AI coding agents use roughly 600x more energy than a chat prompt

Climate scientist Zeke Hausfather tracked eight weeks of his own Claude Code usage: 3.2 billion tokens and about 170 kWh of data-center electricity. He estimates that comes to roughly 600 times the energy of a typical single AI chat prompt.

Why it matters: Official per-prompt energy figures from Google and OpenAI describe simple chat queries, not the multi-step, tool-calling agent sessions that are becoming the default way people use AI. As agentic coding tools spread, this suggests actual AI energy demand is being significantly understated by the numbers companies publicize.