About Gigawatt Map

Gigawatt Map is a public atlas of the physical infrastructure behind the AI buildout: every datacenter we can verify, the power plants and substations feeding it, the submarine cables connecting regions, and the operators and tickers with exposure.

It is not a trading tool, a real-estate listing service, or a permit database. It is a reference map for anyone trying to understand where AI capacity is actually being built, who is paying for it, and what grid is absorbing the load.

The project is open-source and open-data. The code is MIT; the data carries the licenses of its upstream sources, listed in full below.

We write the page with five readers in mind. If you are not one of these, it probably still works, but these are the use cases we actively design against:

1. AI infrastructure investors — locating operator capex exposure in 30 seconds and confirming a name against a site and a substation before sizing a position.
2. Analysts & journalists — cross-referencing announced MW against built MW, grounding a story in a map they can cite.
3. Developers & operators — scouting power and interconnect adjacency for new builds.
4. Curious technologists — seeing where the model they used last night physically lives, and what it is plugged into.
5. Educators & students — an on-ramp to the material reality of AI compute.

A Python pipeline (data-pipeline/) fetches each source on its own cadence, normalizes it against Pydantic schemas, and writes merged artifacts — per-source GeoJSON for ODbL compliance, plus PMTiles for the web map. Every pipeline run emits a manifest.json with source versions, row counts, and hashes so a given map tile is traceable back to the exact upstream snapshot.

The web app (apps/web/) is a Next.js App Router build served at the edge. API routes query the artifacts by bounding box; nothing hits a live upstream at request time.

Refresh cadences vary by source and are documented in the inventory below. “Last updated” at the top of this page reflects editorial review, not the freshest tile.

Every layer the reader sees on the map is sourced from a public dataset or a hand-curated list maintained in this repo. License terms are honored: ODbL sources are redistributed per-source (never merged into a single download), and CC BY-NC-SA sources are used only for non-commercial display.

We prefer to be loud about what the map does not cover:

• Chinese datacenter coverage is limited. OSM tagging is sparse in mainland China and regulatory filings are not public, so our Chinese footprint is best-effort from press coverage and operator disclosures.
• Indian coverage is improving as hyperscaler builds are announced, but lags the US and EU.
• Russia, Belarus, and Central Asia are poor — we would rather show gaps than guess.
• MW values for pre-2020 colo facilities are frequently estimated from footprint, not disclosed. Treat them as ranges.
• Neoclouds announce and lease capacity faster than our refresh cadence. If you see a fresh announcement, it probably beats the map.

The repository lives on GitHub. Open a pull request against data/curated/ to add or correct a facility; every row needs a primary source_url. For larger contributions — a new source module, a methodology change — open an issue first.

Press and partnership inquiries: hello@gigawattmap.com.

Sudhendra Kambhamettu, lead. The project is open to collaborators who want to own a data source, a region, or a methodology question — see the GitHub issues tagged help-wanted.

None yet. When coverage appears, it will be listed here with a link to the original.

Code is released under the MIT license. Data retains the license of its upstream source, listed in the inventory above. Attribution is required when redistributing any layer — the attribution string for each source is embedded in the corresponding manifest.json entry.