New research has shown how AI agents can use more than 100 times the energy of conventional queries processed using generative AI.
Conducted by the Korea Advanced Institute of Science and Technology (KAIST), the study is claimed to be the first quantitative analysis of AI agents’ computational cost, response latency, energy consumption, and data-centre-scale power demand. An AI agent using a 70-billion-parameter LLM comparable to current commercial AI services was found to consume an average of 348.41 watt-hours per query: 136.5 times higher than the energy consumed by a conventional generative AI system performing simple question answering.
The study also revealed how AI agents baked inefficiency into their process by the way in which they operate. As AI agents repeatedly call LLMs while executing their multi-step tasks, their response latency increases significantly. The KAIST team found that response time can increase by up to 153.7 times, with GPUs sitting idle for as much as 54.5 per cent of the total execution time, while external tools including web search and code execution platforms perform their tasks. The work is published in IEEE Xplore.