Dish of the Day

Ondokuz Mayıs University (OMU) Faculty of Engineering, Department of Environmental Engineering faculty member Yüksel Ardalı, shared an assessment of the impacts of increasing water consumption driven by artificial intelligence technologies on global water resources.

“Artificial intelligence consumes far more natural resources than commonly assumed”

Prof. Dr. Yüksel Ardalı stated that artificial intelligence has become widely used over the past decade across many sectors, from search engines to healthcare and defense, but that behind this digital revolution lies a serious level of natural resource consumption that is often overlooked.

“Artificial intelligence (AI) has become one of the most transformative technologies humanity has encountered in the past decade. From search engines and medical imaging systems to the defense industry and universities, billions of people interact with AI-based systems every day. However, behind this digital revolution lies a reality that is frequently ignored: artificial intelligence consumes far greater amounts of natural resources than is commonly assumed, with water being foremost among them. At a time when the world is already facing a deep water crisis due to climate change, drought, and population growth, the increasing water demand of data centers is creating a new area of environmental pressure. The water footprint of digitalisation has not yet received as much attention as energy or carbon footprints, but this invisibility has now reached an unsustainable point,” she said.

Prof. Dr. Ardalı also noted that while approximately 2.2 billion people worldwide lack access to safely managed drinking water, nearly 40 percent of the global population experiences water stress for at least part of the year, and according to data from the Food and Agriculture Organization of the United Nations (FAO), the amount of potable freshwater available per capita has declined by approximately 7 percent over the past decade.

“Artificial intelligence does not ‘drink’ water directly, but the infrastructure it relies on is highly water-intensive”

Prof. Dr. Ardalı explained:

“Artificial intelligence does not ‘drink’ water directly, but the infrastructure on which it operates is extremely water-intensive. Large language models, image-processing systems, and other AI applications run in data centers composed of thousands of servers. Water is used to cool these servers, and electricity generation indirectly consumes water as well. Although this consumption does not yet match agriculture or industry in terms of total global share, it constitutes a critical risk because it is growing very rapidly, occurs at high intensity, and is often concentrated in regions already experiencing water stress. Moreover, a significant portion of the water used evaporates and does not return to the same watershed, meaning it is effectively lost.”

Highlighting the substantial volumes of clean water consumed during AI model training, Prof. Dr. Ardalı stated:

“According to studies conducted by the University of Massachusetts Amherst, training a single large-scale AI model can consume between 200,000 and 700,000 liters of clean water. Researchers at the University of California, Riverside have modelled the water footprint of user interactions with large language models similar to ChatGPT and estimate that approximately every 40–50 queries correspond to nearly one liter of water consumption, including data-center and power-plant cooling. These figures are model-based rather than direct measurements, but they clearly demonstrate that the water footprint of artificial intelligence is far greater than generally perceived.”

“Water consumption linked to AI and data centers is expected to increase explosively in the coming years”

Emphasizing future projections, Prof. Dr. Ardalı said:

“The International Energy Agency (IEA) and sector analyses predict that water consumption linked to artificial intelligence and data centers will increase explosively in the coming years. Various global assessments indicate that AI-related data centers could consume 300 to 800 billion liters of freshwater annually between 2025 and 2027—an amount equivalent to the urban water use of many mid-sized countries. According to a 2024 analysis by Morgan Stanley, the outlook is even more striking: water consumption by AI-focused data centers could increase nearly 11-fold by 2028, reaching approximately 1 trillion liters. This trend is also reflected in the reports of major technology companies. Google disclosed that it consumed approximately 22.7 billion liters of water in its data centers in 2024, while Microsoft and Meta have reported water-use increases exceeding 30 percent in their global operations over recent years. Both companies acknowledge that the primary driver of this increase is the training and operation of AI models.”

“Türkiye aims to rapidly expand digitalisation and data-center investments”

Prof. Dr. Ardalı underlined that one of the most critical aspects of AI-related water consumption is where this consumption occurs:

“Large data centers are often established in regions such as the southwestern United States, Spain, Ireland, Chile, and the Middle East; areas that offer investment incentives but are also experiencing water stress. This situation places technology companies in direct competition with agriculture and cities for the same limited water resources, intensifying social tensions. In Türkiye’s case, with approximately 1,300 cubic meters of available water per capita, the country is classified among ‘water-stressed nations’. Climate change, inefficient irrigation practices, excessive groundwater extraction, and pollution further deepen this stress. At the same time, Türkiye aims to rapidly expand digitalisation and data-center investments. If these investments are not planned on the basis of water efficiency, digital transformation could become a serious risk to water security.”

Stressing that this trajectory is not inevitable, Prof. Dr. Ardalı outlined several solution pathways:

• Advanced cooling technologies and alternative water sources: The most direct way to reduce water consumption is to transform cooling systems in data centers. Technologies such as on-chip liquid cooling and closed-loop systems significantly minimise evaporation and water loss.
• Efficient and smaller models: Excessive resource use in AI training and operation can be directly reduced through algorithmic efficiency. Developing targeted, optimised models with fewer parameters can perform the same tasks using far less computational power—and therefore much less cooling demand.
• Strategic site selection based on water and renewable energy availability: The geographical placement of AI infrastructure is of strategic importance. Data centers should be incentivised to locate in regions with relatively abundant water resources and access to renewable energy, rather than in high water-stress areas.
• Use of alternative water sources: Greywater use can significantly reduce direct pressure on clean water resources. For coastal facilities, desalinated seawater treated via reverse osmosis represents another sustainable option.
• Water-positive corporate commitments and transparent accountability: Major technology companies have announced water-positive commitments aimed at replenishing more water than they consume within watersheds.
• AI-enabled water management: AI systems used in climate modelling, agricultural irrigation optimisation, leak detection in water networks, and drought forecasting can enable far more efficient management and protection of global water resources.
• Policy and regulatory frameworks: Technological solutions and corporate commitments can only be effective within a strong regulatory environment. Local and national regulations should mandate minimum water-efficiency standards and alternative water-source requirements for new data centers. To increase consumer awareness and guide markets, “water footprint” labels for devices and services or performance certifications for data centers should be developed. Large technology companies could be legally required to report basin-level water withdrawal and consumption data transparently and subject to independent audit.
• While the recovery of billions of liters of water by Google and Meta represents a positive step, it is critically important that these “water-positive” commitments are transparent, independently verified, and meaningful at the local watershed level. Otherwise, the risks of greenwashing, lack of transparency, and incomplete accounting highlighted by experts may undermine their real impact.