AI’s Energy and Water Footprint, Explained: What a Prompt Actually Costs
Google says a median AI prompt uses 0.24 Wh and 0.26 mL of water. Data centers used 415 TWh in 2024. Here is the data, plus a calculator for your own use.
WINSS has covered AI extensively as a tool for sustainability — including 100 AI solutions driving sustainability across industries and KLM’s AI-driven food waste reduction. This piece looks at the reverse question: what AI itself costs in electricity and water. Global data centers consumed approximately 415 terawatt-hours (TWh) of electricity in 2024, about 1.5% of world electricity consumption, according to the International Energy Agency’s “Energy and AI” report. The IEA projects that figure will more than double to around 945 TWh by 2030 — roughly Japan’s entire current electricity consumption — with AI as “the most important driver of this growth.”
What a single AI prompt costs, according to Google
On August 21, 2025, Google published what it described as a full-stack measurement of the median Gemini Apps text prompt, in a post co-authored by Amin Vahdat, SVP and Chief Technologist for AI and Infrastructure, and Jeff Dean, Chief Scientist at Google DeepMind. The disclosed figures: 0.24 watt-hours (Wh) of energy, 0.03 grams of CO2-equivalent emissions, and 0.26 milliliters of water — an amount Google compared to “less than nine seconds” of television viewing in energy terms. Google’s methodology counted full system dynamic power at production-scale chip utilization, idle machines kept provisioned for reliability, CPU and RAM consumption beyond the AI accelerator chips themselves, and data center overhead including cooling and power distribution. Google also reported that the same median prompt’s energy use fell 33-fold and its total carbon footprint fell 44-fold between May 2024 and May 2025, a period during which the company says response quality also improved — evidence that per-query efficiency and total sector-wide electricity demand are moving in opposite directions simultaneously, since aggregate use is rising even as individual query costs fall.
Where the water goes
Water figures in AI’s footprint mainly through cooling: data centers use water-based cooling systems to manage heat generated by dense server and chip clusters, and some also draw on water indirectly through the electricity grid, since certain power generation methods themselves consume water. Microsoft’s own 2024 Environmental Sustainability Report data fact sheet recorded total water consumption of 7,844 megaliters in fiscal year 2023, up 23% from 6,399 megaliters in fiscal year 2022, with total water withdrawal rising 21% to 12,951 megaliters over the same period; the company reported that 41% of FY23 withdrawals came from areas already experiencing water stress. Microsoft has since published efficiency-focused updates, including a December 2024 announcement of next-generation data center designs intended to use zero water for cooling, though these apply to new facilities rather than the existing fleet reflected in the FY23 figures above.
Calculate your own AI usage footprint
Using Google’s disclosed median per-prompt figures, this tool estimates the energy, water, and CO2 footprint of your own AI prompt usage. It is a simple linear estimate based on one company’s disclosed median — actual per-prompt costs vary by model size, prompt length, and provider, so treat the result as an order-of-magnitude estimate rather than a precise personal footprint.
The policy angle OECD tracks
Beyond company-level disclosures, the OECD.AI Policy Observatory tracks how governments are approaching AI’s energy and environmental footprint at a policy level, including which countries are moving toward mandatory data center energy or water reporting rather than relying on voluntary corporate disclosures like Google’s. This is a useful complementary angle to the company-specific figures above, since it captures the regulatory direction rather than any single company’s numbers.
What remains unresolved
Google’s disclosed figures apply specifically to Gemini Apps text prompts and its own infrastructure; the company has not published equivalent per-prompt figures for image or video generation, which independent researchers generally find far more energy-intensive than text, nor has it disclosed how the figure varies for longer or more complex prompts. The IEA’s 945 TWh 2030 projection depends on assumptions about AI adoption rates and efficiency gains that the agency itself frames as a base scenario rather than a certainty, and Microsoft’s most recently published detailed water figures are for fiscal year 2023, meaning more current company-level water data was not identified for this article.
Sources: Energy and AI, full-stack measurement of the median Gemini Apps text prompt, 2024 Environmental Sustainability Report data fact sheet, OECD.AI Policy Observatory
Featured image: photo by panumas nikhomkhai on Pexels (free Pexels license).
Become a Sponsor
Our website is the heart of the mission of WINSS – it’s where we share updates, publish research, highlight community impact, and connect with supporters around the world. To keep this essential platform running, updated, and accessible, we rely on the generosity of you, who believe in our work.
We offer the option to sponsor monthly, or just once choosing the amount of your choice. If you run a company, please contact us via info@winssolutions.org.
I specialize in sustainability education, curriculum co-creation, and early-stage project strategy. At WINSS, I craft articles on sustainability, transformative AI, and related topics. When I’m not writing, you’ll find me chasing the perfect sushi roll, exploring cities around the globe, or unwinding with my dog Puffy — the world’s most loyal sidekick.
