Agentic AI: 600x More Energy Than Simple Prompts

Agentic AI uses about 600 times more energy than simple AI prompts. This is a key difference from earlier estimates. Previous calculations suggested a median AI text prompt used very little energy. For example, a Google Gemini text prompt used 0.24 watt-hours, and an average ChatGPT query used about 0.34 watt-hours. These numbers are still accurate for those types of simple prompts. However, modern AI use, particularly with agentic tools like Claude Code and Codex, involves complex reasoning chains and many model calls per human prompt. One user's experience over eight weeks with 1,138 prompts in Claude Code led to over 14,000 model calls and an estimated 170 kilowatt-hours of electricity use. This translates to about 150 watt-hours per prompt, which is significantly higher. The bottom line is that the energy consumption of AI tasks varies enormously, with agentic workflows consuming substantially more power. This matters because it reframes the understanding of AI's overall energy footprint.