Short answer
ChatGPT uses electricity whenever it processes prompts and generates responses. The exact amount is not publicly disclosed in real time, so any public estimate should be treated as an approximation rather than an audited measurement.
Most ChatGPT usage is inference
When a user sends a prompt, the model performs inference: it processes the input, predicts likely output tokens and returns a response. Each interaction requires compute, and that compute runs on servers equipped with specialized AI accelerators.
Electricity use grows with daily activity
A single request may be small compared with industrial energy use, but ChatGPT operates at global scale. Millions of daily users, repeated prompts, longer conversations and multimodal features can turn small per-request energy costs into meaningful infrastructure demand.
Data centers add overhead
The electricity footprint is not limited to the AI chip itself. Supporting servers, memory, networking, storage, power delivery and cooling systems also consume energy. This is why data center efficiency matters when estimating the electricity used by AI services.
Why estimates vary
Different estimates depend on assumptions about model architecture, hardware type, batching, response length, utilization, data center efficiency and electricity sourcing. TheAIMeters presents directional estimates and links them to a transparent methodology rather than claiming exact real-time measurement. See the Methodology.