In 2025, the digital revolution is no longer measured simply by computing power or the volume of data processed, but by billions of liters of water used by the infrastructures supporting artificial intelligence. At a time when AI is becoming omnipresent in our lives, its consumption of natural resources, especially water, reaches an unprecedented level. A recent study conducted by Alex de Vries-Gao, a researcher at VU University Amsterdam, reveals a striking reality: the amount of water used by artificial intelligence approaches, or even exceeds, that which humans consume annually in bottled water. This unsuspected thirst raises profound questions about AI’s ecological impact and urges a rethink of the energy and water models it relies on.
As technological giants like Microsoft, Google, or Meta intensify their investments in generative AI, this digital advancement comes with an often underestimated environmental cost. Beyond electricity, the water consumption linked to artificial intelligence, notably for cooling thousands of servers and associated power plants, becomes a critical issue. This tension around water resources is already evident in some regions where data centers compete with the essential water needs of local populations for drinking and irrigation. Thus, the promise of a technology capable of contributing to solving the climate crisis clashes with its own ecological footprint, complicating the race for a sustainable digital transition.
- 1 Water consumption linked to artificial intelligence: a striking quantified reality
- 2 Why does AI require so much water? Understanding direct and indirect water consumption
- 3 Energy consumption and environmental impact: water, an essential yet overlooked link
- 4 The ecological impact of AI water consumption on global natural resources
- 5 The real cost of an AI query: beyond electricity, significant water consumption
- 6 Transparency and responsibility: a political issue to regulate AI water consumption
- 7 The paradoxical link between AI, water consumption, and the fight against climate change
- 7.1 Why does AI consume so much water?
- 7.2 How can the increase in AI water consumption impact local populations?
- 7.3 What solutions exist to reduce this water consumption linked to AI?
- 7.4 Does each AI query have a concrete impact on water consumption?
- 7.5 Can AI help solve the climate crisis despite its impact?
Water consumption linked to artificial intelligence: a striking quantified reality
The figures revealed in 2025 challenge traditional perceptions of the ecological impact of artificial intelligence. According to data compiled by Alex de Vries-Gao, AI’s water demand could reach between 312 and 765 billion liters in a single year. To put this volume into perspective, it equals or exceeds the total annual global consumption of bottled water by the entire human population. This figure highlights a little-publicized issue: the gigantic water consumption necessary to maintain these vast digital infrastructures that power all AI applications, from simple chatbots to the most advanced language models.
The table below illustrates the comparative volumes of water consumed by different uses, emphasizing the growing weight of AI in the global water resource landscape:
| Usage | Annual consumption (in billion liters) |
|---|---|
| Global bottled water consumption | 500 |
| Water consumption linked to AI (low estimate) | 312 |
| Water consumption linked to AI (high estimate) | 765 |
| Cooling of non-AI related cloud data centers | 350 |
This comparison shows that the direct and indirect water consumption generated by artificial intelligence rivals traditional human consumption uses. This not only disrupts expectations regarding the environmental impact of digital technology but also sharpens the debate about the sustainability of these technologies in the near future.
Why does AI require so much water? Understanding direct and indirect water consumption
AI’s significant water needs are primarily explained by two fundamental factors related to technological infrastructures: server cooling and the electricity production that powers these facilities.
The cooling system: a drop of water becomes essential
Data centers hosting artificial intelligences use a phenomenal amount of complex electronic circuits. These centers operate thanks to high-performance chips, notably GPUs, which generate intense heat during learning or content generation by artificial intelligences like GPT-4 or Gemini. Without effective cooling, these components risk overheating and failing.
To control this temperature, data centers employ water-based cooling systems, often by immersion or circulation in cooling towers. This water circulates through circuits to absorb heat and is then discharged as vapor, resulting in massive evaporation. This evaporation—sometimes up to 80% of the water used—represents a real loss for local water resources, contributing to the depletion of water reserves in some regions.
The less visible but equally crucial side of AI’s water consumption comes from the electricity used. The majority of global electric energy still depends on thermal, nuclear, or fossil fuel power plants, whose operation requires large amounts of water to cool reactors or turbines.
According to available analyses, the indirect water consumption of the plants supplying data centers often represents a larger share than the water used directly to cool servers. It is therefore a double “water and energy” effect weighing on the planet’s water resources.
Energy consumption and environmental impact: water, an essential yet overlooked link
If AI’s energy consumption is often mentioned, with absorbed power potentially reaching 23 gigawatts by the end of 2025, the impact on water resources remains too unknown to the general public and sometimes even to decision-makers.
Indeed, intensive electric energy consumption, combined with water cooling, makes water management a major challenge. This is not limited to a localized problem in server rooms but extends to watersheds and aquifers that can be put under pressure.
This process increases the overall ecological footprint of digital infrastructures and raises the following question: how to reconcile technological development and preservation of natural resources? Here are some major observed impacts explained in detail:
- Regional water stress: Data center installations in already water-stressed areas, such as some US states, Spain, or South America, where competition between industry and agriculture intensifies;
- Thermal and chemical pollution: Water used in cooling towers is often discharged at high temperatures, affecting local aquatic biodiversity;
- Indirect CO₂ emissions: Through electricity production from fossil sources, water consumption is also linked to high environmental emissions.
