La Jornada: The voracious artificial intelligence

The development and use of Artificial intelligence (AI) systems use more energy than almost all countries in the world, which translates into a devastating impact on the use of water, land, and the emission of carbon dioxide and other gases that cause climate chaos. In 2025, the electricity consumption of the artificial intelligence industry was 448 terawatts, which exceeds the electricity consumption of all of Mexico by about 1.3 times, triples that of Argentina and is 40 times the electricity demand of Uruguay or Costa Rica.

Even more worrying: a recent United Nations report estimated that by 2030 the energy demand of data centers – where the AI ​​infrastructure is housed – will double to 945 terawatt hours of electricity, equivalent to the current total consumption of the entire African continent with a population of 1.6 billion people.

If the estimated AI electricity demand for 2030 is considered as a country, it will occupy sixth place in the world ranking, only behind the United States, China, India, Russia and Japan, far surpassing high-consuming countries such as Germany, France or Saudi Arabia.

The report “Environmental cost of energy use of artificial intelligence: carbon, water and soil footprints” was published in June 2026 by the United Nations University and coordinated by Kaveh Madani, director of the Institute of Water, Environment and Health at the United Nations University. (https://tinyurl.com/UNU-IA-impacto).

Together with the very high energy demand, they report that, by 2030, the associated water footprint will be 9.3 billion liters of water, which is equivalent to the annual basic needs of 1.3 billion people in sub-Saharan Africa. The appropriation of land for infrastructure is estimated at 14,500 square kilometers, 10 times the surface of Mexico City.

They point out that the environmental footprint of artificial intelligence has been systematically underestimated. Not only because the powerful technology companies that dominate this industry avoid providing that information, but also because most evaluations focus on only measuring the carbon emissions associated with the training and execution of large language models. They also do not consider that each kilowatt of electricity used carries a significant water footprint, among others in generation, energy use and the cooling required by data centers, as well as its footprint in land use, energy infrastructure, facilities and supply chains.

Measuring impacts only in carbon emissions and trying to offset them with supposedly “low-carbon” energies hides the consideration of water and soil footprints and does not consider other impacts such as the extraction of mineral resources and the production of electronic waste, which is estimated at more than 2.5 million metric tons annually.

An added problem is that the energy, water and land footprint do not necessarily move in the same direction. “Switching from fossil fuels to bioenergy, for example, could give the impression that the carbon footprint is reduced by up to 70 percent, but at the same time it multiplies the water footprint by 30 and the land footprint by a hundred. “Low in carbon” is not synonymous with “low in water” or “low in territory,” explains Dr. Miriam Aczel, one of the authors of the report.

They also point out that the use of energy, with its consequent water and environmental impact, is very relevant in the AI ​​training process and increases exponentially with the use of AI systems, especially generative AI, such as ChatGPT and similar. Between 80 and 90 percent of energy and water demands are caused by the daily interactions of billions of users with these systems, currently 2.5 billion in the case of ChatGPT. A text query uses up to 60 times more energy than a non-AI search query, and 200 times more energy than a simple AI application, such as search engine classification. spam. Generating images with AI uses 1,450 times more energy and making a short video requires up to 200,000 times more.

They also report the enormous inequality between those who benefit and those who suffer the impacts. It is a myth that the location of a data center means greater access for the local population that suffers the impacts. The trend on the part of companies in the United States – which together with China have 90 percent of the AI ​​production infrastructure – is to locate data centers in places that already suffer from water stress, which causes a shortage of electricity supply and increased costs for the populations of those places. One of the examples in the report is the installation of data centers in Querétaro, Mexico.

There are many deeply negative aspects in the development of artificial intelligence, environmental, social, labor, economic, political, dependency. The promotion of its supposed benefits is based, too often, on ignoring or hiding the brutal water, climate and exploitation impacts that it entails on thousands of local communities, among others.