Find out how much water AI uses to answer your questions

Artificial intelligence (AI) is here to stay. Easy to use and responding to everything in seconds, the tool is already present in the routine of millions of people who use and abuse all the features of the system. However, to produce the simple image of a cute kitten riding a bicycle, the AI moves a series of mechanisms that require the use of energy and water.

Read also

Business

Thanks to advances in artificial intelligence, HP will lay off 6,000 people by 2028
Metropolitan music

“It fascinates me and scares me”: Caetano and the age of artificial intelligence
Federal District

UnB approves the creation of a new degree in artificial intelligence
Celebrities

SBT recreates Hebe and Silvio Santos with artificial intelligence; Check it out

Tools like ChatGPT, Google Gemini, Grok or Deep Seek work in huge Data Centers. These veritable “transformation factories” house multiple servers which require a constant supply of electrical energy. Most importantly, to keep these structures in perfect working order and prevent overheating, a large amount of water is required for the cooling systems. With each request or creation generated by the AI, this complex and expensive mechanism is activated.

A study by Riverside University, US, found that every 20 to 50 personal interactions with AIs uses around 500 ml of water. The Exploding Topics 2025 website highlighted that GPT Chat has approximately 800 million weekly active users and over 122 million daily visits, totaling over a billion interactions per day.

Based on this data, an article from the Federal University of Santa Maria (UFSM) did the calculation and highlighted that the use of AI represents a consumption of 10 to 25 million liters of water per day. Still taking into account figures from Riverside University, for 20 to 50 interactions, the AI consumes 500 ml of water, an average of 25 to 10 milliliters per question.

For Columbia University economist, specializing in AI, Nicole Grossmann, when we talk about generative AI, we are talking, in practice, about large data centers with an enormous concentration of high-performance hardware. She details that these data centers consume a lot of electricity and, as a result, generate a lot of heat.

“Water is precisely part of the cooling systems necessary for the safe operation of this infrastructure,” he explains.

Water appears at three levels in the AI:

In the data center itself, in the cooling systems (cooling towers, chillers, chilled water circuits);
In the production of electrical energy, since many thermoelectric power plants use large volumes of water for their operation and cooling;
In the hardware production line, such as in chip and server manufacturing.

In Brazil, the largest consumer of water remains the agro-industry. According to data from the National Agency for Water and Basic Sanitation (ANA), the total water withdrawal in Brazil is estimated at around 90,000 billion liters per year (2022 reference). The irrigated agriculture, urban supply and manufacturing sectors together account for around 85% of this volume. Projections indicate that by 2040, sectoral water demand is expected to grow by around 30% compared to 2022 levels.

With AI potentially being that extra element in that water usage queue, Felipe Giannetti, AI specialist and director of StackAI Brasil, ensures that it is not a fact that AI will always need water. It ensures the existence of data centers by using cold air, reusing water or adopting special fluids.

“The models are also becoming much more economical. Innovation is moving towards more sustainable and more efficient AI,” explains Gannetti.

Specialist Nicole Grossmann reinforces the need for more efficient cooling systems, based on closed liquid circuits, or even immersion, in which the fluid circulates without being constantly evaporated and replaced.

“The challenge is to make this cost transparent, measurable and as low as possible. In my opinion, reconciling the expansion of AI with environmental limits involves exactly this: recognizing that there is a physical price behind the cloud and working to ensure that this price does not fall on already critical water resources in several regions of the world,” he emphasizes.