AI agents are driving a new wave of data center and power building

In short: More AI companies are building and using “AI agents” that can run long, multi step tasks, and that is increasing demand for data centers and electricity.

What's going on

People often think of AI as a chatbot that answers one question at a time. Wired reports that many AI labs are moving toward “agents,” which are AI systems that can make their own small decisions while working on a bigger job (like a digital assistant that keeps going without you watching every step).

Instead of one quick question, an agent might create hundreds of follow up instructions for itself. For example, if you ask an agent to build a website, it might work for hours, repeatedly checking its own work and generating new pages and features. That extra work needs far more computer processing, which means more electricity and often more water for cooling.

Wired points to an OpenAI example where more than 10,000 agents sent 2.7 million messages while working on a math problem. Some mathematicians disputed the company’s claims about the result, but the attempt still shows how much computing power these “swarms” can consume. Climate scientist Zeke Hausfather also estimated that his daily use of agent style tools could use more energy than running two refrigerators.

What to watch

A key issue is that the public has limited, reliable data on how much energy agents use, since companies often share only partial information. Another thing to watch is wider rollout. Meta has introduced a personal agent called Muse and said it is designed to scale to billions of people, with a “dedicated computer in the cloud” for each user. If agents become a normal background feature, like email syncing, the push to build more data centers and new power sources, including natural gas plants and possibly nuclear later, may grow quickly.

Source: Wired