Google claims its Gemini AI models use significantly less water per text prompt than previously estimated, citing software advancements that have reduced consumption to around 0.26 milliliters (ml).
Why it matters: The environmental impact of AI has been a growing concern, and Google’s report provides a rare glimpse into the resource usage of its AI operations, sparking a debate about the true water footprint of AI.
The details:
- Google states that its Gemini applications consume 0.24 watt-hours of electricity per text prompt, much lower than previous estimates by researchers.
- The company’s new test methodology focuses on on-site water usage, which it claims has been significantly reduced through software advancements.
- Google’s figures are being challenged by researchers who argue that the comparisons are misleading because they fail to account for the total water consumption involved in AI operations, including off-site usage.
- Off-site water consumption occurs in the generation of electricity that powers datacenters, whether it comes from gas, coal, or nuclear plants.
What they’re saying:
- “The issue isn’t that Google failed to consider off-site water consumption entirely. It’s that they made an apples-to-oranges comparison: Their new figure is for on-site consumption only, while the discredited figure included all water consumption,” said Shaolei Ren, associate professor of electrical and computer engineering at UC Riverside.
- “The critical flaw in the generalized UC Riverside study is that it assumes a grid powered predominantly by traditional, water-cooled thermoelectric plants, which does not hold true for Google’s data center operations,” said Ben Townsend, head of infrastructure strategy and sustainability at Google.
The other side: Researchers argue that even Google’s figures of 0.26ml per prompt are lower by orders of magnitude compared to the on-site average of 2.2ml per request found in US datacenters, based on 2023 data.
What’s next: The debate over the true environmental impact of AI operations is likely to continue as more companies release data and researchers scrutinize their claims. Transparency and consistent assessment methods will be crucial in understanding the full scope of AI’s water footprint.
Recent from X
As AI becomes more integrated into our lives, understanding its environmental footprint is essential. ⚡️
That’s why we’re sharing our comprehensive methodology for measuring the energy, emissions, and water impact of Gemini prompts. ↓ pic.twitter.com/ctGmHUDt44
— Google DeepMind (@GoogleDeepMind) August 22, 2025
We have data on the environmental impact per AI prompt:
Gemini: 0.00024 kWh & 0.26 mL water
ChatGPT: 0.0003 kWh & 0.38 mL
…the same energy as one Google search in 2008 & 6 drops of water.
Seems to be improving, too: Google reports a 33x drop in energy use per prompt in a year. pic.twitter.com/wcfMYA5rU5
— Ethan Mollick (@emollick) August 22, 2025
How much energy does our AI use? We did the math.
Today, a median Gemini Apps text prompt uses 0.24 watt-hours of energy and 0.26 milliliters of water. That's equivalent to watching TV for less than nine seconds and five drops of water.
Learn more → https://t.co/G7mYDioibm pic.twitter.com/zTwkaf0AbZ
— Google Cloud (@googlecloud) August 21, 2025
