Agentic AI Driving Massive Data Center Expansion

Silicon Valley is shifting its focus from simple chatbot queries toward "agentic AI"—systems based on large language models designed to make autonomous decisions to complete tasks. As Wired first reported, this shift is a primary driver behind the current rush to construct massive data centers and power plants, often funded by billions of dollars in debt.

Unlike standard queries, AI agents can generate hundreds of internal prompts to execute a single user request. Maxwell Zeff, writer of the Model Behavior newsletter, notes that an agent tasked with building a website might run for hours, re-prompting itself repeatedly to develop various datasets and features. This autonomous capability allows for a potential expansion of power use that is decoupled from the number of human users; Boris Gamazaychikov, CEO and co-founder of Sustainable AI, observes that AI leaders are envisioning "unicorns" operated by a single human employee supported by thousands of background agents.

The energy costs of these systems are significant. Wired reports that OpenAI recently utilized a swarm of over 10,000 agents sending 2.7 million messages to solve a complex math problem, a process Zeff estimates cost tens of millions of dollars in processing power. While OpenAI CEO Sam Altman has previously compared the water usage of individual ChatGPT queries to that of harvesting a single almond, these metrics are complicated by the intensity of agentic workflows. Climate scientist Zeke Hausfather recently estimated that his own daily use of Claude agents could consume more energy than two refrigerators.