Cartoon: a small robot and a maker calmly sort colored squares into trays labeled DESK, HOME, ROAD, GRID and a separate tray labeled PAID ONCE, under a banner reading ONE ROOM AT A TIME.

THE ELECTRICITY CHART

the calm edition — one room at a time

0.3 Wh per AI chat query

About the same as one Google search, and less than a minute of your laptop. That's the whole marginal cost of asking.

~100 TWh per year, all AI data centers

Roughly three Irelands of electricity — about 0.3% of world generation, and growing fast. A real industrial load.

Both numbers are true. Most arguments quote one and ignore the other. Below, each scale gets its own small chart, measured against things from the same part of your life.

How to read this: every chart compares a few familiar things at one scale — bar length is energy, and every bar is labeled. Amber = AI you use, an ongoing cost. Blue = one-time training costs (their own section, near the end). Everything else is ordinary life, for scale. Hover any row for the fine print.
🎈 Prefer pictures? See the whole story as an illustrated poster — five leaps, almost no reading. the-electricity-chart-calm.pages.dev/poster
Fine print & sources

I am not an energy auditor. These are rough public estimates — real values vary by grid, hardware, climate, and who is arguing. Chatbot-query and training-run numbers are the most contested; queries use 2025-era measured estimates, training runs use the most-cited public figures. Please don't size a solar array off this page.

Google search (~0.3 Wh): Google (2009) · AI query (~0.3 Wh): Epoch AI (2025); Altman (0.34 Wh); the older ~3 Wh figure: de Vries, Joule (2023) · AI image: Luccioni, Jernite & Strubell (2023) · Streaming: IEA (2020) · Household & appliances: U.S. EIA / RECS · Gasoline = 33.7 kWh/gal: EPA · U.S. home (~10,600 kWh/yr) & per-person energy: EIA · Flights: ICCT · BLOOM 433 MWh (metered): Luccioni et al. (2022) · GPT-3 1,287 MWh: Patterson et al. (2021) · DeepSeek-V3: technical report (2024) · Llama 3.1: Meta model card · GPT-4 / Gemini / Grok estimates: Epoch AI · H100 node ~10.2 kW: NVIDIA · Bitcoin: Cambridge CBECI · U.S. data centers ~176 TWh (2023): LBNL · Global data centers & AI share: IEA, Energy and AI (2025) · U.S. & world generation: EIA / Ember · Eiffel Tower: official site · Hospitals: EIA CBECS · NYC subway: MTA · Google ops ~24 TWh (2023): Google Environmental Report · Ireland ~35 TWh: IEA.