The ecological impact of AI water consumption on global natural resources
The scale of water consumption by artificial intelligence invites a thorough examination of its effects on global natural resources. In a context of climate emergency and increasing drought phenomena, this growing demand for drinking and industrial water has major repercussions.
To better understand the issue, it is necessary first to distinguish the different types of water involved. Freshwater, essential for human consumption, agriculture, and industry, is the most threatened resource. Thus, a large part of the water used to cool data centers is automatically drawn from underground or surface freshwater resources, creating additional pressure on aquifers and rivers.
In some arid zones, the increased uses related to AI exacerbate local water stress already aggravated by climate change. In this regard, usage conflicts arise between data centers and populations over water access, raising questions about responsible management of natural resources at a global scale.
Examples of regions where AI water consumption intensifies tensions
- Iowa, USA: State with seasonal water stress, where cooling pools have been contentious with agricultural operators;
- Spain: Semi-arid zones where some technological operations use considerable volumes of water, sparking debates between authorities and companies;
- Chile and Uruguay: Countries already affected by water shortages seeing the establishment of data centers increase local demand for drinking water.
The balance between digital development and sustainable availability of water resources now stands as a crucial issue and an unavoidable political challenge for the coming years.
The real cost of an AI query: beyond electricity, significant water consumption
Beyond global infrastructures, it is essential to apprehend water consumption at the scale of individual and daily uses of artificial intelligence. For the hundreds of millions of users interacting daily with conversational agents, virtual assistants, or smart search engines, each query generates an energy and water expenditure.
Researchers at the University of California, Riverside, have thus quantified the water impact of a simple interaction with an AI chatbot. An average session containing between 10 and 50 questions can consume nearly 50 centiliters of water, which might seem small for an individual but translates into significant waste when multiplied globally.
To make this more tangible, it is useful to convert these data into a daily comparison: if one billion users each ask about ten questions, this means the use of hundreds of millions of liters of pure water per day just to keep servers cool. Google has acknowledged that its Gemini AI alone consumes the equivalent of five drops of water per simple query.
This consumption raises a major question about the sustainability of the ever-growing popularization of AI models. How to maintain technological performance while limiting pressure on water resources?
Some avenues to reduce water consumption per AI query:
- Develop more optimized and less resource-hungry models;
- Use closed-loop cooling techniques to limit evaporation;
- Favor renewable energy sources less dependent on water;
- Locate data centers in regions with low water stress.
Transparency and responsibility: a political issue to regulate AI water consumption
Faced with the environmental challenges posed by water consumption linked to artificial intelligence, the question of transparency from tech giants arises sharply. Today, information on AI’s specific ecological impact remains fragmented and mixed with other cloud activities, preventing a clear assessment of the issues.
Local communities, environmental authorities, and citizen groups demand broader access to data on the water and energy consumption of digital infrastructures. This would make it possible to precisely evaluate social and environmental costs and guide public policies toward a fairer management of resources.
Furthermore, the establishment of a sustainable AI coalition, as initiated during the international summit in Paris in 2025, aims to bring together states, companies, and civil society to promote ambitious environmental standards. This coalition encourages the adoption of sober, responsible technologies respectful of planetary boundaries.
Without strong and concerted commitment, the advancement of artificial intelligence could worsen issues related to natural resources, particularly water, seriously compromising its expected role in the global ecological transition.
The paradoxical link between AI, water consumption, and the fight against climate change
One of the major paradoxes of the digital revolution lies in the fact that artificial intelligence is both one of the proposed solutions to mitigate climate change and a significant source of pressure on natural resources. Indeed, AI facilitates climate phenomena modeling, energy optimization, intelligent infrastructure management, and fosters innovation in renewable energies.
However, the water and energy consumption it generates slows these advances. The AI production and usage cycle, involving also internet networks, data centers, and associated industries, presents a considerable ecological impact. Thus, without sustainable guidance in its design and operation, AI could become a “vicious circle” where its own means of action are limited by its environmental footprint.
The sector is aware of this contradiction and works on integrating less resource-intensive approaches. Advanced research on waterless cooling, massive use of renewable energies, and development of more efficient AI systems are underway. The challenge is to achieve a balance where artificial intelligence remains a lever for the environment, without jeopardizing the vital resources on which all life depends.
Why does AI consume so much water?
AI requires enormous server infrastructures that produce a lot of heat. To cool them, large quantities of water are used, often via evaporation systems. Moreover, the electricity production powering these centers itself uses a lot of water for cooling the plants.
How can the increase in AI water consumption impact local populations?
In certain regions already under water stress, the presence of data centers can worsen water scarcity, causing tensions with agricultural and domestic uses by the populations.
What solutions exist to reduce this water consumption linked to AI?
Closed-loop cooling techniques, the use of renewable energies, the location of data centers in low water stress areas, and improved AI model efficiency are among the explored avenues to reduce ecological impact.
Does each AI query have a concrete impact on water consumption?
Yes. Each interaction with AI consumes a measurable amount of energy and water, notably for cooling servers. Although small at the individual level, the overall impact becomes huge with hundreds of millions of users.
Can AI help solve the climate crisis despite its impact?
AI is a powerful tool to model and combat climate change, but this requires its development to be guided by sustainability principles to limit its ecological footprint, notably in terms of water and energy consumption.
