Data & methods · data version 26.08-c2
Data dictionary & sample
The columns of the free sample, then the full schema behind the atlas. One row per physical asset, measured values only (no estimates). Every substantive field <field> is paired with a src_<field> column holding its source. Filled = the number of facilities that carry the column, out of the 8,460 facilities of the atlas (the same total as on the home page and the map), with the share in brackets.
free sample
Free sample · release 26.08-c2
See what the data looks like: 200 facilities across 61 countries, 67 columns and the source of each value. Rows are hand-picked for completeness (mostly grade A, 51 with measured IT-load), so the sample shows the best of the dataset, it is not statistically representative of the full 8,460.
ODbL 1.0: attribution required, share-alike on derived databases. Facilities © OpenStreetMap contributors · PeeringDB (AUP) · Wikidata (CC0) · country reference GADM.
Please cite: Datacenter Atlas (2026), data version 26.08-c2. https://www.datacenter-atlas.com
What you can get, and what you cannot
Available
- A free sample of 200 data centers in 61 countries, as CSV or GeoJSON: every column, and the source of each value.
- Every data center one at a time on the map, with its sources, and the list of the data centers of each country, city and operator on its page, to read there.
- Maps, charts and texts, reusable with credit (licences).
Not available
- The full inventory of 8,460 data centers as a file, in any format (CSV, GeoJSON, GeoPackage).
- A public API or any other bulk export.
- Wider access on request: none is granted, to researchers or to companies.
Why. The inventory merges sources whose terms do not allow redistributing them in bulk, notably PeeringDB. The atlas therefore shows each record with its sources and lists them page by page, and offers a fixed sample, but not the whole file.
sample file columns: 67 columns, CSV & GeoJSON
The free sample uses exactly these columns, in this order. A GeoJSON feature carries every column except lat and lon, which are its geometry (null when missing). dc_uid is the record's permanent id.
Measured record
Stated by a source, or computed from the record (dqs, dqs_grade, the mw_ campus columns). Most names match the full schema below; the sample adds lat, lon, address and continent, and leaves out the internal ids.
| Column | Description |
|---|---|
dc_uid | Permanent asset id: survives rebuilds & source-id changes. Use this to join across releases. |
ref | Fixed, link-resolvable source reference (osm:node/…, pdb:…, disc:…). Confirms a record is the one detected. |
osm_id | OpenStreetMap element id (node/way/…). |
pdb_id | PeeringDB facility id. A site that PeeringDB lists under two records is merged into one row: pdb_id keeps one of them, source_urls lists both, and a src_ field can point to the other. |
wikidata | Wikidata QID of the facility. |
name | Facility name as published by the source. |
operator | Operator name as stated by the source (raw). |
operator_group | Alias-resolved operator (NTT ×5 → NTT; SoftLayer/IBM Cloud → IBM) without merging look-alikes. Group on this. |
operator_subbrand | Original brand/entity kept when consolidated into a parent (e.g. group Digital Realty / subbrand Interxion; group IBM / subbrand IBM Cloud). Null when the name is already the parent. |
asset_type | datacenter / network_pop / unknown. Filter to datacenter for capacity work. |
asset_type_reason | The deciding rule for asset_type. |
lat | Latitude (WGS84) of the facility point. |
lon | Longitude (WGS84) of the facility point. |
geo_precision | building (mapped polygon), point (a point at the site) or municipality (the operator does not publish the address: the point marks the town; no site-level derived values, rule GEO-2). |
geo_flag | disputed_territory or suspect_coordinate; blank = no issue. |
address | Street address as stated by the source. |
city | City / locality. |
region_state | State / province / region. |
country_name | Country name (GADM). |
country_iso2 | ISO-3166 alpha-2 country code. Blank only for disputed territories. |
