Overview
- Google estimates a median Gemini text prompt uses 0.24 watt‑hours of electricity, emits 0.03 gCO2e, and consumes 0.26 mL of water, roughly five drops.
- The company reports about a 33x reduction in energy and a 44x reduction in carbon per median prompt compared with a year earlier.
- Its "full‑stack" methodology includes accelerator power, host CPU and memory, idle capacity, and data center overhead such as cooling and power conversion.
- The study covers text inference only and excludes training and more energy‑intensive tasks like image or video generation, so totals for overall AI use remain unknown.
- Experts criticize the omission of indirect water use and location‑based emissions and note the reliance on median reporting; the paper has not been independently peer‑reviewed.