Water is essential to life—that’s not something new. It’s clear as daylight that humans all around the world open faucets daily to utilize water to shower, cook, and drink. But water is a privilege in drought-stricken areas; it is like finding a flower in the desert. In towns across these regions, the water that once flowed freely, created by nature to satiate human thirst, has become scanty like pearls in an ocean—something that demands effort, anxiety, and vigilance to be sought.
This problem is further enhanced by the so-called digital and magical era of AI—yes, the LLMs (Large Language Models) that complete all forms of writing, creative enough to mimic human tones and create art of any kind within seconds. The drawback: these vast databases and machines, smoke-screened by the power of AI, are consuming millions of litres of water.
This isn’t a tale of nature versus technology. It’s the uncomfortable intersection of two truths: water is life, and AI increasingly depends on industrial-scale water access. And as generative models like GPT-4o and future AI agents multiply across our lives—echoing the ambitions outlined by Sam Altman in his blog—they intensify the hidden thirst of our world.
Water and Energy Consumption of AI Data Centres
When people use AI—whether generating a business summary, translating a text, or asking for a poem—they evoke a vast energy infrastructure. AI lives in data centres: concrete warehouses equipped with servers, cables, and humming fans. These places are refrigerators for the digital age, battling never-ending heat.
Over the past three years alone, over 160 new AI-focused data centres were built across the U.S.—a 70% increase from the prior three-year period. Many of these are located in water-scarce regions. It is not only in the U.S., but countries like Saudi Arabia and the U.A.E. are also welcoming these data centres with open arms in order to be part of the AI race, even though they are well aware that they are water-deficit countries.
Experts have warned that tech giants neglect water availability and instead choose sites that best suit their pockets—cheap and favourable in regulations. This results in many AI data centres being ‘water guzzlers’ located in drought-prone areas, posing risks to local water supplies, agriculture, and energy production.
How Do These Data Centres Use Water?
Well, data centres literally absorb water in every possible way—both directly and indirectly.
Cooling Servers (Direct Water Use):
Large data centres withdraw and utilize enormous volumes of water to prevent overheating. According to estimates, data centres have the capacity to evaporate about 1–9 litres of water per kWh of energy for computing. The drawback is that they evaporate 80% of the water they take in—only 20% of the water is returned as wastewater. In contrast, households typically return 90% of the water they use to the source.
Power Generation (Indirect Water Use):
Most of the energy used to drive data centres comes from thermoelectric power plants (coal, gas, nuclear) or hydroelectric dams that require water. Both withdraw significant amounts of water for cooling and steam. According to researchers, almost 60% of a data centre’s total water use comes indirectly through electricity production.
Imagine a thin trickle from a rural faucet, just enough to fill a single morning mug. That same volume, in some places, is consumed silently in a server room cooling loop. While we see our reflection in a mirror, the mirror’s image depends on water that communities can’t afford to spare.
Communities Trapped in a Bottleneck
The statistics hide real lives. In places like Uruguay and Chile—where people already walk to school with empty water jugs—communities are alarmed when an AI-driven data centre proposes to drink millions of litres per day. In fact, data centres have become one of the top 10 water-consuming industries.
Teachers keep water jugs because taps fail; mothers collect bottles for bedtime sips; large shops limit purchases because demand outpaces supply. Then news arrives: “Cloud centre planned nearby.” Suddenly, the question is not “what if,” but “why first us?”
People don’t hate AI; they hate being excluded from decisions about whether water—the water they depend on—should be diverted to heat-handling machines.
According to reports, a 100-megawatt data centre in the U.S. consumes about 2 million litres of water per day for cooling—the same amount used by 6,500 average American households. In Memphis, Tennessee, a surge in AI-related industrial activity was met not with cold code, but with hot turbines and exhaust fumes. Families felt it in their throats and lungs—a reminder that compute power has consequences when it meets human ecosystems.
Scale and Momentum: Sam Altman’s Vision
In a blog post titled “The Gentle Singularity,” Sam Altman described a cascade: AI agents become cheaper, smarter, and more capable—by 2025, able to perform cognitive tasks; by 2027, embedded in robotics. He imagines a world of “abundant intelligence,” where humans are turbocharged, free from repetitive tasks.
What he failed to mention is that data centres currently draw 560 billion litres of water per year, and this could double to 1.2 trillion litres by 2030 if AI adoption accelerates.
That vision is thrilling—but it implies infrastructure that is equally vast. If every human—and then every household—relies on AI at scale, then every AI question, request, or task compounds water and energy demands. That gentle singularity starts to feel thirsty.
Altman doesn’t foreground water in his optimism, but intelligence without sustainability is brittle. If we don’t chart a path that accounts for water, that generosity of predictive text could come at the expense of daily cups of clean water.
