How much energy does the IA use? The people who know are not saying

“People are often Curious to know how much energy a chatgpt query uses, “wrote Sam Altman, CEO of Openii Long blog post last week. The average query, Altman wrote, uses 0.34 hours of energy: “About what an oven would use in just over a second, or a high efficiency bulb would use in a couple of minutes”.

For a company with 800 million active weekly users (e growing), the question of how much energy are using all these research is becoming increasingly pressing. But the experts say that the figure of Altman does not mean much without much more public context from Openi on how it has come to this calculation, including the definition of what is an “medium” query, regardless of whether or not it includes the generation of images and whether altman includes or not use of additional energy, such as the formation of the AI ​​and cooling models of the Openai servers.

As a result, Sasha Luccioni, the protagonist of the climate at the Ai Company who embraces the face, does not put too many actions to the Altman number. “He could have pulled him out of his ass,” he says. (Openii did not respond to a request for more information on how it came to this number.)

Since the IA takes over our life, it is also promising to transform our energy systems, infringement carbon emissions in the right way while we are trying to fight climate change. Now, a new and growing research corpus is trying to put hard numbers on how much carbon we are actually emitting with all our use of artificial intelligence.

This effort is complicated by the fact that the main actors such as Openi reveal little environmental information. An analysis presented for the revision of equal this week by Luccioni and three other authors examine the need for greater environmental transparency in the AI ​​models. In the new analysis of Luccioni, she and her colleagues use data from OpenRouterA ranking of the traffic of Big Language Model (LLM), to find that 84 percent of the use of LLM in May 2025 was for models with zero environmental dissemination. This means that consumers chose overwhelmingly models with completely unknown environmental impacts.

“It drives me crazy that you can buy a car and know how many miles for Gallone consumes, yet we use all these artificial intelligence tools every day and we have absolutely no efficiency metrics, emissions factors, nothing”, says Luccioni. “It is not mandatory, it is not regulatory. Given where we are with the climatic crisis, it should be at the top of the agenda for regulators from all over the world.”

As a result of this lack of transparency, says Luccioni, the public is exposed to estimates that make no sense but that are taken as a gospel. You may have heard, for example, that the average chatgpt request requires 10 times more energy than Google’s average research. Luccioni and his colleagues trace this claim of a public observation that John Hennessy, the president of Alphabet, the Mother company of Google, made in 2023.

A complaint made by a member of the Board of Directors of a company (Google) on the product of another company to which it has no report (Openii) is at most tenuous – again, there is Luccioni’s analysis, this figure has been repeated over and over again in press and politician relationships. (While I was writing this piece, I had a tone with this exact statistics.)