“`html
Gemini AI models from Google have recently unveiled important statistics regarding their environmental impact. According to the data, each text query requires merely 0.24 watt-hours of energy, results in an emission of 0.03 grams of carbon dioxide equivalent, and consumes 0.26 millilitres of water. This is equivalent to watching television for under nine seconds.
The findings, shared in new technical papers and blog posts by Google researchers including Chief Scientist Jeff Dean, form part of a larger initiative aimed at accurately measuring the environmental costs of AI at scale. Previous assessments of AI’s ecological implications often relied on incomplete data, frequently neglecting elements like idle servers, cooling systems, or water usage.
Google asserts that its updated methodology delivers a more thorough evaluation of AI inference, the procedure of generating a response from a trained model, and hopes to foster consistency throughout the industry.
Simultaneously, the company reports remarkable efficiency improvements. In the past year, the median energy consumption of a Gemini text prompt decreased by a staggering 33 times, while its carbon emissions dropped by 44 times, all while the quality of output has improved.
The advancements are credited to enhancements across Google’s complete technological framework: more efficient model structures like Mixture-of-Experts, optimised inference techniques such as speculative decoding, custom-designed Tensor Processing Units (TPUs), and better data centre operations. The latest TPU, named Ironwood, is purported to be 30 times more efficient than its initial public TPU.
Yet, despite these strides, energy demands continue to rise. In 2024, electricity consumption in Google’s data centres increased by 27 percent year-on-year, even as emissions decreased by 12 percent, thanks to cleaner energy sources and improved efficiency. The company has committed to operating entirely on carbon-free energy and to replenishing more freshwater than it consumes over time.
As AI adoption accelerates, Google emphasises the importance of transparency for accountability. The company stated its intention to share its methodology to establish a clearer standard for AI’s genuine energy footprint, highlighting efficiency as vital to the responsible development of AI.
“`
