South Korean Study Warns AI Agents Could Overwhelm Global Power Grids

A recent study by South Korea's KAIST university reveals the enormous energy demands of autonomous artificial intelligence agents, highlighting a potential global electricity crisis. Unlike traditional chatbots that provide a single response, these AI agents operate through multiple iterative steps, planning, searching, computing, and decision-making, resulting in energy consumption up to 136.5 times higher per query. For example, a complex query processed by a large AI model with 70 billion parameters can consume approximately 348 watt-hours on average.

Processing times for these AI agents can be 150 times longer than standard chatbots, with significant delays caused by waiting for external data sources while GPUs continue to draw power. Researchers estimate that if daily requests to AI agents reach 13.7 billion, data centers' electricity consumption could hit 199 gigawatts, nearly half of the entire United States' power usage.

To put this in perspective, U.S. data centers consumed about 176 terawatt-hours in 2023, roughly 4.4% of national electricity use. The International Energy Agency projects global electricity demand to double by 2030, and the rise of autonomous AI could accelerate this trend beyond current expectations. KAIST researchers caution that software improvements alone won't suffice to address this challenge. They call for fundamental changes including more energy-efficient AI models, low-power chips, and upgraded electrical infrastructure.

As the global race to develop autonomous AI agents intensifies, the study underscores a critical question: can the world afford the energy costs of deploying these advanced technologies at scale?