Is AI sustainable?
Does an AI prompt really use a bottle of water? We look at where the claim came from, newer data on AI water and energy use, and why better environmental reporting matters.
A line often quoted is that an AI prompt uses a bottle of water. This stems from a 2023 paper that estimated training GPT3 in Microsoft US data centres could evaporate 700,000 litres of freshwater. The same paper projected global AI-related water withdrawal could reach approximately 4.2–6.6 billion cubic metres by 2027.
In 2025 Google reported that its median Gemini text prompt model used approximately 0.24 Wh electricity, 0.03 g CO₂e and 0.26 mL direct cooling water. A later independent academic analysis expanded Google's accounting boundary to include estimated water consumed in electricity generation, raising the estimate from 0.26 mL to approximately 0.725 mL per median prompt.
Google compares the electricity used by its median prompt, 0.24 Wh, to less than nine seconds of television.
So, there is your bottle of water. But, is it the same bottle of water we started with?
How thirsty is AI? It depends on what, and how, you measure.
Can you avoid AI?
No, and yes-ish.
You can absolutely choose not to use generative-AI tools, but AI is increasingly embedded in other digital services.
For example in banking, email, games, phones, computers, security and many of the other digital services we use.
What does this have to do with streetwear?
It’s a question of sustainability. And reporting. AI is now firmly embedded across industries; fashion is no exception. From fake prom dresses to game-changing 3D design software, to the changing way we shop.
Sustainability, transparency and reporting are currently changing across the fashion industry. The EU, UK and other jurisdictions are developing and introducing measures aimed at greater environmental and supply-chain transparency.
Sustainability has been at the core of everything I have built from the start. There are gaps in our current AI environmental impact measuring, reporting and transparency. Until those are ironed out, we don’t have a clear picture.
While this isn't currently a requirement, and may never be, our global sustainable journey is trying to start, albeit through smoke and fire. Understanding the impact of what we do is as big a part of that as anything.
Is using AI sustainable or environmentally harmful?
If electricity per prompt is the metric we're using, Google's comparison puts it at less than nine seconds of TV. But electricity isn't the only metric.
Using AI can look about as environmentally dramatic as watching a few seconds of TV. Change the metric, and the picture changes.
AI is not going away
I hope we see none of the adverse AI futures we have been warned about. I fear we’ll see at least some of them. I'm not convinced that everyone in power has a vested interest in outcomes beyond their own. I'm not an optimist.
I, perhaps foolishly, hope that AI can help fix some of our biggest problems.
Until then, we need better data than water bottles and seconds of TV.
Sources
Li et al. (2023). “Making AI Less ‘Thirsty’: Uncovering and Addressing the Secret Water Footprint of AI Models”
Google (2025). “Measuring the environmental impact of AI inference”
“The Hidden Thirst of AI: A Framework for Estimating Direct, Indirect, and Scarcity-Adjusted Freshwater Consumption per LLM Query” (2026)