“An irreversible cost.” With these words, UN experts warn about what the development of Artificial Intelligence could imply by the end of the decade. Estimates from the report by the United Nations University Institute for Water, Environment and Health (UNU-INWEH), published this Wednesday, indicate that by 2030 data centers will consume three times more electricity than Pakistan, Bangladesh, and Nigeria: 945 terawatt-hours. These three countries combined have more than 650 million people.
Read more U.S. destroys two Iranian drones operating in the Strait of Hormuz
In terms of water footprint, what AI would consume would be similar to what the entire population of Sub-Saharan Africa needs today in terms of water: 1.3 billion people. As for land use, it would be double the size of the Jakarta metropolitan area, the capital of Indonesia: 14,500 square kilometers dedicated solely to the development of Artificial Intelligence.
These are some of the conclusions of the report “Environmental Cost of AI Energy Use: Carbon, Water, and Land Footprints.” Its authors indicate that, until now, the emphasis on measuring the carbon footprint of large AI models ignored both the water and land they consumed. Water is mainly used to cool data centers and produce the electrical energy they consume.
Among the data to consider is that, for now, only the energy needed to train large models, such as GPT-3 and GPT-4, had been measured. The issue is that this perspective would be obsolete, as the models are “deployed,” so they spend energy mainly on “inference,” i.e., the continuous processing of user queries, which today represents between 80% and 90% of AI consumption in terms of electricity.
ChatGPT processes about 2.5 billion queries a day, which translates to about 383 gigawatt-hours per year. If an attempt were made to offset its carbon emissions, 2.6 million tree seedlings would be required for 10 years: an area the size of Manhattan. Meanwhile, the water footprint is equivalent to the annual water needs of half a million people in Sub-Saharan Africa.
It should be considered that not all queries are equal. A typical conversation with a chatbot consumes 200 times more energy than a simple text classification. Generating a single image consumes 1,450 times more. A short AI-generated video can consume as much electricity as 200,000 spam classifications.
The director of UNU-INWEH and research leader, Professor Kaveh Madani, assures that this report “is not a manifesto against Artificial Intelligence.” “It is a call to use it responsibly and to proactively address its unintended impacts, to make it sustainable and equitable,” he states.
Regarding one of the problems indicated in the report, there is the issue of focusing too much on the carbon footprint. This, considering that trying to lower it can end up increasing the others. Switching from coal to bioenergy, for example, reduces the carbon footprint by 70%, but multiplies the water footprint by 30 and the land footprint by 100. “Low carbon” is not synonymous with “low water” or “low land.”
“What surprised us most was how often options that seem greener from a carbon perspective end up being worse for water or land,” said Miriam Aczel, UNU-INWEH researcher and lead author of the report, in a press release. “If we continue to judge AI sustainability solely on carbon, we might think that renewable energies make AI infrastructure clean, but that solves one problem while creating others, often in places where they are not requested,” she indicated.
As early as 2025, global data centers consumed 448 terawatt-hours. If they were a country, they would be the eleventh largest electricity consumer in the world, behind France and ahead of Saudi Arabia, the study points out.
Read more Kim Jong-un’s sister states that North Korea’s nuclear status is «irreversible»
Another problem, the report indicates, is how this affects the territories where data centers are installed. “The massive expansion of AI is creating very unequal localized tensions. In Ireland, data centers accounted for 21% of measured electricity in 2023, surpassing all urban households. The grid operator has halted new approvals in Dublin until 2028,” the text states.
In Querétaro, Mexico, meanwhile, the expansion of this technology is depleting water supplies, precisely in an area where there have been prolonged droughts. Similar is the case of Uruguay, where a data center was projected precisely at a time when freshwater reserves in Montevideo were running out,
To this is added waste: AI infrastructure could generate up to 2.5 million tons of electronic waste per year by 2030. Most of this material would be processed in low-income economies with scarce safeguards, while critical minerals are extracted in jurisdictions with little environmental oversight.
If that’s not enough, the UN also highlights the unequal way countries share their role and profits in AI development: Only 32 countries in the world host specialized AI data centers. 90% of that capacity is concentrated in two countries, while more than 150 nations currently lack access to sovereign AI computing.
The problem is that, precisely, the countries excluded from AI development are those that bear the extraction of critical minerals and electronic waste. Strategic benefits, meanwhile, are concentrated in the United States and China.
“The concentrated development of AI infrastructure in privileged areas of the world is creating a large digital divide that poses profound challenges for the equitable development of AI,” declared Tshilidzi Marwala, Rector of the United Nations University and Under-Secretary-General of the United Nations. “AI can undoubtedly drive prosperity and human well-being. Whether it does so equitably is now a matter of governance, not technical.”
All in all, the report is auspicious for the development of this technology, and estimates that the global AI market will multiply approximately 25-fold in a decade, from 189 billion dollars in 2023 to nearly 5 trillion in 2033.
Business investment in AI exceeded 580 billion dollars in 2025, while generative AI alone attracted almost 34 billion in private investment.
Read more United States: At least 12 injured, two of them seriously, after shooting near a festival in Ohio