continent | Continent / world region, filled from the country for every row. |
it_load_mw | Measured IT / critical power (MW): key capacity field. Range 0.4–1400. |
mw_shared | true when it_load_mw is a campus figure published once and carried by several buildings of the same operator within 2 km. Count it once per campus when summing. |
mw_campus_n | Number of buildings carrying the same campus figure (blank when not shared). |
mw_campus_id | Identifier shared by all buildings of one campus whose it_load_mw is a single campus figure (blank when not shared). Group by it and count the figure once. |
total_power_mw | Measured total facility power draw (MW). Range 0.4–1000. |
pue | Power Usage Effectiveness (stored as text, e.g. 1.2). |
gross_floor_area_sqm | Gross floor area (m²). Range 200–445,000. |
tier | Resiliency tier as stated (free text, e.g. Tier III). |
certifications | Certifications held (free-text list). |
cooling_type | Cooling approach as described (free text). |
status | Operational status, normalised from the source text to operational / under_construction / planned (blank = not stated; rule STAT-1). |
commissioned_year | Year commissioned / opened (source- or Wikidata-stated). |
net_count | Networks present at the facility. Range 0–610. |
ix_count | Internet exchanges present. Range 0–37. |
carrier_count | Carriers present. Range 0–42. |
website | Operator / facility web page. |
sources | Contributing datasets (OpenStreetMap / PeeringDB / OpenStreetMap+PeeringDB / Discovery). |
source_urls | Public provenance link(s), one per contributing source. |
last_verified | Date the row was last built/verified (YYYY-MM-DD). |
dqs | Data Quality Score 0–100 (geometry, provenance, country, completeness, capacity). Computed by the atlas from the row, not measured. Range 39–100. |
dqs_grade | Letter grade from dqs: A / B / C / D. |
Provenance: src_<field>
Source URL or label of the paired field. Empty when the field is empty, or when the source gave no link for that value.
| Column | Description |
|---|---|
src_name | Source of name. |
src_operator | Source of operator. |
src_location | Source of the coordinates (lat / lon). |
src_address | Source of address. |
src_city | Source of city. |
src_region_state | Source of region_state. |
src_country | Source of the country assignment (GADM). |
src_it_load_mw | Source of it_load_mw. |
src_total_power_mw | Source of total_power_mw. |
src_pue | Source of pue. |
src_gross_floor_area_sqm | Source of gross_floor_area_sqm. |
src_tier | Source of tier. |
src_certifications | Source of certifications. |
src_cooling_type | Source of cooling_type. |
src_status | Source of status. |
src_commissioned_year | Source of commissioned_year. |
Derived indicators: derived_*
Modelled or joined context from open layers, not measured facts about the facility. Kept apart by the derived_ prefix; methods on the indicator register.
| Column | Description |
|---|---|
derived_wsi | Water Stress Index for data centers, 0–100 (higher = more stress). Atlas composite index. |
derived_sfi | Social friction index, 0–100: the share of data centers with a lower absolute value (relative score, method v0.3). Atlas composite index. |
derived_bws | Aqueduct 4.0 baseline water stress score, 0–5, at the site. |
derived_bws50 | Aqueduct 4.0 projected water stress score, 0–5, for 2050 (business as usual, SSP3 RCP7.0). |
derived_fc18_h | Free-cooling hours per year: hours with outdoor dry-bulb temperature ≤ 18 °C (NASA POWER, 2015–2024). |
derived_cdd | Cooling degree days per year, base 18 °C (NASA POWER, 2015–2024). |
derived_hot_days | Days per year with a daily maximum ≥ 35 °C (NASA POWER, 2015–2024). |
derived_pop5k | Resident population (2020) within 5 km (GHSL GHS-POP). |
derived_sub_km | Distance to the nearest mapped high-voltage substation (≥ 110 kV), km (OpenStreetMap). Blank when none within the search radius. |
Full schema
The columns of the full inventory, with the share of records that carry each one. This file is not distributed: the tables document what the map, the pages and the sample are built from.
