The Hidden Cost of a Query
Every time you use an AI chatbot or a generative AI service, you are activating rows of powerful servers housed in a massive data centre. These servers generate an immense amount of heat, and keeping them from overheating is a critical operational challenge.
While the process feels virtual, its physical demands are very real. Estimates suggest a simple chat session with an AI involving around 20 questions can consume up to a bottle of fresh water for cooling. When scaled to millions of users making billions of queries, this adds up to a significant new demand on global water supplies.
The Cooling Conundrum
The primary reason data centres need so much water is for cooling. The most common method, especially in large-scale facilities, is evaporative cooling. This process works much like the human body sweating to cool down. Water is pumped into cooling towers, where hot air from the server rooms is passed through it. As the water evaporates, it absorbs heat, and the resulting water vapour is released into the atmosphere. While energy-efficient, this method means the water is consumed—it is not returned to the local water system. A large data centre can consume up to 5 million gallons of water every single day, which is comparable to the water usage of a city of 50,000 people.
Big Tech's Growing Water Footprint
As the demand for AI grows, so does the water consumption of the tech giants that power it. Many companies have become more transparent about their usage, and the numbers are staggering. Google reported that its data centres used 19.5 billion litres of water in 2022, a 20% increase from the previous year. Similarly, Microsoft's water usage jumped 34% in the same period to 6.4 billion litres. This water is often drawn from the same sources that local communities rely on for drinking water and agriculture, creating competition for a finite resource, especially in water-stressed regions.
More Than Just Cooling
The water footprint of AI extends beyond the direct cooling of servers. A significant amount of water is also used 'indirectly' to generate the massive amounts of electricity these facilities require. Thermoelectric power plants, including those that burn fossil fuels, use water for their own cooling processes. One analysis estimated this indirect water consumption for US data centres was roughly 211 billion gallons in 2023, far exceeding the 17 billion gallons used for direct on-site cooling. Furthermore, the manufacturing of the sophisticated silicon chips that power AI is also a water-intensive process, requiring millions of gallons of ultrapure water each day.
The Search for Sustainable Solutions
The industry is acutely aware of this growing challenge and is exploring innovative solutions to curb water consumption. One of the most promising is the shift to closed-loop cooling systems. These systems circulate a contained liquid to absorb heat, which is then cooled by outside air without evaporation, drastically reducing the need for a constant water supply. Other emerging technologies include direct-to-chip cooling, which uses microfluidic channels to cool processors from the inside, and immersion cooling, where entire servers are submerged in a non-conductive fluid. Companies are also strategically locating new data centres in cooler climates where air-based cooling is more viable. Both Microsoft and Google have pledged to become 'water positive,' meaning they aim to replenish more water than they consume through investments in water restoration projects.
