So some Gemini uses more energy than that: Dan offers an example of dozens of books to feed in Gemini and to give a detailed summary of his content. “This is the kind of thing that will probably take more energy than a median prompt,” says Dean. The use of the reasoning model can also lead to high energy -related demand as these models take further steps before preparing the response.
This report was also strictly limited to the text indicators, so it does not represent what is needed to produce the image or video. (Other analysis, including in one MIT Technology ReviewEarlier this year, the hungry series of electricity, it shows that these tasks may require a lot of energy.)
The report also found that the total energy used to remove Gemini’s queries had dropped dramatically over time. According to Google, Medin Gemini used 33 times more energy than May 2025 in May 2024. The company refers to progress in the improvement of its models and other software improvements.
Google has also estimated the emission of greenhouse gas -related green house, which they have put on 0.03 grams of carbon dioxide. To reach this number, the company increased the total energy used to respond immediately through the average emissions of the average power emission.
Instead of using the US grid average, or the average -based emissions, where Google operates, the company uses market -based estimates, which keeps in mind the purchase of electricity that the company does with clean energy projects. The company has signed contracts to buy 22 gigwat Since 2010, power power from sources, including solar, wind, geothermal, and modern nuclear projects.
AI data centers also use cooling water, and Google estimates that each gesture uses 0.26 ml of water, or about five drops.
Dean says the target for the work was to provide consumers with AI to provide a window in the use of their interaction with AI.
“People are using (AI tools) for all kinds of things, and they should not have major concerns about the energy use or water use of gymnasium models, because in our original measure, what we were able to show was, in fact, it is in fact akin to working without thinking about it.”