The Five Drops Fallacy: Unpacking Google’s Controversial Claims About AI’s Environmental Footprint

In a landmark disclosure that marks the first time a major tech company has released detailed metrics about AI’s resource consumption, Google recently announced that a typical text prompt to its Gemini AI assistant uses approximately five drops of water (0.26 milliliters) and consumes about as much electricity as watching television for nine seconds (0.24 watt-hours). While these figures might sound reassuringly minimal, the environmental impact report has sparked intense debate among experts who claim that Google’s approach to measuring AI’s environmental footprint is fundamentally misleading and potentially harmful to broader sustainability efforts.

Google’s technical paper, released as part of a broader transparency initiative, provides unprecedented insight into the company’s internal measurement methodology. The company claims its median Gemini text prompt produces just 0.03 grams of carbon dioxide equivalent emissions, painting a picture of an increasingly efficient technology with a negligible environmental impact. Furthermore, Google reports that through recent optimizations, it has achieved a remarkable 33-fold reduction in energy consumption per prompt over the past year, with carbon emissions dropping by 44 times during the same period.

“We wanted to be quite comprehensive in all the things we included,” said Jeff Dean, Google’s chief scientist, in an interview with MIT Technology Review (MIT Technology Review Magazine). The company’s assessment includes not only the energy used by AI-specific hardware (which accounts for 58% of the total) but also supporting infrastructure like host CPUs and memory (25%), idle machines for backup (10%), and data center overhead including cooling systems (8%). This approach, Google argues, provides the most complete view of AI’s overall environmental impact.

However, environmental researchers and AI sustainability experts have pushed back strongly against Google’s claims, highlighting critical omissions in the company’s methodology. Shaolei Ren, an associate professor of electrical and computer engineering at the University of California, Riverside, who has extensively studied AI’s water consumption, didn’t mince words: “They’re just hiding the critical information,” he told The Verge. “This really spreads the wrong message to the world.”

One of the most significant criticisms centers on Google’s decision to exclude indirect water usage from its calculations. While the company accounts for water directly used in data center cooling systems—a longstanding environmental concern—it omits the substantial water consumption associated with electricity generation. As Alex de Vries, founder of Digiconomist and a PhD candidate at Vrije Universiteit Amsterdam, explained, “A majority of the water a data center consumes stems from its electricity use—which Google overlooks in this study.” This omission leads to a severely incomplete picture, with de Vries noting, “You only see the tip of the iceberg, basically.” For those interested in personal water conservation, products like Water Purification Tablets highlight growing awareness of water as a precious resource.

Another contentious methodological choice involves how Google measures carbon emissions. The company reports only “market-based” emissions, which factor in Google’s investments in renewable energy projects and carbon offset initiatives. What’s missing is the “location-based” measure—the actual emissions associated with the power grids where Google’s data centers operate. This location-based metric typically shows higher emissions figures and offers more transparency about a company’s real environmental impact in specific regions.

“This is the ground truth,” Ren emphasized, pointing to internationally recognized standards set by the Greenhouse Gas Protocol that recommend reporting both metrics. By providing only the more favorable market-based figures, Google presents an artificially reduced environmental footprint that doesn’t accurately reflect the real-world impact of its operations.

Further complicating matters is Google’s use of median rather than average measurements, and its lack of transparency regarding how it calculated those medians. Ren points out that without knowing the word count or token length used for the median text prompts in Google’s study, it’s impossible to make meaningful comparisons with previous research. This makes Google’s claims that its water usage is “orders of magnitude less” than earlier estimates of up to 50ml per prompt misleading, as these earlier studies included both direct and indirect water consumption.

It’s worth noting that Google has yet to submit its paper for peer review, though a company spokesperson indicated they remain open to doing so in the future. The absence of external validation further compounds concerns about the reliability of the company’s claims.

Moreover, while Google celebrates the improved efficiency of individual Gemini prompts, this narrative obscures a more troubling big-picture reality. The company’s overall carbon emissions increased by 11 percent last year and have grown by 51 percent since 2019, coinciding with its aggressive push into artificial intelligence. This phenomenon, known as Jevons paradox, demonstrates how efficiency improvements at the individual level can still lead to greater overall resource consumption as usage expands. For those looking to better understand artificial intelligence, books like Mind Crime: The Moral Frontier of Artificial Intelligence explore these complex ethical dimensions.

“If you look at the total numbers that Google is posting, it’s actually really bad,” de Vries told The Verge. “This is not telling the complete story.”

The debate over Google’s environmental claims comes at a critical juncture for the AI industry. As concerns mount about the resource-intensive nature of large language models and their potential environmental impacts, there are growing calls for standardized, transparent reporting on AI’s energy and water consumption. Sasha Luccioni, an AI and climate researcher at Hugging Face, told MIT Technology Review that while Google’s disclosure represents progress, “People want to know what the cost is,” and a standardized AI energy score—something akin to the Energy Star rating for appliances (like energy-efficient appliances) would better serve public understanding.

What’s clear is that as AI continues to evolve and become more deeply integrated into our digital infrastructure, the environmental questions will only grow more pressing. Google’s unprecedented disclosure, despite its limitations, may signal the beginning of a more transparent era in how tech companies report their environmental impact. However, for that transparency to be meaningful, it must encompass the full scope of AI’s resource consumption—both direct and indirect—and acknowledge the cumulative effect of billions of those five-drop prompts adding up to oceans of water and significant carbon emissions. Building more efficient computing systems begins with better hardware choices, which even extends to components like computer hardware kits for DIY builders concerned about energy efficiency.

As we navigate the complex relationship between technological advancement and environmental sustainability, demanding complete and honest accounting from the companies building our AI future remains essential. Only then can we make truly informed decisions about the real costs and benefits of these transformative technologies.