Short answer
A single AI query may consume only a small amount of electricity individually, but modern AI platforms process millions or billions of requests every day. At global scale, this results in significant electricity demand.
AI inference requires computation
Every AI prompt triggers inference: the process where GPUs calculate predictions token by token. Large language models continuously perform mathematical operations across billions of parameters in order to generate responses.
GPUs consume large amounts of power
Modern AI systems rely on specialized GPUs designed for parallel computation. High-end AI accelerators can consume hundreds of watts each, especially when operating continuously inside large data centers.
Image generation uses even more electricity
Generating AI images generally requires more computation than simple text generation. Diffusion models repeatedly refine images through many processing steps, increasing GPU usage and electricity consumption.
The infrastructure around AI also consumes energy
Electricity is not only used for GPUs themselves. AI infrastructure also requires networking equipment, storage systems, cooling systems, backup power systems, and data center operations.