Never forget the 9-second detail when using AI

Google has released its most comprehensive field measurement yet of the environmental impact of AI chatbots. The report shows that a single chat request made through Gemini results in just 0.03 grams of carbon emissions and 0.26 milliliters of water consumption, which the company describes as equivalent to "five drops of water."

EQUAL TO 9 SECONDS

According to Google's measurements, an average prompt sent via Gemini consumes 0.24 watt-hours of energy—the same amount of electricity a modern television uses in just nine seconds.

The company stated that chat energy consumption decreased by 33-fold and carbon emissions by 44-fold between May 2024 and May 2025. According to the report, 23-fold savings achieved through software efficiency and model improvements played a critical role in this reduction. Furthermore, improved machine utilization achieved 1.4-fold energy efficiency, while clean energy procurement reduced emissions intensity by another 1.4-fold.

"THE FIRST MEASUREMENTS WERE NARROW-SCOPE"

Google engineers emphasized that many independent studies to date have only considered the energy consumption of active AI processors, a "narrow-scoped and misleading" approach. The company also included processor idleness, memory power, and data center cooling systems in its own methodology.

This means that while the narrow-scope method calculated 0.10 watt-hours of energy for the same conversation, Google's method calculated this figure as 0.24 watt-hours. The company stated that this difference highlights the importance of transparent measurement standards.

WAY BELOW EXPECTED

The report noted that the average water footprint of Gemini chats is 0.26 milliliters. This amount is hundreds of times lower than the 45 to 50 milliliters estimates some previous studies put forward. Google also noted that by 2023, new data centers in regions with high water stress will use air-cooled systems, reducing water consumption for AI services to "near zero."

A HARD FALL OCCURRED

According to Google data, Scope 2 emissions, which represent the indirect carbon footprint, have decreased by 47-fold in the last 12 months, while Scope 1 and 3 hardware-related emissions have decreased by 36-fold. This demonstrates that energy consumption and carbon emissions have become partially independent in AI services.

The report emphasized that while the impact of a single conversation is small, the total global impact should be carefully monitored due to the billions of users. Google engineers stated, "As AI's capabilities evolve, its environmental efficiency must also advance at the same pace."