AI’s Growing Thirst: Data Centers Could Consume Water for 1.3 Billion People

A stark warning has emerged from a new report by the United Nations University (UNU), detailing the potentially catastrophic environmental impact of artificial intelligence, particularly its voracious appetite for water. By 2030, the data centers powering AI technologies could consume a staggering amount of water, equivalent to the basic needs of 1.3 billion people living in sub-Saharan Africa.

The study, conducted by the UNU Institute for Water, Environment and Health (UNU-INWEH), also projects that these data centers will demand 945 terawatt-hours of electricity annually by 2030. This figure nearly triples the combined annual electricity consumption of Pakistan, Bangladesh, and Nigeria, nations that are home to over 650 million people.

Beyond carbon emissions, the report emphasizes that the cost of AI extends to its “water footprint” and “territorial footprint.” Every kilowatt-hour used to train or run AI systems requires water for cooling and land for infrastructure and supply chains. Crucially, these environmental impacts do not move in lockstep. For instance, switching electricity sources from coal to bioenergy might reduce carbon footprints by 70%, but it can more than triple water consumption and multiply land use by a hundredfold.

The environmental toll in 2030 is projected to be immense. The water required for AI data centers could match the domestic water needs of 1.3 billion people in sub-Saharan Africa. Furthermore, the land occupied by these facilities could exceed 14,500 square kilometers, an area nearly double that of the Jakarta metropolitan region.

In 2025, global data centers consumed approximately 448 terawatt-hours of electricity. If considered a country, this would have ranked them as the eleventh largest electricity consumer worldwide, ahead of Saudi Arabia and behind France.

Focus on Inference, Not Just Training

Professor Kaveh Madani, Director of UNU-INWEH and a recipient of the Stockholm Water Prize, clarified that the report is not an indictment of artificial intelligence itself but a call for responsible development. The public debate has largely focused on the energy needed for training large AI models. However, the UNU report argues this perspective is outdated, as the actual daily operation, or “inference,” of AI systems accounts for 80% to 90% of their total energy consumption once deployed.

For example, ChatGPT alone processes around 2.5 billion queries daily, consuming approximately 383 gigawatt-hours of electricity annually. The energy cost per query varies dramatically: a text conversation uses about 200 times more energy than a basic classification task, while generating a single image can be 1,450 times more energy-intensive than that baseline. Creating a short AI-generated video can consume as much electricity as 200,000 spam email classifications.

Rebound Effect and Local Tensions

The report also highlights the “rebound effect,” where increased efficiency in AI models leads to lower costs, encouraging even greater usage and thus amplifying overall consumption. Concrete examples of local strain are already evident. In Ireland, data centers accounted for 21% of the nation’s electricity consumption in 2023, surpassing all urban households. This has led to a moratorium on new data center permits near Dublin until 2028. In Querétaro, Mexico, data center expansion is straining water reserves during a severe drought. Similarly, a project in Uruguay coincided with the 2023 drought that left Montevideo without safe drinking water.

Unequal Distribution of Burdens and Benefits

The report underscores a significant disparity in how the burdens and benefits of AI are distributed. Only 32 countries host specialized AI data centers, with the United States and China dominating 90% of this capacity. Meanwhile, over 150 countries lack sovereign computing power but face the environmental consequences of critical mineral extraction and electronic waste management. By 2030, electronic waste from these operations could reach 2.5 million metric tons annually, equivalent to discarding 250 Eiffel Towers each year.

Recommendations for a Sustainable AI Ecosystem

To foster a responsible AI ecosystem, the UNU report proposes six core principles: transparency, efficiency by design, equity, lifecycle responsibility, global cooperation, and sustainable use. It also offers specific recommendations for governments, businesses, and users. The researchers conclude that an AI that operates within planetary boundaries is achievable, emphasizing that technological advancement and environmental responsibility can progress in tandem through measurement, transparency, and shared commitment.

Datos clave
| Aspecto | Proyección 2030 | Fuente |
|—|—|—|
| Consumo eléctrico de centros de datos | 945 TWh | UNU |
| Equivalencia hídrica | Necesidades básicas de 1.300 millones de personas | UNU |
| Ocupación de suelo | Más de 14.500 km² | UNU |
| Generación anual de residuos electrónicos | 2.5 millones de toneladas | UNU |

This development is significant for FootballGames10 readers as it highlights a critical sustainability challenge directly linked to the technology underlying many modern digital platforms, including sports analysis and data dissemination. Understanding the environmental cost of AI is crucial for appreciating the long-term viability and ethical considerations of the digital infrastructure that supports global sports content.

Fuente: okdiario.com – https://okdiario.com/okgreen/huella-ambiental-inteligencia-artificial-onu-17998514

Datos clave