Identity & lineage
| Column | Type | Filled, of 8,460 | Description |
|---|---|---|---|
dc_uid | TEXT | 8,460 (100 %) | Permanent asset id: survives rebuilds & source-id changes. Use this to join across releases. |
ref | TEXT | 8,460 (100 %) | Fixed, link-resolvable source reference (osm:node/…, pdb:…, disc:…). Confirms a record is the one detected. |
dc_id | TEXT | 8,460 (100 %) | Content hash (name+coords) for lineage: not stable. |
fid | INTEGER | 8,460 (100 %) | GeoPackage internal id, per layer, not stable across files; never use as a key. |
geom | geometry | 8,460 (100 %) | Feature geometry (Point / MultiPolygon / either), EPSG:4326. |
Naming & operator
| Column | Type | Filled, of 8,460 | Description |
|---|---|---|---|
name | TEXT | 7,903 (93 %) | Facility name as published by the source. |
operator | TEXT | 7,189 (85 %) | Operator name as stated by the source (raw). |
operator_group | TEXT | 7,197 (85 %) | Alias-resolved operator (NTT ×5 → NTT; SoftLayer/IBM Cloud → IBM) without merging look-alikes. Group on this. |
operator_subbrand | TEXT | 473 (5.6 %) | Original brand/entity kept when consolidated into a parent (e.g. group Digital Realty / subbrand Interxion; group IBM / subbrand IBM Cloud). Null when the name is already the parent. |
operator_canonical | TEXT | 1,343 (16 %) | Wikidata-canonical operator name (where a match exists). |
Location
| Column | Type | Filled, of 8,460 | Description |
|---|---|---|---|
country_iso2 | TEXT | 8,457 (100 %) | ISO-3166 alpha-2 country code. Blank only for disputed territories. |
country_name | TEXT | 8,460 (100 %) | Country name (GADM). |
city | TEXT | 6,197 (73 %) | City / locality. |
region_state | TEXT | 3,421 (40 %) | State / province / region. |
Capacity & physical metadata: measured only, sparse
Stated by a source, never estimated. Treat missing as unknown, not zero.
| Column | Type | Filled, of 8,460 | Description |
|---|---|---|---|
it_load_mw | REAL | 299 (3.5 %) | Measured IT / critical power (MW): key capacity field. Range 0.4–1400. |
total_power_mw | REAL | 156 (1.8 %) | Measured total facility power draw (MW). Range 0.4–1000. |
pue | TEXT | 66 (0.8 %) | Power Usage Effectiveness (stored as text, e.g. 1.2). |
gross_floor_area_sqm | REAL | 391 (4.6 %) | Gross floor area (m²). Range 200–445,000. |
tier | TEXT | 109 (1.3 %) | Resiliency tier as stated (free text, e.g. Tier III). |
certifications | TEXT | 335 (4.0 %) | Certifications held (free-text list). |
cooling_type | TEXT | 345 (4.1 %) | Cooling approach as described (free text). |
status | TEXT | 418 (4.9 %) | Operational status, normalised from the source text to operational / under_construction / planned (blank = not stated; rule STAT-1). |
commissioned_year | REAL | 61 (0.7 %) | Year commissioned / opened (source- or Wikidata-stated). |
Interconnection (PeeringDB)
| Column | Type | Filled, of 8,460 | Description |
|---|---|---|---|
net_count | REAL | 4,951 (59 %) | Networks present at the facility. Range 0–610. |
ix_count | REAL | 4,951 (59 %) | Internet exchanges present. Range 0–37. |
carrier_count | REAL | 4,951 (59 %) | Carriers present. Range 0–42. |
website | TEXT | 4,911 (58 %) | Operator / facility web page. |
Geometry, flags & classification
| Column | Type | Filled, of 8,460 | Description |
|---|---|---|---|
geo_precision | TEXT | 8,460 (100 %) | building (mapped polygon), point (a point at the site) or municipality (the operator does not publish the address: the point marks the town; no site-level derived values, rule GEO-2). |
asset_type | TEXT | 8,460 (100 %) | datacenter / network_pop / unknown. Filter to datacenter for capacity work. |
asset_type_reason | TEXT | 8,460 (100 %) | The deciding rule for asset_type. |
geo_flag | TEXT | 3 (< 0.1 %) | disputed_territory or suspect_coordinate; blank = no issue. |
Source identifiers
| Column | Type | Filled, of 8,460 | Description |
|---|---|---|---|
osm_id | TEXT | 4,159 (49 %) | OpenStreetMap element id (node/way/…). |
pdb_id | REAL | 4,951 (59 %) | PeeringDB facility id. A site that PeeringDB lists under two records is merged into one row: pdb_id keeps one of them, source_urls lists both, and a src_ field can point to the other. |
wikidata | TEXT | 117 (1.4 %) | Wikidata QID of the facility. |
operator_wikidata | TEXT | 1,343 (16 %) | Wikidata QID of the operator. |
Provenance: source of each value
Every substantive field <field> has a paired src_<field> holding its source URL/label.
