Technology news outlet NeoWin published an analysis on August 21st revealing Google's recent environmental impact report for its Gemini AI model. The report claims each text request consumes approximately 0.26 milliliters of water (equivalent to five drops) while significantly reducing carbon emissions.
A group of environmental experts has challenged Google's interpretation of water usage statistics, highlighting the company's methodology only accounts for on-site data center cooling systems rather than including indirect water consumption from power generation facilities.
Google's newly released environmental report provides detailed metrics for the Gemini AI model's operational efficiency using a comprehensive "end-to-end methodology." Key figures disclosed by the tech giant include:
- Average water consumption of 0.26 milliliters per text query (approximately five drops)
- Energy consumption at 0.24 watt-hours, less than 9 seconds of television viewing
- Carbon emissions measuring 0.03 grams of CO2 equivalent
- Google reported total emissions reduced to 1/44th of previous levels between May 2024 and May 2025
Professor Shaolei Ren from University of California Riverside criticized Google's narrow focus, emphasizing the methodology omits significant water usage in power generation required to support data center operations. His prior research indicated training models like GPT-3 in premium U.S. data centers consumes up to 700,000 liters of water, with a 20-50 round ChatGPT conversation potentially requiring 500 milliliters of indirect water usage.
Experts also raised concerns about Google's use of "market benchmark method" for carbon accounting, which allows companies to offset purchased green energy certificates to lower emission figures. The more transparent "location-based benchmark method" calculates emissions based on actual grid carbon intensity, providing a more accurate regional environmental impact assessment. Critics argue Google's approach creates an artificially favorable emissions profile while obscuring real-world environmental consequences.
