Finding · July 18, 2026

384 TWh: how much of AI's footprint is actually disclosed?

Every current-year energy figure in our dataset is a model, ours included. Only 3 of 27 tracked companies have ever published a company-wide energy total, and only 6 have disclosed any environmental metric at all.

The headline on our homepage says the 27 companies we track consumed roughly 384 TWh of electricity in their latest reported year. Here is what that number is made of: all 27 of the current-year energy figures behind it are estimates. Every single one carries confidence tier C, meaning it comes from our capacity model rather than from a company telling anyone what it used.

That is not a weakness we are hiding. It is the finding.

What companies have actually disclosed

Across every fiscal year in our records, exactly three of the 27 companies have ever published a company-wide total for their own electricity consumption: Google (24.3 TWh, FY 2023), TSMC (23.6 TWh, FY 2023), and Meta (15.3 TWh, FY 2024). Together that is 63.2 TWh of company-reported energy, against a modeled current-year total of 384 TWh. Roughly one sixth of the picture rests on something a company actually said.

Widen the lens to any environmental metric at all, energy, any carbon scope, or water, and only 6 of the 27 tracked companies have a single company-reported figure in our dataset. For Scope 3 emissions, the category that usually dominates a hardware or infrastructure company's footprint, 26 of 27 current-year values are modeled. For water withdrawal, all 27 are.

Why this matters for every number you read

It is not just our numbers. Any aggregate claim about AI's energy or water use, from a newspaper, a research lab, or an agency, is built on the same disclosure base, because there is no other one. When the underlying companies publish nothing, everyone downstream is modeling, whether they say so or not.

We handle this in two ways. First, every figure we publish carries a confidence tier, so an estimate can never silently pass as a disclosure; the tier composition of any total is visible in the data tables and the downloadable CSV. Second, we grade the companies themselves on disclosure, T-1 through T-5, because in a field this opaque, the most informative fact about a company is often not our estimate of its footprint but how hard it makes estimating necessary.

The fastest way to make our 384 TWh figure obsolete is for the companies on the list to publish their own. We would consider that a success.

Every figure referenced here is in the observatory dataset with its confidence tier and source. Methods are documented on the methodology page. Corrections and record changes are logged on the updates page. If you can show a figure here is wrong, write to research@openbutterfly.com and we will correct it publicly.