Definition
The electricity used to train AI models and answer requests, drawn almost entirely by the data centers housing the chips.
At a glance
- One typical chatbot question uses about 0.3 watt-hours — roughly an old Google search, not the once-popular 10x claim.[2]
- The cost is scale and training, not one question: training GPT-4 reportedly used about 50 gigawatt-hours.
- Data centers used about 415 TWh in 2024 (around 1.5% of global electricity); AI servers were about 15% of that.[1]
- That figure is projected to roughly double to about 945 TWh by 2030 — just under 3%.[3]
Where the energy goes
The work happens in data centers, not your device. Each high-end AI chip draws 250 to 700 watts, plus power and water for cooling.[4] AI energy use is really data center energy use.
What it means for a business
Using AI tools is a tiny direct cost, like other cloud software. The real issue is industry-wide demand straining local grids, which can lift prices and emissions in some regions. If sustainability matters, ask vendors about data center efficiency and clean power.
Bottom line
One AI question costs almost nothing — the footprint is scale, training, and cooling.
References
- Energy demand from AI - Energy and AI Analysis — International Energy Agency. IEA www.iea.org
- How much energy does ChatGPT use? — Epoch AI Epoch AI epoch.ai
- Global data center power demand to double by 2030 on AI surge — S&P Global. S&P Global www.spglobal.com
- We did the math on AI's energy footprint — MIT Technology Review. MIT Technology Review www.technologyreview.com
Comments
Questions, corrections, and links welcome. Be specific and civil.