EVIDENCE BEFORE ESTIMATES

A map you can
trace back to a source.

This atlas maps physical computing infrastructure using public geographic records and official operator information. It is a sourced snapshot, not a complete census or a live monitoring service.

What’s in this first edition

The snapshot contains 4,614 OpenStreetMap objects before reconciliation, selected Google development locations, Google Hamina, and six Google Cloud regions. Only developments with a source coordinate or an explicitly approximate locality anchor are mapped. Other Google developments await geographic review.

OpenStreetMap

Objects tagged telecom=data_center are retrieved using Overpass. Names and operator labels are preserved, and each record links to its original object. A node uses its mapped point; a way or relation uses its bounding-box center. These are community records, not independently surveyed locations.

© OpenStreetMap contributors. The OSM-derived database is available under the Open Database License 1.0. Download the OSM-derived database. Operator facts are separately attributed and not relicensed as OSM data.

Official operator records

Google’s location directory supplies the selected development projects and Hamina’s operational listing. “In development” is preserved as a broad planned status; it does not imply a building permit, active construction, secured power or a confirmed opening date. Operator regional coordinates and editorial town anchors are labeled as approximate.

Cloud regions are a separate geography

The initial cloud layer contains six Google Cloud regions, cross-checked against Google’s regions and zones documentation. These markers represent deployment regions, not buildings. They are excluded from physical facility totals. AWS, Azure and Oracle coverage remains to be added.

What a count means

Counts describe mapped records. OSM can represent a campus as one area and its buildings as separate objects; these have not yet been fully reconciled. Nearby matching Google OSM records are superseded by a curated Google campus record. This is not yet a deduplicated census of unique campuses, a market-share estimate or an operator ranking.

Operator filters and directories group explicit aliases: Amazon, Amazon Web Services and its case/region variants are one AWS group; Google LLC is grouped with Google, and Meta Platforms with Meta. Original source labels are retained on records. AWS has 305 attributed mapped records in this snapshot (232 documented by 2024), not four and not a verified global facility total. Seven additional Amazon-named records with no source operator attribution remain unassigned rather than inferring ownership from a name.

Status, capacity and confidence

A data-center tag alone does not establish that a site is operational. Unverified status stays unverified. Capacity and AI role remain unknown unless supported by a specific source. This edition includes a small set of official lifecycle histories and published annual electricity measurements, but no capacity estimates or environmental scores. Retrieval dates mean the source was accessed, not that each assertion was independently verified.

Geography and basemap

The map uses bundled Natural Earth public-domain geography, via world-atlas 2.0.2. It works without a tile-service account. Country assignment uses generalized boundaries and can be unresolved near coastlines or borders. The basemap is intended for regional exploration, not parcel-level navigation.

Time slider and historical coverage

The annual timeline runs from 2007 to the September 2026 snapshot. Past years show the latest sourced lifecycle event by year-end, carried forward until another event is documented. This is not continuous monitoring: the panel links the event and its date. Records without dated evidence are shown only in the latest snapshot. A missing marker in the past does not prove that the facility did not exist.

Ten curated campuses or projects have official dated lifecycle histories. Other dates usually describe when a record was first mapped in OpenStreetMap, or evidence from Wikidata and source tags. These dates establish documentation, not an opening date. Current names, operators and geographic anchors are retained when browsing past years.

A published electricity measurement can establish earlier campus evidence for its representative map record. It does not establish when any particular building opened or that the campus had reached full operation. Sources, evidence dates and the original mapping dates remain available in each record.

Electricity: global estimates and reported consumption

The map’s electricity chart shows only the world total, from IEA (2026), Electricity consumption by data centres, 2020–2035 (CC BY 4.0; chart updated 16 April 2026). The published annual historical estimates are 269.1, 297.0, 324.1, 360.4, 415.6 and 485.1 TWh for 2020–2025. The 2026 figure, 560 TWh, is the IEA base-case projection, shown with a striped bar and a projection label. Values were extracted from the chart’s published data; we do not interpolate missing years or construct our own model.

The world total describes all data centres, independent of the map’s incomplete inventory, search, country or operator filters. It is not the selected operator’s electricity use. World, national and campus scopes overlap and are never summed. Published campus consumption appears on each campus record page and in the Google and Meta guide.

The campus sample contains 102 annual measurements for 17 Meta campuses across 2017–2024. The 2017–2019 figures come from Meta’s 2023 Sustainability Report, PDF page 51; 2020–2024 use Meta’s 2025 Environmental Data Index, PDF page 6. Each observation links to its report edition. These are operator-reported consumption measurements, not modelled demand or live meter readings. MWh are divided by 1,000 for display in GWh.

Figures describe the whole campus, not an individual building. Reviewed campus-to-record associations let multiple buildings reference the same report, while counting each campus once in a filtered total. Report names Richmond, Papillion and Newton County correspond to Henrico, Sarpy and Stanton Springs respectively. Mapping relationships remain separate from the original OSM records.

Totals include only campuses with a published value for the selected year and a matching map record. The reporting-campus count is shown beside the number. A dash or missing year stays unknown: there is no interpolation, forward fill, capacity conversion or renewable-purchase substitution. This is a small, changing sample, not a global total. It contains no 2025 or 2026 campus measurements.

The separate Ireland series uses CSO’s 2025 release, Table 1, covering 2015–2025. The full series uses that edition’s revised values. It is a national metered total, independent of map filters, and is never added to the overlapping campus sample. The previous release is archived separately. Download the sourced electricity series.

The Power use guides add quoted context from the IEA, Berkeley Lab, EPRI, the CSO, Google and Meta. Every quote in data/energy-context.json can be re-checked against its live source with pnpm data:verify-quotes; PDF values are tested against pinned extracts in data/raw/.

PeeringDB

PeeringDB is a valuable candidate for interconnection coverage. Its acceptable-use policy restricts bulk redistribution. No PeeringDB records are included in this public dataset. Permission and intended-use review are needed before adding it.

Corrections and next data work

For a community record, follow its source link to review or correct the original OSM object. The next data work is broader historical coverage, campus reconciliation, lifecycle-tag ingestion, additional official operator directories and field-level verification. This edition does not accept direct public edits.

Snapshot: 2026-09-13. Return to the atlas →