Artificial intelligence and water: consumption, impact and how AI can save this resource

AI-generated summary

Recent studies reveal AI’s sizeable but often overlooked water footprint: training large models like GPT-3 can consume roughly 700,000 liters, and each user interaction with chat systems may indirectly use 0.1–0.5 liters. The water use isn’t in the algorithms but in data centers, where evaporative cooling and other systems manage heat from thousands of servers. As AI demand grows, so does the need for more computational capacity and associated water use, prompting concerns where water is already scarce.

Big tech is responding by redesigning data centers—air cooling, closed circuits, recycled or non-potable water use, and hybrid systems—to lower water intensity. At the same time, AI offers powerful solutions for water management: smart urban networks that detect leaks and predict demand, real-time distribution optimization, and precision irrigation in agriculture, where sensors and predictive models can cut water use by 20–50%. Advanced applications extend to treatment, desalination and recycling, improving efficiency and monitoring. The result is a paradox: AI both increases resource demand and provides tools to manage water sustainably. The challenge is balancing infrastructure design, site choices, reuse strategies and AI-driven optimization so innovation helps avert, rather than worsen, global water stress.

Did you know that every use of artificial intelligence consumes water? From ChatGPT to data centers, the water impact of AI is real. But so is its potential to optimize water use in cities, agriculture, and industry. We analyze the problem and, above all, the solutions.

Artificial intelligence consumes water. And much more than it seems.

Training advanced models such as GPT-3 required approximately 700,000 liters of water, according to recent research. And each interaction with systems such as ChatGPT can involve the indirect consumption of between 0.1 and 0.5 liters of water.

The data is surprising. It also generates an immediate reaction: if artificial intelligence needs so much water, are we aggravating one of the greatest challenges on the planet?

The answer, as is often the case in major technological transitions, is more complex. AI is part of the problem. But it can also be one of the most powerful tools to solve it.

At the Bankinter Innovation Foundation we have been analysing this duality in emerging technologies for some time: its direct impact and its transformative capacity. In the case of water, an essential and increasingly stressed resource – more than 2,000 million people live in water-stressed regions – artificial intelligence is positioned as a critical lever to improve efficiency, anticipation and resilience.

How much water does artificial intelligence consume? The surprising fact

The conversation about AI and sustainability has grown in parallel with the expansion of models such as ChatGPT. Energy has been the main focus. Water, however, is beginning to take center stage in the debate.

Why Data Centers Need So Much Water

The water consumption of artificial intelligence is not direct. There is no “water inside the algorithm”. The use is concentrated in the physical infrastructure: data centers.

Companies such as Microsoft, Google or Meta operate thousands of servers that generate constant heat. To avoid failures, that heat must be dissipated. One of the most efficient ways is evaporative cooling, which uses water.

In 2022, Microsoft reported a consumption of more than 6 million cubic meters of water in its data centers globally. Google has also published water efficiency metrics, known as Water Usage Effectiveness (WUE), which allow you to measure how much water is used per unit of computing.

The key takeaway here is that the growth of AI means more computational capacity. And that means more data centers.

How much water does ChatGPT use per query

A query to ChatGPT may seem trivial. A prompt, a response, a few seconds of processing.

However, behind it there are servers working remotely. According to academic estimates based on the Making AI Less ‘Thirsty’ study, several dozen interactions with models such as ChatGPT can involve the consumption of approximately half a liter of water, which allows estimating an impact per query in the order of tenths of a liter, depending on the location of the data center, the cooling efficiency and the time of use.

It is not a visible consumption for the user. But it is cumulative. Millions of daily queries generate a relevant impact.

This type of data has turned the water consumption of AI into a media topic. But it has also opened up a more interesting conversation: how to redesign this infrastructure to make it more efficient.

What Big Tech is doing to reduce its water consumption

Big tech companies have begun to act. Not only because of regulatory or reputational pressure. Also because of economic efficiency.

Big tech companies have begun to redesign their infrastructures to reduce their water footprint. Microsoft has announced the development of data centers capable of operating without water consumption in cooling under certain conditions, through air cooling systems and closed circuits. Google is working on optimizing its Water Usage Effectiveness (WUE) metric and using recycled or non-potable water to reduce pressure on water resources. Meta, for its part, is exploring hybrid systems that combine air and water depending on the environment, adapting consumption to each location.

