AI Agents Are Quietly Guzzling Massive Amounts of Energy

Here's the shift everyone's missing: AI used to mean asking a chatbot a question and getting one answer. Now the big labs are building "agents" — AI that can plan and act on its own, like an intern who keeps working after you go home. Ask an agent to build you a website, and it might run for hours, quietly re-prompting itself dozens of times to create pages, menus, and data behind the scenes.

That autonomy is the whole point — and the whole problem. A single agent task can burn hundreds of times more computing power than a one-line chatbot reply. OpenAI recently announced that a swarm of over 10,000 agents, sending 2.7 million messages, cracked a long-standing math problem. The feat was genuinely impressive, but the electricity bill was probably in the tens of millions of dollars. This is why tech companies are borrowing billions and building some of the world's biggest power plants.

Here's the catch: nobody outside these companies really knows how much energy agents use. AI firms are famously secretive about their environmental numbers. OpenAI's Sam Altman has compared one ChatGPT query to a tiny fraction of the water needed to grow a single almond — a comparison experts dispute. The bigger worry is scale. If one person can command thousands of AI workers running around the clock, energy use stops being tied to how many humans are typing.

So what does this mean for you? More AI capability, yes — but also rising electricity demand that could show up in your power bill, strain local grids, and complicate climate goals. Independent researchers are now trying to do the math themselves, because the companies won't. Until better disclosure arrives, treat the "AI is basically free" pitch with healthy skepticism.