“People are often Curious about how much energy uses ChatGPT inquiry, ” Sam Altmanthe general manager of Openaiwrote aside in a Long blog pas Last week. The average survey, Altman wrote, uses 0.34 Wat hours of energy: “About what an oven would use in a little more than one second, or high efficient light would use in a few minutes.”
For a company with 800 million weekly active users (and Growing up), The question of how much energy all these searches are becoming increasingly pressing. But experts say altman’s figure does not mean much without much more public context of Openai about how it has arrived at this calculation – including the definition of what is an “average” survey, whether it includes or not an image, and whether Altman includes additional energy use, such as from training AI models and cooling servers.
As a result, Sasha Luccioni, the climate lead at AI Company hugging face, does not put too much stock in Altman’s number. “He could have taken it from his ass,” she says. (Openai did not respond to a request for more information on how it came to this number.)
As AI takes over our lives, it also promises to transform our energy systems, overloading carbon emissions right as we try to fight climate change. Now new and growing research is trying to put hard numbers on how much carbon we actually broadcast with all our AI use.
This effort is complicated by the fact that leading players like Openai reveals few environmental information. Analysis presented for Peer Review this week of Luccioni and three other authors look at the need for more environmental transparency in AI models. In Luccioni’s new analysis, she and her colleagues use data from OpenrouterLeader of a large language model (LLM) traffic, to find that 84 percent of LLM use in May 2025 was for models with zero environmental disclosure. This means that consumers are extremely chosen models with completely unknown environmental effects.
“It blows my mind that you can buy a car and know how many miles per gallon it consumes, yet we use all these ai -tools every day and we have no effective metrics, emission factors, nothing,” Luccioni says. “It’s not ordered, it’s not regulatory. Considering where we are with the climate crisis, it must be the peak for regulators everywhere.”
As a result of this lack of transparency, Luccioni says, the public is exposed to estimates that have no sense, but which are taken as the Gospel. You may have heard, for example, that the average ChatGPT request takes 10 times more energy like the average Google search. Luccioni and her colleagues track this claim to a public remark that John Hennessy, the president of an alphabet, the Google parent company, made in 2023.
Demand made by a board member of one company (Google) on the product of another company, to which he has no (openi) connection – however, Luccioni’s analysis, this figure has been repeated over and over again in the press and political reports. (As I wrote this piece, I got a pitch with this correct statistics.)