This change reflects a relevant evolution: the focus is no longer only on the energy efficiency of data centers, but on their water impact, which introduces new metrics, new location decisions and a growing integration between digital infrastructure and sustainability.

The challenge is to scale these solutions at the pace of growth in AI demand.

AI as a solution: smart water management

The other side of the story is more promising. Artificial intelligence can help manage water more efficiently than ever before.

In urban networks, a significant part of the water is lost before it reaches the user. In some cities, leaks can account for up to 30% of the supply.

These types of solutions are already being developed. The Bankinter Innovation Foundation has analysed how the combination of sensors, algorithms and data makes it possible to create “smart water” systems capable of anticipating leaks, optimising networks and improving water resilience in urban environments.

Prediction of consumption and detection of leaks in urban networks

AI systems make it possible to analyse real-time data from sensors distributed in the network. From this data, they can detect anomalous patterns that indicate leaks, breakages or irregular consumption.

Companies in the water sector and utilities are incorporating these systems to anticipate problems instead of reacting when they have already occurred.

This shift in focus – from reactive to predictive maintenance – is one of the major impacts of AI on critical infrastructure.

Real-time distribution optimisation

Artificial intelligence also makes it possible to adjust water supply according to real demand.

Instead of operating with static models, systems can integrate variables such as weather, historical consumption, local events or the state of reservoirs to redistribute water dynamically.

This improves efficiency and reduces stress on resources.

The Bankinter Innovation Foundation’s Future Trends Forum insisted on the need to integrate technology, governance and data to manage water sustainably. AI acts here as a layer of intelligence that connects all these elements.

AI and agriculture: precision irrigation to save water

Agriculture accounts for approximately 70% of global freshwater consumption. Any improvement in this sector has a systemic impact.

This is where artificial intelligence shows one of its clearest use cases.

Smart sensors and predictive irrigation models

Precision irrigation combines in-ground sensors, weather data, and predictive models to adjust water use to the real needs of the crop.

Companies such as CropX or Hortau develop systems that allow farmers to decide when and how much to irrigate with great precision. These types of solutions are part of a broader transformation towards sustainable agriculture based on data, sensors and predictive models, a line that the Bankinter Innovation Foundation has analysed as key to the future of the agricultural sector.

The result is twofold: reduced water consumption and improved productivity.

In some cases, savings can be between 20% and 50%.

Use cases in Spain: Almeria and the CSIC

Spain is a natural laboratory for this type of solution.

In regions such as Murcia, where water is scarce, smart irrigation systems are already part of everyday agricultural life. The use of sensors, climate models and automation allows every drop to be optimised.

The CSIC has been working for years on precision agriculture and agricultural digitalization, a line that combines sensors, advanced analytics and predictive models to adjust irrigation and improve water efficiency in environments increasingly affected by climate stress.

These cases show something important: innovation is not only in Silicon Valley. It is also being developed in local ecosystems where the need is urgent.

The future: desalination, recycling and water treatment with AI

Artificial intelligence is beginning to be applied in more advanced stages of the water cycle.

In treatment plants, algorithms can optimize chemical and energy processes. In desalination, they can improve the efficiency of membranes and reduce costs. In recycling, they allow water quality to be monitored in real time.

These applications are key in regions with structural water scarcity.

This approach is part of a broader transformation of the sector. The Bankinter Innovation Foundation has analysed how technological innovation – from digitalisation to desalination or reuse – will be key to addressing the water crisis from a systemic perspective.

The report “Water: our vital resource in check” by the Bankinter Innovation Foundation stresses that these technologies will be essential to guarantee access to water in the coming decades. And AI can accelerate all of them.

The paradox of AI and water: problem or solution?

Artificial intelligence introduces an interesting paradox.

On the one hand, it increases the consumption of resources. Data centers require energy and water. The growth of AI can amplify this demand. On the

other, it offers unprecedented tools to manage those same resources more efficiently.

The key is in balance. In designing more efficient AI systems. In locating data centers in areas with less water stress. In using recycled water. And in applying artificial intelligence itself to optimize its impact.

This type of paradox is not new. It is part of all major technological revolutions. What is relevant is how it is managed.

In the case of water, artificial intelligence can make the difference between a scenario of increasing scarcity and one of intelligent and sustainable management.