Methodology
What we track
We assess five dimensions of environmental transparency for each major AI provider:
- Water Usage
- Does the provider publish total water consumption figures for their data center operations?
- Energy Usage
- Does the provider publish energy consumption data, including metrics like PUE (Power Usage Effectiveness)?
- Per-Facility Data
- Does the provider break down environmental data by individual facility or region?
- AI-Specific Data
- Does the provider publish water or energy data specific to AI/ML workloads, as opposed to aggregate data center figures?
- Water Pledge
- Has the provider made a public commitment to water stewardship, replenishment, or positivity?
How we assess
Each dimension receives one of three ratings:
- Yes — The provider publishes substantive data in this area, sourced from an official report or statement.
- Partial — Some data is published but is incomplete, inconsistent, or lacks granularity compared to peers.
- No — No relevant data has been published as of the last review date.
How we verify
Every assessment links to its source: a published sustainability report, official blog post, or regulatory filing. We do not use anonymous tips, leaked documents, or third-party estimates without clear attribution. If we cannot cite a source, the assessment is "No."
Build water
We estimate the water footprint of building this site using published research on data center water consumption (Li et al., 2023; Macknick et al., 2012): per-token energy estimates, regional PUE/WUE values, and grid water intensity. The coffee comparison uses 140 liters per cup from the Water Footprint Network.
Contributing
This is an open-source project. If you find an error or have data we missed, please open an issue or submit a pull request on GitHub. All contributions must include a verifiable source.