AI agents use roughly 600 times more energy than a simple chat prompt

Climate scientist Zeke Hausfather tracked his usage of Claude Code over eight weeks, consuming 3.2 billion tokens and about 170 kWh of data center electricity. His data reveals that the low figures reported by Google and OpenAI distort the reality of energy consumption in agent-based AI. Each prompt requires significantly more energy than previously assumed, highlighting the gap between reported figures and actual usage. This indicates that companies relying on AI must reassess their energy strategies, as environmental costs could be much higher than expected. These findings may also influence how AI models are developed in the future.