Amid ongoing debates about the environmental impact of artificial intelligence, Google has released a study asserting that its Gemini AI assistant consumes minimal water and energy per text prompt. However, experts have raised concerns, labeling the tech giant’s claims as misleading.
According to Google, a typical Gemini text prompt uses approximately five drops of water, or 0.26 milliliters, and consumes about as much electricity as watching TV for less than nine seconds, equating to roughly 0.24 watt-hours (Wh). This results in approximately 0.03 grams of carbon dioxide emissions. These figures are significantly lower than previous research on the resource-intensive nature of data centers supporting generative AI models. Google attributes these reductions to efficiency improvements made over the past year. However, experts argue that Google’s study omits critical data, leading to an incomplete picture of Gemini’s environmental footprint.
Expert Criticism and Missing Metrics
Shaolei Ren, an associate professor of electrical and computer engineering at the University of California, Riverside, criticized Google’s approach. “They’re just hiding the critical information,” Ren stated, emphasizing the importance of transparency in environmental reporting. Ren, who has extensively studied AI-related water consumption and pollution, is also one of the authors cited in Google’s study.
Alex de Vries-Gao, founder of the website Digiconomist and a PhD candidate at Vrije Universiteit Amsterdam Institute for Environmental Studies, echoed these concerns. He noted that Google’s water usage estimate represents only “the tip of the iceberg.” De Vries-Gao has researched the energy demands of data centers used for cryptomining and AI, and he highlights the need for a more comprehensive understanding of AI’s environmental impact.
Google’s study also lacks a critical metric: a “location-based” measure of carbon emissions. This measure considers the energy mix of local power grids where data centers operate, offering a more accurate depiction of a company’s local environmental impact. Both Ren and de Vries-Gao advocate for the inclusion of this metric, in line with internationally recognized standards set by the Greenhouse Gas Protocol.
Comparisons and Methodological Concerns
Google’s paper cites previous research by Ren and de Vries-Gao, claiming it provides a more accurate representation of environmental impact than other studies lacking first-party data. However, Ren argues that Google is making an “apples-to-oranges comparison,” as previous research was based on averages rather than the median used by Google. Furthermore, Google did not disclose the data (word count or tokens for text prompts) used to calculate the median, raising questions about the study’s transparency.
Regarding water consumption, Google claims its finding of 0.26 ml per text prompt is “orders of magnitude less than previous estimates” that reached as high as 50 ml in Ren’s research. Ren contends this is misleading, as his research accounts for a data center’s total direct and indirect water consumption.
Google’s Response and Future Commitments
While Google has not submitted its study for peer review, spokesperson Mara Harris mentioned that the company is open to doing so in the future. Although Google declined to respond to further questions from The Verge, the study and accompanying blogs state that Google aims to enhance transparency regarding the water consumption, energy use, and carbon emissions of its AI chatbot. The company claims to provide more standardized parameters for measuring environmental impact, considering the energy used by idling machines and supporting infrastructure at data centers, such as cooling systems.
“While we’re proud of the innovation behind our efficiency gains so far, we’re committed to continuing substantial improvements in the years ahead,” Amin Vahdat, VP/GM of AI & Infrastructure for Google Cloud, and Jeff Dean, chief scientist of Google DeepMind and Google Research, stated in a blog post.
The Broader Environmental Picture
Google claims to have significantly improved the energy efficiency of a Gemini text prompt between May 2024 and May 2025, achieving a 33x reduction in electricity consumption per prompt. The company also reports a 44x reduction in the carbon footprint of a median prompt over the same period. These improvements explain why Google’s current estimates are much lower than those from previous years.
However, despite these efficiency gains, the broader environmental picture remains concerning. Efficiency improvements can paradoxically lead to increased pollution and resource use, a phenomenon known as Jevons paradox. According to Google’s latest sustainability report, the company’s “ambitions-based carbon emissions” increased by 11 percent last year and 51 percent since 2019, as Google continues its aggressive AI expansion. The report also notes that Google has started excluding certain categories of greenhouse gas emissions from its climate goals, citing them as “peripheral” or beyond the company’s direct control.
“If you look at the total numbers that Google is posting, it’s actually really bad,” de Vries-Gao remarked. Regarding the estimates released on Gemini, “this is not telling the complete story.”
As Google pledges to improve its environmental transparency, the debate over AI’s ecological impact continues, highlighting the need for comprehensive data and responsible corporate practices.
