Measuring the environmental impact of delivering AI at Google Scale
Summary
This paper proposes a comprehensive methodology to measure energy, carbon emissions, and water consumption for AI inference in production, using Google's Gemini Apps as a primary case study.
Key quotes
the median Gemini Apps text prompt consumes 0.24 Wh of energy—a figure substantially lower than many public estimates.
Google’s software efficiency efforts and clean energy procurement have driven a 33x reduction in energy consumption and a 44x reduction in carbon footprint for the median Gemini Apps text prompt over one year.
The report introduces a ‘Comprehensive Approach’ to measuring AI environmental impact, including active accelerators, host systems, idle capacity, and data center overhead (PUE). It compares these findings against narrower ‘Existing Approaches’ commonly found in academic literature.