Methodology
Directories disagree about what counts as a data center by a factor of three. This is where we draw the line, where our data comes from, and how we keep it deduplicated across sources.
What we map
A facility makes the map only if it passes all five tests. PeeringDB listings are included as interconnect facilities even when power or floor area is unpublished — the listing itself is the citable evidence. The 500 kW / 50 cabinet / 2,500 sqft floor is applied when we ingest operator and hyperscale pages that would otherwise pull in smaller sites.
Inclusion matrix
| Category | Status | Notes |
|---|---|---|
| Commercial colocation | Include | Equinix, Digital Realty, CoreSite, DataBank… |
| Hyperscale buildings | Include | Each named building, not just the campus |
| Enterprise (≥ 500 kW) | Include | Banks, retail, healthcare — when documented |
| Carrier hotels with tenants | Include | One Wilshire, 60 Hudson, Telehouse… |
| Standalone HPC centers | Include | NCSA, ORNL, etc. Purpose-built and at scale |
| Standalone edge facilities | Include | Vapor IO, Compass Edge — actual buildings |
| Cloud regions | Include | Tracked as a separate layer on the map |
| Government (when public) | Include | DoE/NASA-published, etc. |
| Pure telco central offices | Exclude | Switching only — not a data center |
| Crypto mining facilities | Exclude | Different category, different operating model |
| University CS server rooms | Exclude | Research compute is not infrastructure-as-product |
| Cabinets at cell towers | Exclude | Edge compute ≠ a facility |
| Server rooms in non-DC buildings | Exclude | Closet test fails |
| Rumored or unannounced sites | Exclude | Wait for a source |
Where the data comes from
PeeringDB
The directory operators use to arrange peering. It is also where networks (34,732 ASNs) and internet exchanges (1,309) come from; we join those to facilities through PeeringDB's published facility links. Most rows start here.
OpenStreetMap
Crowd-tagged buildings with telecom=data_center. 97 OSM records; 92 are facilities PeeringDB does not list. Useful for sites outside the interconnect ecosystem.
Operator websites
Per-facility pages from Equinix, Digital Realty, DataBank, Cologix, CoreSite, CyrusOne, QTS, Iron Mountain, H5, Vantage, Aligned/ODATA, NEXTDC, and STACK. A match enriches the existing row with specs PeeringDB does not publish — power, space, UPS, certifications — when the operator discloses them. Many do not: Equinix still publishes no MW. Unmatched pages (STACK campuses, some NEXTDC, H5, Vantage, and Aligned buildings) become new canonical facilities, not just overlays.
Google and Meta location pages
60 Google and 33 Meta buildings from their public location lists. Those pages give city and region, not a street pin — we geocode the city and keep the named building. Microsoft, AWS, and Apple buildings are not in the facility table yet; those clouds appear only as the region layer.
Cloud provider region pages
AWS, Google, Azure, Oracle. Tracked as a separate map layer, not in the facility table. A region is a logical grouping of buildings, not a building.
No user submissions. We do not scrape directories whose terms forbid it (datacentermap.com, Cloudscene).
How we match and dedupe
Each canonical facility row can have many source records — one for every directory or page where we found that building. New sources are attached to an existing row before a new one is created.
PeeringDB and OpenStreetMap use a two-step match: exact (operator, name), then any facility within 100m. Unmatched OSM tags are inserted as new facilities.
Operator pages and hyperscale buildings use a stricter matcher scoped to the same operator: exact name, then name prefix, then facility code. There is no spatial step — a 100m search was merging neighboring buildings of different operators. Equinix, Digital Realty, DataBank, Cologix, CoreSite, CyrusOne, and QTS pages that still miss are held as orphans for review. H5, Vantage, Aligned, NEXTDC, STACK, Iron Mountain, Google, and Meta pages that miss are inserted as new facilities.
That is why an Equinix site known by three names across PeeringDB, OSM, and Equinix’s own website is still one row, with three source records linked to it.
Data from PeeringDB (CC-BY-SA), OpenStreetMap (ODbL), and operator-published facility pages · Map tiles by Mapbox · Privacy