Google revealed in a blog post the resource consumption rates of its AI model ‘Gemini’, including water, energy, and carbon footprint, according to a report by Ars Technica. The company states that each text command to the ‘Gemini’ AI model consumes 0.24 watt-hours of energy, equivalent to 9 seconds of watching television. This report comes amid growing concerns about AI and data centers’ impact on global energy consumption and environmental damage. It is the first transparent report addressing these concerns and discloses the actual environmental impact of AI technologies. Jeff Dean, Google’s chief scientist, said, “We wanted to be completely comprehensive in everything we included.” The report details that AI chips consume 58% of the total energy, RAM uses 25%, backup systems 10%, and other data center systems like cooling consume 8%.

The energy consumption of the ‘Gemini’ model has decreased 33 times compared to last year, significantly reducing the company’s overall carbon footprint and that of the AI model specifically. The report does not specify the total number of commands sent to ‘Gemini’, making it difficult to determine the model’s full consumption. It also only mentions energy consumption for text generation commands, not for video or image generation.