Here's how much energy a single Gemini query consumes

Google has revealed fresh figures on the environmental footprint of its Gemini artificial intelligence (AI) models, claiming that each text query consumes just 0.24 watt-hours of energy, emits 0.03 grams of carbon dioxide equivalent, and uses 0.26 millilitres of water. The company equates this to watching television for less than nine seconds.

The data, published in new technical papers and blog posts by Google researchers, including Chief Scientist Jeff Dean, is part of a broader attempt to quantify the true costs of AI at scale. Until now, many estimates of AI’s environmental impact have relied on incomplete measurements, often excluding factors such as idle servers, cooling systems, or water consumption.

Google argues its new methodology offers a more comprehensive assessment of AI inference, the process of generating a response from a trained model, and hopes to encourage consistency across the industry.

At the same time, the company claims dramatic efficiency gains. Over the past year, the median energy use of a Gemini text prompt dropped by 33 times, while its carbon footprint fell by 44 times, even as output quality improved.

Google attributes the progress to advances across its entire technology stack: more efficient model architectures such as Mixture-of-Experts, optimised inference techniques like speculative decoding, custom-built Tensor Processing Units (TPUs), and improvements in data centre operations. Its newest TPU, Ironwood, is said to be 30 times more efficient than its first public TPU.

Despite these advances, energy demand continues to grow. In 2024, Google’s data centres saw electricity use rise 27 percent year-on-year, even as emissions fell 12 percent thanks to cleaner energy sources and efficiency gains. The company has pledged to run entirely on carbon-free energy and replenish more freshwater than it consumes in the long term.

As AI adoption accelerates, Google said it wants transparency to drive accountability. “By sharing our methodology, we aim to set a clearer benchmark for AI’s true energy footprint,” the company noted, positioning efficiency as critical to responsible AI development.