How much water is hidden in your AI Ghibli image?
Following trends, consuming water: the hidden silk of artificial intelligence
When we think of artificial intelligence, we imagine codes, algorithms, and machines learning to solve problems. But there's one detail that's rarely mentioned: the water that makes it possible. In recent days, artificial intelligence has taken a new leap forward with the ChatGPT-4 update, which now includes the ability to generate images from descriptions or even transform photographs into different artistic styles, like the iconic Studio Ghibli. However, beyond the innovation and enthusiasm for these new features, a crucial question arises: How much water does AI consume?
In recent days, Sam Altman, CEO of OpenAI, expressed through his X account: "It's a lot of fun watching people enjoy the images on ChatGPT, but our GPUs are melting. We're going to temporarily introduce some usage limits while we work on making it more efficient." It is estimated that, in less than a week, this activity has consumed approximately 216 million liters of water, equivalent to the monthly consumption of a small city due to the "Ghibli" trend.
Why does AI need water?🔍
The servers that process AI models work tirelessly in gigantic data centers. These facilities generate so much heat that they need cooling systems to prevent failures. This is where water plays a crucial role: many infrastructures use large volumes of water to dissipate heat and keep the temperature under control.
To put this into perspective, for every kilowatt-hour (kWh) of energy used in these processes, up to 9 liters of water can be used. Research suggests that 20 to 50 queries can waste up to 500 milliliters of water, used in data center cooling systems. It may seem like an insignificant amount, but when millions of people interact with AI daily, consumption multiplies. As AI continues to expand, its water footprint becomes an environmental issue we cannot ignore.
The study, conducted by researchers at the University of California, Riverside, reveals that training AI models, such as GPT-3, can consume up to 700,000 liters of water in advanced data centers in the US. Furthermore, it projects that by 2027, global AI demand could represent between 4.2 and 6.6 billion cubic meters of water withdrawals, exceeding the total annual consumption of countries like Denmark or half of the United Kingdom. These findings underscore the need to address AI's water footprint to ensure its environmental sustainability.
Recommendations to Reduce Your Water Footprint🤖
In their research, Li and his team offer key recommendations for addressing the water impact of artificial intelligence:
More transparency on water consumption 🌊
Although carbon emissions are reported in AI models, water consumption remains a poorly documented topic. It is recommended to include this information in AI model cards and cloud dashboards to raise awareness among users and encourage more sustainable solutions.
When and where to train models matters🕒🌍
Water use efficiency varies by time and place. Training AI during cooler hours or in data centers with better water efficiency can reduce its environmental impact. Users could also choose to use AI during times of lower water demand.
To follow or not to follow the sun?☀️💧
To reduce their carbon footprint, data centers are seeking solar power ("follow the sun"). However, to minimize water consumption, it might be better to avoid the hottest hours ("not follow the sun"). Finding a balance between water and energy efficiency is key to more sustainable technological development.
AI is an incredible tool, but like any resource, its use comes at a cost. While much of the debate about responsible AI focuses on verifying accuracy and avoiding misinformation, there is another dimension we cannot ignore: its environmental impact.
Every unnecessary query, AI-generated image, or cold-blooded use contributes to global energy and water stress.
This doesn't mean avoiding AI altogether but being mindful of when and how we use it. If we want a sustainable future, we must treat technology as a resource, not a limitless toy.
What do you think? How do you balance the benefits of AI with its impact? Let's talk.
Echoes of the Earth: This newsletter is the English translation of Ecos de la Tierra. If you'd prefer to receive the newsletter in Spanish, you can sign up for the original version: https://ecosdelatierra.substack.com/subscribe