Google has released a technical report that, for the first time, provides precise data on the amount of energy consumed by its Gemini artificial intelligence models when processing queries. According to the company, an average query requires 0.24 watt-hours of electricity – equivalent to the energy consumption of a standard microwave oven running for one second.
In addition to energy consumption figures, Google also published estimates of water usage and carbon dioxide emissions per single text query, becoming the first major technology company to present its environmental footprint in such a transparent manner.
The report states that an average query generates about 0.03 grams of carbon dioxide. This value was calculated by multiplying the total energy required to process the query with the average carbon intensity of electricity generation.
At the same time, the data centers powering artificial intelligence consume water for cooling. Estimates show that one query uses around 0.26 milliliters of water – roughly five drops.
More:
- European Parliament Calls for a Strategy on Energy Flexibility
- Nature and Climate Protection Takes Another Major Hit in EU Proposal to “Simplify” Agricultural Policy
- China Surpasses Wind and Solar Targets Ahead of Schedule – What Comes Next in Its Energy Transition?
The analysis also reveals that AI chips (TPUs) account for 58 percent of total electricity consumption, while processors and host machine memory contribute 25 percent. Backup equipment in standby mode consumes about 10 percent, and the remaining 8 percent is spent on overhead costs such as cooling and energy conversion in data centers.
“We wanted to be comprehensive in everything we included,” said Google’s Chief Scientist Jeff Dean in an interview with MIT Technology Review.
Experts note that this report marks an important step toward better understanding the environmental footprint of artificial intelligence.
However, the presented figures refer only to an “average query.” More complex tasks – such as processing and synthesizing content from dozens of books or running advanced reasoning models – can require significantly more energy.
Moreover, the report applies exclusively to text queries, while image and video generation demand much higher energy consumption, as shown in previous analyses.
Google also stated that the energy consumption of an average Gemini query in May 2025 was 33 times lower than a year earlier, thanks to model improvements and software optimizations.
Energy portal
