Google has published a technical report on the energy usage of artificial intelligence queries. According to the company, an average query in Gemini applications consumes 0.24 watt-hours of electricity, which is roughly equivalent to running a standard microwave for one second.
“Tehsil365”, citing foreign sources, reports that this document represents a significant step toward understanding AI’s energy consumption. Until now, major technology companies had not disclosed such data. Google’s report covers energy demand comprehensively, accounting not only for AI chips but also for the supporting infrastructure.
According to the information, 58% of energy consumption goes to Google’s custom TPUs (Tensor Processing Units – AI chips), 25% to CPUs (Central Processing Units) and memory, 10% to backup equipment, and 8% to data center cooling and energy conversion.
Experts believe this report will serve as a key resource in the field of AI energy usage, as such precise measurements can only be conducted by large-scale companies.
While the figures released by Google do not reflect all queries, this move provides researchers with the long-awaited transparency.