Google has released new research quantifying the environmental footprint of its Gemini artificial intelligence (AI) models, offering a rare, detailed look at the energy, carbon, and water costs of large-scale AI systems.
According to the company, each Gemini 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—an impact Google equates to watching television for less than nine seconds. The findings were outlined in newly published technical papers and blog posts by Google researchers, including Chief Scientist Jeff Dean.
The company said its methodology provides a more comprehensive view of AI inference—the process of generating responses from trained models—by including factors often overlooked in previous assessments, such as idle server use, cooling infrastructure, and water consumption. Google hopes these disclosures will help establish industry-wide benchmarks for measuring AI’s true environmental cost.
At the same time, Google highlighted significant efficiency improvements. Over the past year, the median energy use per Gemini prompt dropped 33-fold, while its carbon footprint fell 44-fold, even as model performance improved. These gains stem from multiple advancements: more efficient model designs like Mixture-of-Experts, optimized inference techniques such as speculative decoding, custom-built Tensor Processing Units (TPUs), and improved data centre operations. Google’s latest TPU, Ironwood, is reportedly 30 times more efficient than its first-generation hardware.
Despite these advancements, overall energy demand is still climbing. In 2024, Google’s data centres recorded a 27% increase in electricity consumption, though overall emissions declined by 12%, aided by cleaner energy sources and operational efficiencies. The company has pledged to operate entirely on carbon-free energy and to replenish more freshwater than it consumes over time.
“By sharing our methodology, we aim to set a clearer benchmark for AI’s true energy footprint,” Google stated, underscoring its view that transparency and efficiency are central to responsible AI development as global adoption accelerates.
Recent Random Post:
