Energy consumption by AI agents remains opaque while driving data center construction
AI agents consume substantially more energy than ordinary chatbot queries, but companies do not disclose the actual figures. A swarm of 10 000+ agents from OpenAI sent 2.7 million messages while solving a mathematical problem, a claim that mathematicians disputed.
According to its own claims, OpenAI deployed a swarm of more than 10 000 AI agents that exchanged 2.7 million messages and solved a long-unsolved mathematical problem. Mathematicians disputed this claim. According to estimates mentioned in the source, this volume of messages required computing power that probably cost tens of millions of dollars in energy, but the source does not provide an exact figure.
AI agents differ from simple chatbot queries in that they generate dozens to hundreds of intermediate prompts themselves to complete a task and can run for hours without human intervention — for example, when building a website. This makes them substantially more energy-intensive than the conventional question-and-answer model. According to the source, companies have long failed to disclose the actual energy and environmental metrics of their products; Sam Altman, CEO of OpenAI, instead used a comparison stating that growing one almond requires as much water as 38 000 queries to ChatGPT — this calculation was disputed.
Climate scientist Zeke Hausfather tried to estimate consumption himself and calculated that his normal daily use of Claude, which makes extensive use of agents, could consume more energy than two running refrigerators. Boris Gamazaychikov, co-founder of Sustainable AI, said that the estimate by Hausfather was based on somewhat outdated data, and his organization is due to publish more accurate calculations of the environmental footprint of agents running on closed models later that same month.
Last week, Meta launched the personal AI agent Muse, which, according to the company, is intended to work for billions of people and maintain a dedicated computer in the cloud for each user even when the user is offline; the agent is to be connected to AI glasses from Meta at a later date. The source text is incomplete; its final section is missing.
Why it matters
The shift from simple queries to autonomous agents changes the order-of-magnitude basis for estimating the energy footprint of AI: an agent running for hours with dozens of intermediate steps consumes many times more than a single query, something that comparisons published so far (for example, consumption per ChatGPT query) do not address. Without credible data from companies, neither users nor regulators have a reliable basis for assessing the actual impact of deploying agents at scale, as planned by companies such as Meta.
Two audiences, two different impacts
What this means
For individuals
People who regularly use the agent features of AI tools (long autonomous tasks, not one-off queries) probably generate a significantly larger energy footprint than simplified company comparisons, such as energy consumption per query, suggest.
More practical updates →For a business
The lack of published data on the energy consumption of AI agents poses a risk of future regulation and a reputational burden for businesses, while capital expenditure on data centers and power plants is already growing regardless of this lack of transparency.
Risks and compliance More business impacts →Check the original
Event sources
only one source so far · 1 publisher, 1 independent. We count feeds from the same owner only once.