Molly Taft, senior writer for Wired, notes that tech companies are taking on billions in debt to construct massive power plants for data centers. This buildout is driven by a shift from simple chatbots to autonomous AI agents. These systems are designed to make independent decisions to execute complex tasks.
Maxwell Zeff, author of the Model Behavior newsletter, explains that agents do not just answer questions. Instead, they generate hundreds of small prompts based on a user’s initial request. If a user asks an agent to build a website, the system might run for hours. It re-prompts itself dozens of times to create features, web pages, menus, and the datasets required to power the site.
OpenAI recently announced that a swarm of more than 10,000 agents sent 2.7 million messages to solve a longstanding math problem. Mathematicians pushed back on the company’s claims. While labs are committed to solving unsolvable problems, the process consumes significant processing power. Max estimates this equates to tens of millions of dollars worth of energy, though exact figures remain difficult to pin down.
Private AI companies have historically been selective about disclosing environmental metrics. Many CEOs point to single queries as a measure of resource use. In a recent podcast interview, OpenAI CEO Sam Altman claimed the water needed to harvest one almond equals 38,000 ChatGPT queries. The calculation has been disputed.
“The people that are scarfing down 12 almonds at a time don’t feel like they’re doing something horrible from a water perspective for the most part,” he said.
Introducing agents, which are far more energy-intensive than simple queries, complicates these calculations. There is a major lack of information regarding the energy use of agents. Their tasks range from simple jobs to a full day of autonomous coding involving parallel “helper” agents. The power use gap between these applications is vast, and expansion could be limitless as tasks grow more complex.
“In other technological growth areas, we’re constrained by how many people are driving a car or streaming Netflix,” says Boris Gamazaychikov, co-founder and CEO of Sustainable AI. “Now, this stuff is kind of decoupled from users—and if you listen to AI leaders, I think that’s what they want. They’re talking about unicorns that have one employee.”
Well, one human employee. In that imagined world, hundreds or even thousands of AI agents could work in the background. Suffice to say that is the future AI companies are working toward, which explains the rush to build data centers.
With little reliable data from companies, some enthusiasts are trying to calculate energy use themselves. Last month, climate scientist Zeke Hausfather wrote a blog post calculating his own AI consumption. He used various sources to determine his average daily Claude session may consume more energy than two refrigerators. Gamazaychikov noted Hausfather made a good effort, but the math relied on somewhat outdated findings. This is unsurprising given how little academic work exists on the topic and how opaque tech companies are about emissions metrics.
Hausfather concludes that his personal AI use is not a world-ending number. However, it represents a net new source of emissions while global temperatures skyrocket and reduction goals fall off track. It is a lot bigger than the fraction-of-an-almond-sized numbers Altman uses.
Hausfather says he uses agentic tools more than most people, but that could change soon. Last week, Meta rolled out a personal AI agent called Muse. The company said in a press release it is “built to work for billions of people worldwide.” Muse will maintain a “dedicated computer in the cloud” for each user, even when offline. Meta plans to integrate Muse with its AI glasses later this year. It is very possible that in the near future, Meta users playing with their glasses or browsing Facebook will outsource tasks to agents without realizing what they are doing.
Compared to regular flights or eating beef daily, the carbon footprint for personal agentic use is still relatively small. But if Meta envisions a future where everyone uses an agent, it explains the massive scale of data centers like the Hyperion project in Louisiana. That facility will be powered by 10 natural gas plants.
“The technology that’s going to be trained by the data centers that are being proposed and built right now is three to five years away,” Gamazaychikov says. “It’s going to be a very different flavor than just the chatbot window.”
What You’re Asking
A reader asks if small nuclear power plants would work for data centers.
The short answer is yes. Small modular reactors could be a great choice for carbon-free power. Startups and developers envision a future where data centers couple with these reactors running off the electric grid.
The problem is speed. No small modular reactors are operating commercially in the US. Just one model has been licensed for sale, despite decades of development. The Trump administration is trying to help the industry mature rapidly, including creating a pilot project in the Department of Energy for 11 startups to hit a key milestone this year. At least a handful have succeeded. Now they begin the long journey to market.
Many data center developers do not want to wait years for these companies to prove themselves. They are installing gas turbines now. In other words, they are not waiting for utopia to come around. We will see what happens in the next few years.
What We’re Reading
For Scientific American, Austyn Gaffney travels to Memphis to document the backlash against SpaceX’s data centers.
The Wall Street Journal reports on how some states giving tax breaks for data centers are now walking back on those deals.
Texas’s KERA News covers how data centers are making some Republican voters consider leaving the party.
