AI Agents Drive Data Center Boom With Massive Power Demands

The shift from chatbots to autonomous agents

Tech companies are investing billions in data center infrastructure not to power simple ChatGPT queries, but to support a fundamentally different class of AI systems called agents. These autonomous tools represent a dramatic escalation in computational demands that helps explain the current data center construction surge.

Unlike traditional chatbots that respond to single prompts, AI agents generate hundreds or thousands of internal prompts to complete complex tasks autonomously. When a user asks an agent to build a website, for example, the system might run for hours, re-prompting itself dozens of times to construct different pages, menus, and datasets, according to WIRED's Maxwell Zeff.

OpenAI recently demonstrated this scale by deploying more than 10,000 agents that exchanged 2.7 million messages to tackle a longstanding mathematics problem. While mathematicians disputed the company's claims about solving the problem, the computational cost was undeniable—likely tens of millions of dollars in processing power.

Why it matters

The energy implications of widespread agent deployment could dwarf current AI power consumption. Meta's recently announced Muse agent will maintain a dedicated cloud computer for each user that operates even when offline, with plans to integrate the technology into AI glasses. If billions of users begin outsourcing tasks to agents without realizing it, the infrastructure requirements explain projects like Meta's Hyperion data center in Louisiana, which will be powered by 10 natural gas plants. This represents a fundamental decoupling of AI energy use from direct user activity—creating potentially limitless expansion as tasks grow more complex.

The opacity problem

Private AI companies have historically disclosed little about their environmental footprint, often citing individual query metrics that obscure the full picture. OpenAI CEO Sam Altman recently claimed that harvesting a single almond requires water equivalent to 38,000 ChatGPT queries—a calculation that has been disputed.

These comparisons become meaningless when applied to agents. Climate scientist Zeke Hausfather calculated that his daily use of Claude agents may consume more energy than running two refrigerators. Boris Gamazaychikov, CEO of research group Sustainable AI, notes that agent energy use ranges dramatically depending on task complexity, from simple jobs to full-day autonomous coding sessions involving parallel "helper" agents.

"In other technological growth areas, we're constrained by how many people are driving a car or streaming Netflix," Gamazaychikov told WIRED. "Now, this stuff is kind of decoupled from users."

Gamazaychikov's organization plans to release more precise calculations on agent environmental footprints later this month, though the research landscape remains sparse due to company opacity.

Building for a different future

The data centers under construction today are designed for technology that won't arrive for three to five years, Gamazaychikov notes. AI industry leaders envision companies with single human employees supported by hundreds or thousands of AI agents—a vision that requires infrastructure on an entirely different scale than current applications.

While individual agent use may produce relatively modest emissions compared to activities like frequent air travel, the aggregate impact of billions of users each maintaining always-on agents represents a significant new emissions source at a time when global temperature goals are increasingly off track.

These details were first reported by Molly Taft in WIRED's Power Play column.