halls are the counted, mapped records and the campus is a context-only dossier — reachable from each hall, but never pinned, ranked, or summed alongside them, so its megawatts are counted once, not twice.

How to read our choices

AI and cloud services can look weightless in software. On the ground they mean grid connections, land take, cooling water, and contested planning. This site exists to make that physical footprint across Greater London and the M25 fringe visible and comparable. The angle is disclosed, not neutral. Within it the rules are strict: every number is sourced or null; claims are shown as claims; we let sourced figures and side-by-side comparisons make the case, and we avoid loaded adjectives.

We lead with stated power size because that is what planning documents and press releases give, and because operators here do not publish load factor (what share of that peak they expect to use day to day). Comparison units (kettles, local geography, homes, water, land) are chosen so communities can feel the scale without a specialist glossary. Evidence that would change a figure: a planning document page, a metered disclosure, or a correction from a named operator via the right-of-reply channel.

Evidence standard

• Press releases and operator pages are not proof of site capacity. We still show operator figures where no planning document gives one, labelled Operator, and a planning-document figure replaces them when found. Most megawatts on this map are operator figures.
• “Sourced” means a specific document and page or paragraph, not merely “the planning portal”.
• A claim with no published data is labelled no public data.
• Directory and secondary sources sometimes quote total electrical capacity (grid substation plus backup generators) rather than IT load — figures that can differ several-fold. Where they conflict with an operator IT-load figure, we use the IT load, so sites stay comparable.
• Campaign sources are labelled as campaign sources.
• No planning-milestone status (in planning, approved) without a resolving link to the live application. Factual states (operating, decommissioned) are evidenced instead by a resolving operator or third-party source.
• When we add a source, we should also save an archived copy. Gaps are listed on the Research page and in the data changelog.

On water

Water is the hardest figure to get, because operators rarely publish it. Data centres are not required to disclose water use, and are excluded from the Environment Agency’s National Framework for Water Resources — so most sites here show water as not disclosed, not as zero. Individual UK facilities have sought up to about three million litres a day for cooling, and a large majority of new schemes are proposed in water-stressed areas.

A few operators claim waterless cooling (Ada Docklands; Vantage’s air-cooled closed-loop standard). We record those as an operator claim of zero on-site cooling water, clearly flagged as unverified — with two caveats: it is not metered, and “waterless” cooling shifts water to the power station instead (higher electricity, more indirect water). We count on-site cooling water only, so a zero here covers the water used at the site, not the extra water used to generate its electricity.

Have a planning water statement, abstraction licence, or metered figure for a site? Send it — water is the disclosure we most want to close.

Where a figure came from

Operator

Operator or promoter figure, never presented as verified fact.

Planning

Taken from a planning application document.

Estimated

Derived with stated assumptions; shown as a range when more than one assumption is required.

Measured

Metered or otherwise measured disclosure.

Stated size is not electricity use

When this site says Stated size, it means how large that data-centre scheme claims to be, in megawatts: usually the biggest size stated for that site, otherwise the grid connection applied for, otherwise the first stage only. The source for the chosen figure sits with the number on the site page and map panel. That is a size on paper for one site. It is not how much electricity the facility will draw while running, and it is not a regional total when two schemes publish the same figure.

No operator on this map discloses expected load factor (the share of peak power used on average), so yearly electricity figures are shown as a range, not as a known total.

• Stated size (nameplate or contracted): fair for like-for-like size comparisons only (for example kettles). Measured in megawatts (MW).
• Assumed average use: stated size × 40-70% load factor. All energy comparisons use this, always as a range.
• Measured energy: supersedes assumptions when available.

Published data-centre literature band (about 40-70% utilisation of nameplate). No operator on this map discloses an expected load factor.

Conversion factors

• Kettle: 3 kilowatts continuous (like-for-like power comparison, not annual use).
• Home: 2,700 kilowatt-hours per year. Source: Ofgem and Department for Energy Security and Net Zero (DESNZ) typical domestic electricity consumption figure used in UK comparisons.
• Local geography: Department for Energy Security and Net Zero (DESNZ) subnational electricity consumption at local-authority level. Download the local consumption figures.
• Water: 141 litres per person per day. Source: Thames Water reported average per-capita domestic use (litres/person/day) — confirm against latest published figure. Olympic swimming pool as a secondary unit.
• Land: acres to football pitches (Football Association pitch about 105×68 metres, roughly 1.76 acres). Wheat-foregone only where Agricultural Land Classification (ALC) grades support a yield comparison. Best and most versatile (BMV) farmland is grades 1, 2, and 3a.

Circle scaling

Map circles scale by the best available power size for that site, in fixed field priority: biggest size stated for that site, then grid connection applied for, then first stage only, then IT-equipment load. Figures in MVA (substation ratings) are never counted as megawatts. Each site page states which field was used and the source of that figure. If two sites publish the same megawatts, both keep that size (joint maximum); compare ranks them equal and lists them A-Z. Conflicting maxima in different documents: we record one figure with its source. We do not average. Map circle sizes follow the megawatt key in the legend, which squeezes the range so a 5.6 MW site stays clickable beside a 1 gigawatt (1,000 MW) scheme. Do not treat circle area as proportional to megawatts on the map. Compare does use true proportions: twice the megawatts is twice the area. Sites with no disclosed megawatts render as dashed circles sized by site area (acres). If area is also unknown, a small dashed marker still appears at the published location (for example Telehouse West). That marks presence only, not a size claim. Colour encodes planning status with a colour-vision-friendly set (warm stone, purple, ochre, teal, blue) so consecutive stages stay in distant hues. Outline style (solid vs dashed) is separate: whether megawatts are disclosed. The compare view draws circles so area matches megawatts (twice the power looks twice as large).

Standing right of reply

Any named operator or developer may respond to any figure or claim on this site. Responses are published in full on the relevant site page, with the date they were received. Send responses to marco@prismaticlabs.ai. Suggested subject: “London Compute Ring: right of reply”. Or open a GitHub issue. Site pages show “Right of reply: open; no response received” rather than silence.

Corrections policy

Material errors are corrected in the data file and noted in the repository’s data/CHANGELOG.md (visible before launch, updated with each correction). Prefer a planning-document page citation. Councillors and people in affected communities without a GitHub account can email marco@prismaticlabs.ai. Suggested subject: “London Compute Ring: correction”. Nothing on the public site writes the data file directly; every tip is checked by editors first.

How fresh the figures are

Each site shows when we last reviewed its figures. If the typical review age across the map goes past 90 days, a banner appears site-wide so outdated data is obvious, not quietly wrong.

The map is maintained by Prismatic Labs. Freshness is mostly event-driven: a site is updated when its status changes, such as a committee decision, a completion, or a sale. The handful of active planning cases are swept about monthly; the wider map is kept within 90 days, after which the staleness banner shows. Corrections from the public or operators are actioned as they arrive and recorded in the changelog. Send one via Contribute.

Claim statuses

On each site page, operator claims carry a status about the public record: unverified, no public data, or no longer stated. That describes what is published, not a verdict that the claim is true or false.

Open data

Sites ship as GeoJSON (data/sites.geojson), coordinates [longitude, latitude]. Schema versioned from day one. Region config lives in data/config.json. Download: sites.geojson, schema, local consumption.