How does AI consume so much water? Here’s why every question heats up massive servers and how data centres keep them cool

Artificial intelligence is changing the way people search for information, write emails, create images and solve everyday problems. Millions of users ask AI tools questions every hour, expecting answers within seconds. However, behind every response is a massive network of data centres filled with powerful computer servers that consume large amounts of electricity and generate significant heat. Keeping these servers cool requires enormous amounts of water, raising concerns about the environmental impact of AI’s rapid growth.

The issue has gained global attention after United Nations Secretary-General António Guterres called for greater transparency from AI companies regarding their use of electricity, water and land. Experts warn that as AI adoption continues to increase, the demand for these resources could rise sharply over the next few years.

What happens when you ask an AI a question?

Whenever a user submits a prompt to AI platforms such as ChatGPT, Gemini or other AI assistants, the request is processed in large-scale data centres located around the world. These facilities contain thousands of high-performance servers that work simultaneously to analyse information and generate responses.

As these servers perform billions of calculations every second, they produce large amounts of heat. Without an effective cooling system, the machines could overheat, reducing performance and increasing the risk of hardware failure. This is why cooling has become one of the most important parts of operating AI infrastructure.

Researchers from the University of California, Riverside have estimated that answering a batch of around 10 to 50 common AI questions may require approximately 500 millilitres of water for cooling. That means a single prompt could indirectly consume around 10 to 50 millilitres of water, depending on the workload and the cooling technology used by the data centre.

Why do AI data centres use so much water?

Water is not poured directly onto computer servers. Instead, it plays a vital role in advanced cooling systems used inside data centres.

One major method involves cooling towers. These towers absorb heat generated by the servers, allowing water to evaporate into the atmosphere while carrying away excess heat. Because much of this water evaporates, it cannot immediately be reused.

Water is also consumed indirectly through electricity generation. AI data centres require enormous amounts of power, and many power plants also rely on water for cooling their equipment. As a result, every AI query has both direct and indirect water consumption associated with it.

Although the amount of water used for a single question appears small, the cumulative impact is substantial. With hundreds of millions of AI requests being processed every day across the globe, the total water consumption reaches millions of litres daily. Training advanced AI models requires even greater resources because thousands of servers operate continuously for weeks or even months.

What are technology companies doing to reduce water use?

Major technology companies have started adopting several measures to reduce the environmental impact of AI infrastructure.

Many companies are increasingly using recycled or non-potable industrial water instead of drinking water for cooling operations. Others are building new data centres in naturally colder regions such as Finland and Ireland, where outside air can help cool servers and reduce dependence on water-intensive cooling systems.

Several firms have also announced long-term sustainability targets. Some aim to replenish more water than they consume by investing in water conservation projects, while others are improving server efficiency to lower heat generation and reduce cooling requirements.

Are governments introducing water regulations?

Regulations governing water use in AI data centres remain limited in many countries. While some proposals in the United States seek mandatory reporting of electricity and water consumption, comprehensive national rules have yet to be implemented.

Certain US states have introduced stricter requirements. For example, large data centres consuming more than specified annual water limits may need permits and environmental assessments before expanding operations.

In India, there is currently no nationwide policy specifically regulating water consumption by data centres. States including Telangana, Tamil Nadu and Karnataka have introduced data centre policies that encourage efficient water management, but strict nationwide regulations are still absent.

This issue is particularly important because many Indian cities, including Bengaluru, Hyderabad and Chennai, already face seasonal water shortages. As India continues to become a major global hub for digital infrastructure and AI services, experts believe sustainable water management will be essential.

With AI usage expected to expand rapidly in the coming years, balancing technological innovation with responsible resource management will become increasingly important. Governments, technology companies and researchers are likely to face growing pressure to develop cooling solutions that reduce water consumption while supporting the future growth of artificial intelligence.