Google has disclosed the resource consumption and environmental impact of its AI model, Gemini, in a blog post shared on its official platform, as reported by tech site Ars Technica. The report covers water usage, energy consumption, and the carbon footprint associated with running the AI system.
According to the report, each text prompt submitted to Gemini consumes 0.24 watt-hours of energy, roughly equivalent to 9 seconds of watching television. This comes amid growing concerns over the environmental impact of AI technologies and the energy demands of the data centers that power them.
Google’s initiative marks one of the first transparent reports addressing these concerns and revealing the real environmental footprint of AI systems. Jeff Dean, Google’s lead AI scientist, emphasized, “We wanted to be fully comprehensive in everything we included.”
Breakdown of Energy Consumption
- AI chips: 58% of total energy consumption (0.24 watt-hours per text prompt)
- RAM (memory): 25%
- Backup systems: 10%
- Other data center systems (cooling, etc.): 8%
The report highlights that Gemini’s energy consumption has decreased 33-fold compared to last year, resulting in a significant reduction in Google’s overall carbon footprint and that of the AI model specifically.
However, the report does not disclose the total number of prompts processed, making it difficult to determine the full scale of energy consumption. Additionally, the figures only account for text generation and do not include AI-generated videos or images.
This report reflects Google’s ongoing efforts to address sustainability concerns in AI development and offers insight into how technological advances can reduce environmental impact without compromising performance.
