AI Footprint · 2026 figures
How much water and energy
does your AI use?
Enter what you do on a typical day. We total it up for a month and show it in everyday terms.
Where your footprint comes from
Each activity's share of your monthly total. Images and video use far more per item than text, so a few of them can outweigh hundreds of searches.
| Activity | Water / use | Water / mo | Energy / use | Energy / mo |
|---|
How the numbers work
Monthly = daily × 30
Your daily entries are multiplied by 30 days. The yearly line multiplies by 365.
Water is on-site cooling
Water counts what data centers use to keep chips cool, about 1 mL per Wh. Counting the water power plants use to make the electricity raises totals roughly 4–5×.
Ranges are wide
Figures vary with the model, image resolution, clip length, and data-center location. We use published medians or midpoints.
Not included
Model training, building the hardware, and your own device's power use.
Sources: Google, Gemini Apps environmental measurement (Aug 2025): median text prompt 0.24 Wh, 0.26 mL · Sam Altman, OpenAI (Jun 2025): average ChatGPT query 0.34 Wh, ~0.32 mL · Jegham et al., "How Hungry is AI?" arXiv 2505.09598, plus 2026 roundups: reasoning models use ~10–70× a standard prompt · Luccioni et al. (2024) and MIT Technology Review / Hugging Face (2025): image generation ~2.9 Wh average · Sustainable AI Group, "Lights, Camera, Carbon" (2026): 5-second video 57.5–114.8 Wh · Shaolei Ren, UC Riverside: water methodology and grid water factors.