The world observes World Water Day every 22nd of March, a United Nations observance that focuses attention on freshwater and the global water crisis. This year’s observance falls at an especially uncomfortable moment: the world is talking more about innovation, speed, and artificial intelligence, even as billions of people still lack access to safe drinking water. 

A Blue Planet with Almost No Drinking Water

Earth looks water-rich from afar, but that impression is badly misleading. About 96.5% of Earth’s water is in the oceans and saline waters, while only about 2.5% is freshwater. Even that freshwater is mostly not readily available to us: the U.S. Geological Survey says most of it is locked in ice or underground, and only about 1.2% is surface water. In other words, the share of all Earth’s water that is freshwater surface water is only about 0.03%. 

That distinction matters because public discussions often blur the difference between “freshwater,” “drinkable water,” and “accessible water.” They are not the same. Not all freshwater is safe to drink, and not all accessible water is equitably available. What we can responsibly say is this: only a tiny fraction of the planet’s total water supply is the kind people can realistically rely on for daily life. At the same time, 2.1 billion people still lacked safely managed drinking water in 2024, according to UNICEF and WHO monitoring data. 

That alone should make World Water Day less ceremonial and more political. Water scarcity is not just about drought. It is also about infrastructure, pollution, inequality, governance, and priorities. The crisis is not that the planet has no water. The crisis is that safe, usable water is scarce where and when many people need it most. 

The Hidden Thirst of Artificial Intelligence

This is where the conversation becomes even more difficult. Generative AI is often treated as if it exists in some weightless digital realm. It does not. It depends on data centers, electricity, cooling systems, and supply chains grounded in the physical world. A 2025 version of the widely cited paper Making AI Less “Thirsty” states that AI’s water footprint includes freshwater consumed for both on-site cooling and electricity generation. The same paper estimates that global AI demand could account for 4.2 to 6.6 billion cubic meters of water withdrawal in 2027, with 0.38 to 0.60 billion cubic meters actually consumed. 

The paper also gives concrete examples that make the issue harder to dismiss. It estimates that training GPT-3 in Microsoft’s U.S. data centers could consume 5.4 million liters of water in total, including 700,000 liters of direct on-site consumption. It further estimates that GPT-3 might “drink” a 500 mL bottle of water for roughly 10 to 50 medium-length responses, depending on where and when it runs. These are model-based estimates, not universal constants for every AI system, but they are serious enough to challenge the fantasy that AI is environmentally frictionless. 

Just as important, we should use careful language here. It is more accurate to say generative AI can consume clean freshwater, often potable water, than to claim every AI output straightforwardly “wastes drinking water.” The same paper notes that, in many cases, data centers require clean, potable water to avoid clogs and bacterial growth in cooling systems. It also cites one technology company whose self-owned data centers in 2023 consumed more than 23 billion liters of freshwater for on-site cooling, with nearly 80% of that being potable water. 

That reality becomes harder to justify when we look at what generative AI is increasingly being used for. The proliferation of AI-generated images is not hypothetical; it is already at an industrial scale. Adobe said in April 2025 that Firefly had generated more than 22 billion assets worldwide in under two years. That is only one platform, not the whole ecosystem. 

So yes, AI can help. It can improve drought forecasting, detect leaks, optimize irrigation, and model water systems. But that is only half the story. The other half is that we are also spending real energy and real water on a flood of novelty images, trend content, style experiments, and disposable visual clutter. The contradiction is glaring: on a planet where safe water remains out of reach for billions, we are scaling technologies whose hidden resource costs are still treated as a footnote. 

Rethinking What We Value

World Water Day should force a more honest question than the usual appeals to shorter showers and closed taps. It should make us ask what kind of economy we are building when one of humanity’s most precious resources is increasingly tied to the production of synthetic content at a massive scale.

If we accept that water is finite, then we must also accept that how we use it is a matter of collective choice, not inevitability. That means demanding transparency from the companies building and scaling AI, pushing for standards that prioritize water-efficient infrastructure, and refusing to treat resource-heavy digital excess as harmless. It also means rethinking our own habits: what we consume, what we create, and what we normalize. Because every system we build reflects what we value. If we can mobilize global resources to generate billions of images, we can mobilize that same urgency to ensure every person has access to safe water. The question is no longer whether we have the capacity. It is whether we have the will.

Water is not abstract. It is not infinitely reproducible. Unlike generated images, water cannot be regenerated with a prompt.