Google releases research on Gemini’s AI environmental footprint

Google has released a report sharing the environmental footprint from Gemini estimating that a single text prompt uses about five drops of water or 0.26 millimetres. The report also stated that a text prompt for their AI chatbot consumes the same amount of electricity as watching TV for less than 9 seconds or around 0.24 watt-hours (Wh), which produces about 0.03 grams of carbon dioxide emissions. 

“Over a 12-month period, while delivering higher-quality responses, the median energy consumption and carbon footprint per Gemini Apps text prompt decreased by factors of 33x and 44x, respectively,” Ben Gomes, chief technologist of learning and sustainability wrote in a blog post. 

The company has claimed that the estimates are much lower than older research had shown due to these efficiencies.

However, the author of the one of the papers cited by Google, Shaolei Ren has responded to the study saying the comparisons made were misleading and the information in the report was incomplete. 

“While Google chose to consider its most recent 2025 onsite-only water, it used our *highest 2023 total number among the 18 locations* to create a misleading comparison, seemingly suggesting that the prior estimates were way off reality and that Gemini is much more efficient. For “onsite vs onsite” comparison, even based on our 2023 result and Google’s most recent 2025 result, the water by GPT-3 in 8 out of 18 locations in our paper is lower than Google’s reported global average for Gemini,” he wrote in a post on LinkedIn.

Google had omitted the indirect water usage in their estimates as compared to Ren’s paper which included “onsite water for cooling, offsite water for electricity generation, and the total,” he noted.