AI Agents Use 136× More Energy Than Standard Chatbots, Study Finds

Autonomous AI agents—systems that can plan, use external tools, and solve complex tasks independently—consume dramatically more energy than conventional chatbots, according to the first comprehensive infrastructure analysis of these emerging systems.

Researchers at the Korea Advanced Institute of Science and Technology (KAIST) measured the computational cost, response latency, and energy consumption of AI agents in real-world service environments. Their findings reveal a stark efficiency gap: AI agents using a 70-billion-parameter language model consumed an average of 348.41 watt-hours per query, 136.5 times higher than simple question-answering systems.

The hidden cost of autonomous reasoning

The research team, led by Professor Minsoo Rhu of KAIST's School of Electrical Engineering, identified why AI agents demand so much more power. Unlike conventional large language models that generate a single response, AI agents repeatedly invoke language models throughout their execution as they plan steps, call external tools like web search or code execution environments, and coordinate multiple actions.

This repeated invocation pattern creates two critical inefficiencies. Response times can increase by up to 153.7 times compared to standard generative AI. Meanwhile, expensive graphics processing units sit idle for as much as 54.5 percent of total execution time while external tools perform their tasks—a new form of waste in AI infrastructure.

Projected data center demands

The KAIST team modeled a scenario matching current Google search volume: 13.7 billion AI agent requests per day. Under these conditions, data center power demand would reach approximately 198.9 gigawatts—roughly half the average power consumption of the United States and far exceeding the few-gigawatt scale of AI data centers currently under development.

These projections arrive as AI agents move from research labs into production environments for software development, research assistance, and workplace automation.

Why it matters

As enterprises adopt AI agents for complex workflows, infrastructure costs will become a competitive differentiator as significant as model performance. Organizations building or procuring AI capabilities must account for dramatically higher power consumption, longer response times, and GPU utilization challenges. The findings suggest that sustainable AI agent deployment will require coordinated optimization across models, semiconductors, data centers, and power infrastructure—not just better algorithms.

"This study is the first to quantitatively show not only how AI is becoming more intelligent, but also how much electricity and cost are required to implement and sustain that intelligence," Rhu said. He emphasized that integrated co-design approaches will be essential to reduce end-user costs while building sustainable AI infrastructure.

The research, with Ph.D. student Jiin Kim as first author, was presented at the IEEE International Symposium on High-Performance Computer Architecture in February. The team has released their AI agent implementations and benchmarks as open source to support follow-up research.

These findings were first reported by KAIST and detailed in a paper presented at HPCA 2026.