Evidence / elsworth2025measuring

Measuring the environmental impact of delivering AI at Google Scale

Elsworth, Cooper, Huang, Keguo, Patterson, David, Schneider, Ian, Sedivy, Robert, Goodman, Savannah, Townsend, Ben, Ranganathan, Parthasarathy
arXiv, 2025

reportedmachine checkedpreprint

Read the source

Authors beyond the eighth are not listed in the scout output. Company-measured production serving. The figure is 0.24 Wh per median Gemini Apps text prompt, not per token, so the card carries no energy-per-token finding; the atlas cites it for the 33x one-year reduction and for production readiness.

Findings

Per prompt, not per token: median Gemini Apps text prompt, full serving stack (0.24 Wh, about 864 J). No metric of the atlas is per prompt, so this finding carries no metric.

the median Gemini Apps text prompt consumes 0.24 Wh of energy

Cited by

Energy-efficient AI inference

Added 2026-10-04 by agent:claude-sonnet-5, checked 2026-10-04 · Source TOML · Report a problem