The 28 companies tracked on this site used about 373 TWh of electricity in their latest reporting year. 199 TWh of that, a little over half, comes from figures that 14 of the companies published themselves (confidence tier B). The other 174 TWh is estimated for the 14 companies that publish no company-wide electricity figure (tier C, from the capacity model on the methodology page).
Which companies report
Company-reported electricity totals come from Google, Samsung, Microsoft, Meta, Micron, Intel, Oracle, Tencent, Alibaba, Huawei, NVIDIA, Broadcom, AMD and Nebius. The largest estimated figure is Amazon's, at about 51 TWh: Amazon reports emissions for the whole company but no electricity total. TSMC, Equinix, SK hynix, Digital Realty and Baidu publish sustainability reports, but their latest figures are not yet in the dataset and are shown as estimates.
The AI labs
None of the seven AI labs tracked (OpenAI, Anthropic, xAI, ByteDance, Mistral AI, DeepSeek and Z.ai) publishes a company-wide figure for electricity, emissions or water. Every figure shown for them is an estimate, about 53 TWh of electricity in total. Much of their computing runs in cloud providers' data centers, whose reports include it without separating it out. Mistral AI has published a lifecycle analysis of one model, which is useful but does not cover the company as a whole.
Other figures
In each company's latest year, direct (Scope 1) and market-based Scope 2 emissions are company-reported for 16 of the 28 companies, Scope 3 emissions for 12 and water withdrawal for 11. The rest are estimates.
Reading industry totals
Any total for the AI industry's energy or water use combines reported and estimated figures, whoever publishes it. On this site each figure carries its confidence tier, so the split is visible in the company data. Companies are also graded from T-1 to T-5 on how much they disclose.