| Column | Type | Filled, of 8,460 | Description |
|---|---|---|---|
sources | TEXT | 8,460 (100 %) | Contributing datasets (OpenStreetMap / PeeringDB / OpenStreetMap+PeeringDB / Discovery). |
source_urls | TEXT | 8,460 (100 %) | Public provenance link(s), one per contributing source. |
last_verified | TEXT | 8,460 (100 %) | Date the row was last built/verified (YYYY-MM-DD). |
src_location | TEXT | 8,460 (100 %) | Source of the coordinates. |
src_name / src_operator | TEXT | 100 / 84 | Source of name / operator. |
src_country | TEXT | 8,457 (100 %) | Country method: GADM / GADM+1km-buffer / GADM+10km-buffer / GADM-disputed. |
src_it_load_mw · src_pue · src_… | TEXT | – | Source URL for each measured metadata field (it_load_mw, total_power_mw, pue, gross_floor_area_sqm, tier, certifications, cooling_type, status, commissioned_year). |
Data quality
| Column | Type | Filled, of 8,460 | Description |
|---|---|---|---|
dqs | INTEGER | 8,460 (100 %) | Data Quality Score 0–100 (geometry, provenance, country, completeness, capacity). Computed by the atlas from the row, not measured. Range 39–100. |
dqs_grade | TEXT | 8,460 (100 %) | Letter grade from dqs: A / B / C / D. |
data quality score (DQS): how it's computed
dqs is a 0–100 completeness & reliability score computed per facility from the assembled row. Points are added across seven weighted components, geometry, location trust, country resolution, identity, address, provenance and data richness, and capped at 100. It rewards a precise footprint, a trustworthy well-resolved location, a named operator, corroborating multi-source provenance and measured capacity.
| Component | Max pts | How points are awarded |
|---|---|---|
| Geometry precision | 25 | Building footprint +25 · point +10. |
| Location trust | 15 | Coordinates trusted +15 · 0 if flagged suspect_coordinate. |
| Country resolution | 15 | GADM exact +15 · buffer match +10 · disputed territory +7. |
| Identity | 15 | Name +8 · operator (raw or canonical) +7. |
| Address | 10 | City +5 · region / state +5. |
| Provenance | 10 | Multi-source (e.g. OSM+PeeringDB) +10 · single source +6. |
| Capacity & interconnection | 10 | Measured IT-load MW +5 · PeeringDB interconnection counts +5. |
| Total | 100 | Sum of the above, capped at 100. |
Conversion to a rating:
dqs_gradeThe numeric score maps to a single letter grade:
A≥ 80: excellentB60–79: goodC40–59: fairD< 40: sparse
Example: a mapped building footprint (+25) with trusted coords (+15), exact GADM country (+15), name + operator (+15), city + region (+10), two sources (+10) and measured MW + interconnection (+10) → 100 = grade A. A bare OSM point with only a name and a buffer-matched country lands in the C/D range.
categorical domains
| Column | Values |
|---|---|
asset_type | datacenter · network_pop · unknown |
geo_precision | building · point · municipality (town-level point, rule GEO-2) |
geo_flag | disputed_territory · suspect_coordinate · (blank) |
sources | list of: OpenStreetMap · PeeringDB · Discovery (joined with "; " in CSV, e.g. "PeeringDB; OpenStreetMap") |
src_country | GADM 4.1 in the web export (the GeoPackage also records the buffer method: GADM+1km-buffer · GADM+10km-buffer · GADM-disputed) |
status | operational · under_construction · planned · (blank = not stated). Normalised from the source text |
continent | Europe · North America · Asia Pacific (includes Australia and Oceania) · South America · Africa · Middle East (derived from the country) |
mw_campus_id | shared by buildings carrying one campus IT-load figure, group by it to count the figure once |
dqs_grade | A · B · C · D |
using it
- Join on
dc_uid(permanent); verify identity viaref. Never usefid/dc_id. - Filter
asset_type = 'datacenter'for capacity work (exclude POPs / unknown). - Power & area are measured: treat missing as unknown, not zero.
- Trace any value at its
src_<field>URL. - OSM-derived → ODbL on redistribution; PeeringDB AUP; Wikidata CC0.