Data

Methodology & data dictionary

Last updated July 27, 2026. What counts as a facility, what our numbers mean, how confidence is graded, and where every figure comes from.

Every figure traces to a source

When our pipeline extracts a value (a capacity figure, a certification, a cooling design) it stores the value together with the document it came from: the source URL, the publisher, the surrounding text, a confidence score, and the date we captured it. We call this the evidence lake, and it now holds more than 55,000 field-level citations.

That record is not an internal convenience. Inside the DC Atlas app, sourced fields carry a source link, so you can open the document behind a number instead of taking our word for it. Closed databases ask you to trust a figure because it is in the database. We think the stronger claim is the opposite one: here is the number, and here is where it came from.

55,000+source-linked field values
Gradedevery source rated by reliability, from operator-official down
Datedeach citation keeps the date it was captured

Definitions: what counts, and what the numbers mean

Comparing databases starts with definitions. These are ours, stated plainly so you can check any figure we publish against them.

What counts as a facility

One DC Atlas record represents one physical thing, at a declared level. A record is always marked as one of the following, and capacity is only ever stated at the level the record represents. A campus total is never stamped onto one of its buildings.

Record levelWhat it representsHow it is counted
CampusA whole multi-building site. Its capacity figure is the site total.Counted once, at the site level.
BuildingA single building. Its capacity figure covers that building only, never the campus around it.Counted individually.
Tenant presenceA provider that leases space inside someone else's building: a reseller footprint or a network point of presence.Excluded from maps and capacity totals, so the host building's power is never counted twice.

Capacity: what a megawatt means here

IT load, not total draw

Capacity figures are IT load: the power available to customer equipment, in megawatts. Total site draw (which includes cooling and losses) is a larger number; quoting it would flatter every facility. Where a source only publishes total power, that goes through review before any figure is applied.

Live capacity vs full buildout

We hold two separate figures: live capacity, what is operational today, and full buildout, what the site reaches if every announced phase completes. Each carries its own confidence grade. A 300 MW ambition is never blended into today's supply, and a facility page shows both explicitly when both are known.

Estimates are labelled

Where no capacity is published anywhere, we may derive an estimate, for example from building size and typical power density. Estimates are always shown with an “est.” label, and market or operator totals disclose what share of the total is estimated rather than silently blending it in.

Development pipeline stages

Pipeline records sit in one of two stages, and the stage is the project's own development phase as our record holds it — announced, or under construction. Nothing else feeds it. That matters for a reason worth stating plainly: how much we happen to know about a project can never change which stage it appears in. How firm the evidence is behind it is a separate reading, and it has its own grade, below.

AnnouncedThe project is on the record and not yet building: publicly announced by the developer or operator, granted a permit, or recorded by us at an earlier phase. Some of these carry a dated public signal and some do not — the confidence grade below says which.
Under constructionGround has been broken and building work is verifiably underway.

Operational is not a pipeline stage. The pipeline measures supply that is still coming, so a site leaves it when it energizes, and from that point its power is counted as live capacity instead.

Each pipeline record also carries a confidence grade, which is what tells two projects in the same stage apart: a granted permit, a groundbreaking or confirmed construction reads high, a bare dated announcement reads medium, and a record we hold with no dated public disclosure behind it — no announcement date, no permit, no groundbreaking — reads low. That last case is neither folded into the firmer numbers nor quietly dropped; it is included and labelled, so the weakest evidence in the pipeline is the part marked most clearly and you can discount it yourself.

Other core fields

  • Building area is total building square footage; data-hall, colocation, and raised-floor areas are held as separate fields where published, never conflated with the building total.
  • PUE is recorded as two fields: design PUE (the engineered target) and measured PUE (a reported operating figure). They are never substituted for one another.
  • Tier certification records the rating a facility actually holds, with the certification date where published. Marketing claims of “tier-equivalent” design are not certifications.
  • Rack density distinguishes maximum supported density from average deployed density, in kW per rack.

Field-by-field definitions for the wider schema live in the glossary.

How confidence is graded

A record carries several independent judgements: whether the facility is real, whether its capacity figure is trustworthy, and how it was checked. We grade them separately, and we show the grade.

The capacity confidence ladder

GradeWhat it meansHow it is displayed
VerifiedConfirmed by an operator-official source, or by multiple independent sources that agree.Shown plainly.
UnverifiedReported by a single online source we have not yet cross-checked.Shown, treated with caution in rollups.
EstimatedOur own derivation, for example from building size and power density, where no figure is published.Always labelled “est.”, never presented as fact.
UnknownNo credible figure found anywhere.Shown as “—”. We would rather show a gap than invent a number.

Record-level confidence works the same way. Facilities are graded from verified down to unverified, and when research finds insufficient public evidence that a listing is a real data center (an office mislabeled by an aggregator, a reseller's virtual address), the record is flagged and withheld from public surfaces rather than deleted, so the judgement itself stays auditable.

Honest aggregates. Market and operator totals are computed from our facility-level records, never imported as someone else's headline. They keep live capacity, construction, and planned pipeline as separate figures. Tenant presences are excluded so the host building's power is never counted twice, and so are records where research could not establish that the site is a real data center. Projects we hold with no dated public disclosure are a different case: they are included in the planned figure and labelled with the lowest confidence grade, not removed. Where a facility's only figure is an estimate, the facility page says so.

How the data is gathered and kept current

Where the data comes from

Our records are compiled from publicly available sources: operator websites and marketing materials, press releases and investor communications, planning and permitting filings, regulatory disclosures, procurement notices, and reputable industry press.

We use AI-assisted research tools to locate, cross-reference, and summarize this public information at scale. Every record describes something that is publicly documented. Our researchers and pipelines are finders and organizers of public facts, not extractors of anyone else's proprietary database.

How records are verified

High-stakes fields go through dual-source triangulation: two independent research systems are asked the same question about the same facility. When both agree against our record, we correct it. When they disagree with each other, the record goes to a human reviewer instead of being changed.

Automated guards block implausible values before they can land: a capacity out of realistic range, a campus total about to be stamped onto a single building, a swing too large to be credible. Every applied change is flagged for review and fully reversible.

How records are normalized

Raw findings are normalized into our own schema: standardized field definitions (power in MW, PUE, tier levels, operational status), one consistent taxonomy for facility types and company roles, and locale-aware naming. Every facility carries a DC Atlas catalog name minted by our own convention, the way a library assigns catalog numbers.

Aggregate figures shown on DC Atlas (a company's tracked capacity, market totals, portfolio breakdowns) are computed by us from our own facility-level records.

How fresh the data is

Monitoring pipelines run daily, tracking announcements, filings, and procurement notices across the markets we cover, and feeding confirmed developments back into facility records. Rolling enrichment sweeps re-research the corpus section by section; each sweep re-checks identity, status, capacity, and location against current public sources.

Capacity history is preserved as quarterly snapshots, so market and operator trends reflect how supply actually developed rather than a single point in time.

Coverage

Facility records across established and frontier markets, the companies behind them, and the public procurement activity around them, monitored across European and US procurement and filing systems with further regions in progress.

Thousands

Facilities tracked

Every record independently researched.

200+

Global markets

From established hubs to frontier regions.

2,100+

Companies

Operators, owners, developers, utilities, tenants.

55,000+

Source citations

Field-level values stored with their source.

Corrections

Data about a fast-moving industry is never perfect; when the public record changes, ours should too. If you operate a facility we've described incorrectly, or spot any error, contact us and we'll review it against the public record and correct it. Corrections go through the same verification loop as everything else, and the prior value stays in our history.