Rough weekly counts
Most people should start by counting the habits they can remember: quick chats, long work sessions, AI images, and generated video seconds.
Personal use guide
Estimate your own AI use without fake precision.
Most consumer AI products do not show clean token totals. That does not mean you are stuck. Start rough: light chat, heavy research, image generation, video, and agentic tasks have different footprints.
Rough calculator
Start with your weekly AI habits.
Enter roughly how much AI you use in a normal week. The calculator turns that into an estimated electricity range so you can see the scale of different habits.
This is not a moral score and it is not exact. Provider hardware, model size, cooling, grid mix, cache hits, and utilization all matter, so the result is best read as an order-of-magnitude guide.
Start with counts you can remember: quick chats, long work sessions, images, and video. If you use an API dashboard, you can switch to a known token total under “advanced.” Current central assumptions: 0.30 Wh per 1,000 text tokens, 3 Wh per image, and 12 Wh per second of generated video. See the methodology, source logic, and review date.Advanced: use a known API token total
Your estimated weekly AI electricity use
Start here
Enter a few numbers above. The result will show a rough weekly electricity estimate and a plain-English comparison.
What you can actually enter today
Known token totals
If you use an API dashboard or another tool that already reports token usage, enter that total under “advanced.” Leave it blank if you do not know it.
Heavier uses
Do not treat all AI use as equal. Image, video, long reasoning, and agentic workflows can be far heavier than a short text answer.
Two practical ways to use this
1. Estimate from habits: count normal chats, heavy research sessions, images, and videos per week.
2. Use known tokens: if a dashboard already gives you token totals, use that number instead of guessing text volume.
What this tool does not do
This calculator does not import chat histories, connect to provider accounts, or analyze outside files for you. It is a lightweight estimator for rough scale, not a forensic audit.
Deeper analysis would need clearer privacy controls and a separate design.
How the calculator works right now
Text
If you leave tokens blank, the calculator estimates text volume as 1,200 tokens for a quick chat and 12,000 tokens for a long work session, then applies 0.30 Wh per 1,000 tokens.
Images
Each AI image is counted at 3 Wh as a rounded central estimate. Different image models vary a lot, so this should be read as a midpoint, not an exact figure.
Video
Each second of generated video is counted at 12 Wh. This is the most weakly constrained part of the model and is reviewed separately as better public evidence appears.