US EditionGoogle Reveals Just How Much Energy Each Gemini Query Uses - but Is It Being Entirely Truthful?
Google reports its Gemini AI model uses 0.24 watt-hours and 0.26 milliliters of water per prompt, but experts argue these figures omit indirect emissions and scale effects.
- On August 25, 2025, Google published a report revealing that each typical Gemini AI query consumes approximately 0.24 watt-hours of electricity and 0.26 milliliters of water.
- The study focuses only on inference energy use for simple text prompts and excludes higher-demand workloads, omitting the effects of billions of daily AI queries.
- Google reports achieving a 33-fold decrease in energy consumption and a 44-fold reduction in CO₂ emissions for each AI prompt within the last year, enabled by AI chips that are 30 times more efficient and data centers operating with only 9% overhead.
- Experts like Shaolei Ren say, "They're just hiding the critical information," warning the study's limited data underrepresents AI's true environmental impact at scale.
- Despite efficiency gains, increasing AI demand may drive overall energy use and carbon emissions higher, reflecting the Jevons Paradox where improved efficiency leads to greater consumption.
11 Articles
11 Articles
Researchers and analysts have long been calling for greater transparency from the AI industry about its energy consumption, and now Google has lifted the veil of secrecy.
Google's total greenhouse gas emissions increased by 48% between 2019 and 2023. And most of this is related to the increase in calculation demand from 2022, which coincides with the advent of the general AI. However, Google relativizes. The company estimates that "the energy and water consumption habits of the AI data centres are exaggerated." According to Google, the processing of a prompt by Gemini now consumes the equivalent of 9 seconds of e…
The enormous power consumption of the AI data centers is always the subject of the attack. Google now refers as one of the largest providers in the area, however, that progress has been made and that demand has been reduced significantly. (Continue reading)
While the sustainability of artificial intelligence imposes itself as a central topic, Google publishes a methodology to quantify the environmental impact of inference. The figures provided reflect significant efficiency gains, but raise methodological criticisms. At the scale of enterprise projects, simulations show significant consumption volumes. Over the past two years, the sustainability of intelligence [...] The post Google publishes an al…
Coverage Details
Bias Distribution
- 100% of the sources are Center
Factuality
To view factuality data please Upgrade to Premium
