AI Agents and the Data Center Power Crunch

Tech companies are pouring billions into data centers and power plants, and the reason isn't simple chatbot queries. The shift toward AI agents-large language model-based systems that make autonomous decisions to complete tasks-is a major driver of Silicon Valley's energy buildout.

Unlike a single chatbot question, agents can generate hundreds of prompts from one user request, running for hours and re-prompting themselves repeatedly. OpenAI recently said a swarm of over 10,000 agents sending 2.7 million messages solved a longstanding math problem, though mathematicians disputed the claim.

Reliable data on agents' energy use is scarce. Climate scientist Zeke Hausfather calculated that his daily Claude sessions could consume more energy than two refrigerators. Boris Gamazaychikov of Sustainable AI noted Hausfather's math relied on somewhat outdated findings, and his group plans research with more precise figures.

Meta's new personal agent, Muse, promises a dedicated cloud computer for each user, working even when they're offline-a sign of the scale behind projects like Louisiana's Hyperion, powered by 10 natural gas plants. As Gamazaychikov put it, the technology being trained now is three to five years away and will be a very different flavor than a chatbot window.

Source: Wired science