Every campus the Data Center Index tracks is in a free file. 866 counted data center campuses, 270,391 MW of planned capacity and $2.2T of stated project cost, across 64 countries and 46 US states, with 66 public companies linked to individual projects. 1,097 rows in total, which includes the phases, buildings, power assets, semiconductor fabs and non-campus programs we track but never count as campuses.
CSV and JSON, no key and no email, published under Creative Commons Attribution 4.0. 11 columns: the name, its slug, its page URL, the developer, the US state, the country, the status, the full-build megawatts, who states that figure, whether the row counts as a campus, and the credit line. The deeper fields stay in the full export.
One row per tracked data center campus, plus the rows we deliberately do not count. Each row carries the campus name, its slug, its page URL, the developer, the US state or the country, the status, the full-build megawatts as a source states them, who states that figure, whether the row counts as a campus, and the credit line. Rows we do not count carry counted = no, so you can reproduce our published campus counts and megawatt totals exactly rather than taking them on trust.
What is not in the free file, stated plainly. Coordinates, the tenant, the city, live megawatts, the stated cost, the expected completion date, the source URL behind each figure and the public companies linked to a project are not in it. Those are on every campus page, and in the full export, which is free with an email address on the projects page. 873 of 1,097 rows have map coordinates from the site or town a source names, so you can see what the fuller file adds before you hand over an address.
Coverage is named campuses, generally 30 MW and up, not every building, colocation suite or carrier hotel on earth. 661 rows carry a megawatt figure and 311 carry a cost figure; a blank means no source states one, and we never derive megawatts from cost, acreage or building count.
| Column | Type | Meaning |
|---|---|---|
name |
text | Campus name as the source names it |
external_id |
text | Stable slug for the campus, and the last segment of its page URL |
url |
url | Page for the campus on datacenterindex.ai, where its sources and remaining fields are published |
developer |
text | Company a source names as building or investing in the campus |
state_code |
text | Two-letter US state code, blank outside the US |
country |
text | Country name |
status |
text | announced, permitted, under_construction, operational, paused or cancelled, per campus |
capacity_planned_mw |
number (MW) | Full-build IT capacity as a source states it, blank where no source states one |
capacity_basis |
text | filed (a filing, permit or utility record states the figure), stated (the developer or press states it) or estimated (modeled or inferred) |
counted |
text | yes when the row is a campus that enters counts and megawatt totals |
attribution |
text | Credit line that must travel with the row under CC BY 4.0 |
A worked example. PORTS-Pike Technology Campus in Piketon, Ohio carries capacity_planned_mw 8,000 with capacity_basis stated and capacity_as_of 2026-08-18. Read that as: a named publisher stated that figure on that date, and its page lists the sources behind it. A row labelled estimated has no such publisher behind the number, and it carries no as-of date.
Scrapers run every weekday after the US market close, and a coverage sweep each Wednesday looks for newly announced campuses and re-reads rows whose sources have gone stale. In the last seven days the directory added 138 sites, moved 25 campuses to a new status and corrected 66 figures.
Rows last changed 2026-10-30, which is the dateModified on the download. Every change is listed on the changelog, and the week in review is on today.
The files are published under Creative Commons Attribution 4.0. Use them commercially, republish them, build on them. The one condition is credit, and the credit line to copy is:
Data Center Index (datacenterindex.ai). Includes Epoch AI Frontier Data Centers data (CC BY 4.0) and AI Data Center Index data, SheetGenius, Inc. (CC BY 4.0).
Two upstream sources are redistributed here under the same licence, and both are credited by name on the rows they touch:
Every row keeps an attribution column, so the credits travel with the data whatever you do with it.
Add ?state=XX to either file for one US market, for example the Texas cut. The full export adds coordinates, tenant and city, live megawatts, stated cost, expected completion, the source URL behind each figure, our internal identifiers, per-row notes and the supplier links, as CSV or GeoJSON. It is on the projects page and asks for an email address; the two files above never do.
The directory is also an MCP server, so ChatGPT, Claude, Cursor or your own agent can call it directly: POST https://datacenterindex.ai/mcp, streamable HTTP, stateless JSON. The same tools are a plain REST twin at https://datacenterindex.ai/api/v1/<tool>, and https://datacenterindex.ai/api/v1/tools lists them. Tools cover project search, a single campus, a US state market, a company, stock moves and recent changes; every result carries the page URL to cite.
Quotas are 60 requests a day with no key, 1,000 on a free key (any confirmed subscriber can have one, it is emailed to you) and 20,000 on a Pro key. The key form and the full tool list are further down this page.
How rows are sourced, what each capacity label means and the error rate our last audit measured are on the methodology page. If a row is wrong, open its campus page, expand sources at the bottom and use "Suggest a correction or add a source", or email hello@datacenterindex.ai with a link that states the right figure.
Yes. The campus files are published under Creative Commons Attribution 4.0, so you can use them commercially as long as you credit Data Center Index (datacenterindex.ai) and keep the two upstream credits that travel with the rows: Epoch AI, Frontier Data Centers (CC BY 4.0) and AI Data Center Index, SheetGenius, Inc. (CC BY 4.0).
Scrapers run every weekday after the US market close, and a coverage sweep each Wednesday adds newly announced campuses and re-verifies stale rows. Every row carries a last-verified date, and what changed in the last week is listed on the changelog.
Because no source states one. 661 of 1,097 rows carry a megawatt figure, and a blank means the developer, the permit record and the press have all stayed quiet on capacity. We never derive megawatts from cost, acreage or building count, and a figure nobody publishes is labelled estimated where we model it.
Coordinates come from the source that published the site: a planning or permit record, a utility filing, the operator's own project page, or an upstream open dataset. 873 of 1,097 rows have published coordinates, and we do not place a pin on a campus whose location no source gives. They are not in the free CSV or JSON: coordinates ship on each campus page and in the full export, which is free with an email address on the projects page.
Cite it as: Data Center Index, global data center campus directory, 866 tracked campuses, datacenterindex.ai/data, CC BY 4.0, retrieved [date]. Include the upstream credits shipped in the attribution column of every row.