AI Agents Drive Data Center Power Boom — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- AI agents are autonomous LLM-based systems that give themselves hundreds of self-prompts to execute tasks like building a website, replacing simple chatbot queries as the dominant use case for frontier AI labs.
- OpenAI announced that a swarm of more than 10,000 agents sending 2.7 million messages solved a longstanding math problem, burning through tens of millions of dollars in processing power, per colleague Maxwell Zeff.
- OpenAI CEO Sam Altman claimed in a podcast that 38,000 ChatGPT queries use the water needed to harvest a single almond — a disputed metric that obscures the far heavier energy load of agentic AI.
- Climate scientist Zeke Hausfather calculated his own daily Claude session using agents may consume more energy than powering two refrigerators, calling AI use a "net new source of emissions" as emissions targets slip.
- Meta rolled out Muse, a personal AI agent with a dedicated cloud computer per user built for "billions of people," with plans to integrate it with Meta's AI glasses later this year.
- Meta's Hyperion data center in Louisiana will be powered by 10 natural gas plants, illustrating the fossil-fuel infrastructure scale that agentic AI for billions of users could require.
- Small modular nuclear reactors could supply carbon-free power to data centers, but none operate commercially in the US and only one model is licensed for sale, pushing developers to install gas turbines now.
Why it matters: As OpenAI and Meta race toward agentic AI for billions of users while disclosing almost nothing about energy use, the emissions footprint of mainstream AI will hinge on whether carbon-free alternatives like small modular nuclear reactors scale fast enough to offset fossil-fuel buildouts like Meta's 10-plant Louisiana campus.
Ask SkimNews
