AI Agents vs. Basic Chatbots: A 137x Hidden Energy Cost

A new, first-of-its-kind study from the Korea Advanced Institute of Science and Technology (KAIST) reveals that AI agents—the autonomous programs that can plan and execute tasks for you—use dramatically more energy than traditional chatbots. The findings, released on July 5, 2026, by the team led by Professor Yoo Min-soo, quantify this gap at roughly 136 to 137 times per query, raising urgent questions about future data center power demand and emissions . Here is what the research found and why it matters.

The Energy Gap: 348.41 Wh vs. 2.55 Wh

The study measured the resource consumption of existing AI agents and found a stark disparity. When using a 70-billion-parameter large language model (LLM)—the scale of today's most capable commercial services—an AI agent consumed an average of 348.41 watt-hours (Wh) per query. A traditional generative AI chatbot performing the same single-turn Q&A only used about 2.55 Wh. That is a multiple of roughly 136.5x to 136.6x . Some English-language reports round this up to