Google has recently released precise statistics about the energy consumption of its Gemini AI.
According to the new study, each text prompt in Gemini consumes only 0.24 watt-hours of energy, equivalent to watching television for less than 9 seconds. Additionally, each query is responsible for producing 0.03 grams of carbon emissions and consumes 0.26 milliliters of water, roughly equivalent to five drops of water.
What’s even more impressive is the rapid improvement in Gemini’s efficiency. Over the past year, Google announced that energy consumption for each text prompt has decreased by 33 times, while the carbon footprint has been reduced by 44 times.
This progress is built upon Google’s continuous efforts to increase energy efficiency in its data centers.
In 2024, the company reduced carbon emissions from its data centers by 12 percent, even though electricity consumption increased by 27 percent.
Google says the secret to this success is efficient design across all parts of the system – from the AI models themselves to the hardware they run on.
Gemini models are built with structures like “Mixture-of-Experts,” which only activate the parts of the model needed to answer specific questions.
Google is also developing specialized hardware such as Tensor Processing Units (TPUs), which are far more efficient than traditional chips.
The latest TPU, codenamed Ironwood, is 30 times more energy efficient than the first one the company released publicly.
Cloudflare