Half a liter of water per hundred words: AI's hidden water footprint

By Momentum IA · June 29, 2026.

Three Law and International Relations students at the Catholic University of Santa Fe (UCSF) did what major AI labs tend to avoid: put concrete numbers on the table and ask people whether they were aware of them. The occasion was the II Meeting of Law School Students of the Network of Catholic Universities, held at the San Juan campus of the Catholic University of Cuyo, with delegations from thirteen law schools across the country. The presentation, titled 'The Invisible Side of Artificial Intelligence', did not come from a think tank or an environmental NGO: it came from young people who, while studying law and papal encyclicals, found a gap between what society thinks it knows about technology and what actually happens inside a data center.

The figure underpinning the research is simple but revealing: generating a text of roughly one hundred words using artificial intelligence can require close to half a liter of water to cool the servers that process that request. It is a figure that, in isolation, may seem small. Multiplied by the billions of daily queries that platforms such as ChatGPT, Gemini or Claude receive, it becomes water consumption on an industrial scale. The authors —María del Pilar García Porta, Paz Trinadori and Anna Paula Gómez, accompanied by professors Sergio Daniel Mencarelli and Diana Moralejo— also conducted an anonymous survey of 255 people in the city of Santa Fe. 97% considered that access to drinking water should be recognized as a fundamental human right. Yet more than half were unaware that AI consumes water in its operation, and only a fraction deemed specific regulation of that consumption necessary.

This gap between stated values and actual knowledge is, in itself, a political finding. It is not enough for a society to support water as a right: it needs to understand where that resource is being drained from. Invisibility is the perfect mechanism for consumption to scale without social friction. No one protests what they cannot see. And the tech industry, in general, has been a master at maintaining that invisibility: data centers are installed far from users, in industrial parks or rural areas, and the 'cloud' has precisely that semantic virtue of evoking something ethereal, weightless, without a footprint.

Our reading is the following: the underlying problem is not AI itself, but the almost total absence of regulatory frameworks that require reporting its environmental impact with the same rigor with which, for example, CO₂ emissions are reported in manufacturing. Today, a cement manufacturer must declare how many liters of water it uses per ton produced. A language-model provider, no. That regulatory asymmetry is what makes research like that of these students almost archaeological: they excavate data that ought to be published in each company's sustainability reports.

The regional context adds a layer of urgency. While the students were developing their work, the province of Santa Fe was reforming its Provincial Constitution and expressly incorporated into Article 34 the right to water and to a healthy environment, recognizing water as an essential collective good. That a territory with the Paraná River at its doorstep —and with a history of floods and droughts that makes water an existential issue— is beginning to enshrine that right at the constitutional level is a sign that the concern is not academic: it is structural. And the question that follows is whether that constitutional protection will ever come to also regulate the water consumption of digital infrastructure that operates, often, in other jurisdictions but whose impact is global.

It is no coincidence that the presentation's theoretical framework appeals to the encyclicals Laudato Si' and Laudate Deum by Pope Francis. Beyond the meeting's faith-based context, those documents articulate with precision a critique of the technocratic paradigm that normalizes intensive resource use in the name of progress. It is a critique that transcends the religious: it points out that technology without an ethics of limits becomes a way of externalizing costs toward the most vulnerable —those who do not have guaranteed access to water and bear the consequences of its scarcity— to benefit those who can afford advanced digital services.

That said, Momentum IA does not subscribe to easy catastrophism. AI also has enormous potential to optimize water use: from drought-prediction models to precision irrigation systems that reduce agricultural consumption —which represents, as sector context, close to 70% of the fresh water used worldwide. The problem is not the tool; it is the direction it points in. An AI that helps a farmer in northern Argentina irrigate with surgical precision can return far more to the water system than it consumes in cooling. But that equation only adds up if there is political will to steer technological development toward those ends, and if citizens are informed enough to demand it.

In the long term, the trajectory is hopeful: AI models are becoming progressively more efficient in energy and water consumption; research into neuromorphic chips and edge computing promises to drastically reduce dependence on large data centers; and regulatory pressure —still timid, but growing in Europe and in some U.S. states— is beginning to demand greater transparency. But the short term is the one we live in today, and in that short term millions of people in regions under water stress share a planet with digital infrastructure that drinks water without regard for others' scarcity.

That this analysis is being carried out by three university students in a conference presentation —not by a team of consultants paid by a foundation— says something valuable about how the next generations of jurists and international relations specialists are being trained in Argentina. Research as a 'professional horizon', in García Porta's words, is exactly the attitude the technological transition needs: people who look beneath the surface, who connect everyday use of a chatbot with Article 34 of a Provincial Constitution, and who ask who pays, and with what resources, the cost of artificial intelligence.