How Much Water Does AI Really Use?
When people think about artificial intelligence (AI), they often imagine futuristic robots or powerful chatbots like ChatGPT. But there's a hidden cost that many don't see: AI consumes a surprising amount of water — and it’s becoming a real environmental concern.
Why does AI need water?
AI models run on powerful computers inside massive buildings called data centers. These machines generate a lot of heat while processing data, and to prevent overheating, they need to be cooled constantly. The most common method is water-based cooling, where water is used to absorb and remove the heat.
In fact, around 70% of the water used in AI operations is for this kind of cooling. This is not just happening during model training — even everyday AI usage (like you chatting with an AI assistant) contributes to this water consumption.
How much water are we talking about?
Training a single large AI model like GPT-3 can use up to 700,000 liters (185,000 gallons) of clean, potable water — that’s enough to fill an Olympic-sized swimming pool.
A single user’s interaction with a large language model might indirectly use half a liter (about a full water bottle) of water, depending on the server and time of day.
While that might not sound like much for one person, remember that millions of people use AI every day — and that number is only growing.
The global impact is growing fast
If the current trends continue, by 2027, the global water consumption of AI systems could reach 4.2 to 6.6 billion cubic meters per year. That’s more water than Denmark uses in an entire year, or roughly half of the UK's total water withdrawals.
What’s more concerning is that many data centers are being built in areas that already suffer from water stress or drought, like parts of the U.S., India, and the Middle East. This creates a localised burden, even if the impact feels invisible to users.
The problem: lack of transparency
While many tech companies report their carbon emissions, few are open about their water use. This lack of transparency makes it difficult to hold companies accountable or to compare how different providers impact the environment.
Experts are now calling for more public reporting on water footprints, just like we do with energy and emissions.
What can be done?
Fortunately, there are some promising solutions:
Use reclaimed or non-drinkable water instead of fresh drinking water for cooling.
Move data centers to cooler regions, like the Nordics, where natural temperatures help reduce the need for water-intensive cooling.
Improve cooling technology, such as using recycled water systems or immersion cooling.
Schedule energy-intensive AI tasks during times or in locations where water is more abundant.
Encourage AI developers to be more efficient, both in training and running models.
Final thoughts
AI is transforming the world — but it comes with invisible environmental costs, especially when it comes to water. As AI becomes more powerful and more widely used, we need to think carefully about how and where it’s developed.
Source: OECD
0 Comments