The Real Cost of AI
Every AI answer draws on a physical supply chain of electricity and water. The figures below are pulled together from the IEA, EIA, Epoch AI and peer-reviewed research — each panel links back to its source so you can check the maths.
How to read this
Single-query numbers are estimates: providers rarely publish per-request energy, and efficiency is improving quickly even as models grow. The more robust story is in the aggregates — total data-center demand, and its share of national grids — where independent bodies broadly agree on the direction and rough magnitude.
Sources
Work with the data
Frequently asked questions
How much electricity does one AI query use?
Estimates vary with model size and task, but a generative-AI text answer is on the order of 3 watt-hours — roughly ten times a traditional web search. Longer reasoning answers, images and video cost substantially more. The infographic shows a comparison.
Where do these numbers come from?
The International Energy Agency (IEA) for grid and data-center totals, EIA for US electricity, Epoch AI for model training compute, and peer-reviewed papers (Luccioni et al. 2024 on inference energy, Li et al. 2023 on water use). Every panel links to its source.
Is AI a significant share of electricity demand?
Globally, data centers are around 1.5% of electricity use today and are projected to roughly double by 2030. In concentrated markets like Ireland the share is already above 20%. AI is the fastest-growing component.
Can I reuse these charts?
Yes, with attribution to the original sources listed on each panel. A link back to this page is appreciated.
Turn this into a saved workflow
Create a free account to save the steps from this guide as a reusable workflow and re-run it on any file, from any device.
Follow HowToCSV on Google
Add us as a preferred source on Google Search so our latest CSV guides and tutorials surface more often in your Top stories.