Globally, the AI-driven demand for water is becoming impossible to ignore. By one estimate, the water consumption of AI data centres and related power generation could soon rival that of other notoriously thirsty industries like cattle farming and textile manufacturing.
This surge threatens to complicate international sustainability efforts. Notably, the world’s largest cloud/AI companies—Amazon, Google, Microsoft—had all pledged to become “water positive” (adding more water to stressed basins than they use) by 2030, as part of corporate sustainability goals. But the way AI is expanding now puts these promises at risk .
It is ironic that regions blessed with renewable resources (like deserts) are often water-scarce, leading to unwanted and difficult trade-offs. Even if data centres switch to air cooling systems to save water, the machines demand more electricity, which is drawn from fossil fuels that consume water elsewhere. In other words, to save one resource, we strain another. It’s a tricky balancing act—water and energy—that AI developers and policymakers are only just beginning to confront.”
Cooling: Not a Free Lunch
Promises to be “water positive” by 2030—meaning projects to replenish more water than is used—feel reassuring. But if those replenishment efforts happen in distant watersheds, while drinking water dries locally, then families still lose. Replenishment helps in the aggregate—but does not necessarily restore what a specific community lost.
The human lens focuses on the faucet, not the aggregate ledger.
Smart Cooling, Smarter Policy
There are brighter ways forward—places like Sweden show the potential. Some data centres there capture server waste heat and send it to district heating systems for homes or wood-drying facilities. They sit in cold climates, with cheap renewable energy, and use ambient air for cooling. Circular design helps communities, not just digitize them.
But technical improvement isn’t enough without governance that reflects human rights. Communities in water-stressed areas deserve:
- Transparent water accounting: We need to know how many litres and from which source.
- Community consultations: Before permits are approved—a genuine, respectful conversation.
- Impact assessments with climate modelling: Not just paper reports but data-driven scenarios.
- Local mitigation promises: Water restoration specifically where withdrawals occur, not only far away.
When regulators denied initial data centre permits in Chile and Uruguay, they showed that legal frameworks, though currently weak, can evolve to protect water rights.
The Human Texture of Policy
Forget the spreadsheets. Think of a grandmother climbing two flights of stairs to fetch water because her faucet went silent. Or students arriving at school faint from dehydration. Or the hum of turbines and smell of exhaust that suddenly became a neighbour’s reality.
These are not the texts of change—they are the voices of harm.
On the policy front, more voices are rising to inculcate the awareness that AI needs to be stopped before the water crisis worsens. Transparency is the greatest step under discussion, i.e., companies honestly disclosing how much water their data centres utilize. Currently, there are no rules in place that can force data centres to share this information, which is why many companies hide it like trade secrets. Due to this, communities and officials are left in the dark and hence are unable to plan for droughts and the impact these will have on their water supplies.
Some progress is underway. The Europe Union in 2024, as part of an updated Energy Efficiency Directive introduced new rules requiring large data centres to track and report their water and energy use to regulators. This EU policy will create a central database of data centre water metrics (accessible to authorities) and sets an important precedent for holding AI infrastructure accountable.
In the United States, a group of lawmakers proposed the Artificial Intelligence Environmental Impacts Act to develop standards for measuring AI’s water and carbon footprint and to establish a voluntary reporting framework. While not yet law, it reflects a recognition at the federal level that AI’s resource consumption must be understood and managed. Individual states and cities are also starting to demand water disclosures in permitting new data centres, spurred by cases like The Dalles, OR, and Phoenix, AZ, where public trust was shaken by secret water deals.
Global civil society has shown it can play a role in checking AI’s water impact. As noted, community protests in places as far apart as Chile, the Netherlands, and the U.S. have led to permits being revoked or projects halted until water issues are addressed. These steady acts of resistance send a clear message to the industry that the locals will not stand silently as AI takes their water.
Some tech companies have responded by investing in water conservation measures: for example, Google and Microsoft have announced initiatives to use recycled wastewater for cooling, to install more water-efficient cooling technology (like closed-loop systems), and even to replenish watersheds in the regions where they operate.
Such steps are welcome, but experts argue they remain insufficient without broader transparency and enforceable limits. Ultimately, the global conversation around AI is beginning to expand beyond algorithms and ethics to include ecological footprints. Water, being finite and irreplaceable, is emerging as a central concern. Unless proactive measures are taken, AI’s growing “thirst” could intensify water insecurity in many parts of the world—a paradoxical outcome for a technology often touted as a tool for sustainable development.
Policy must protect those invisible lives, not just spreadsheets. It can—but only if leaders embrace values rooted in caring. Building data regulations with human-centred care, not just bytes and terawatts.
Small Acts, Big Impacts
As a user ,one should question before giving a prompt, ‘Is it necessary?’ Do we need massive AI models for “tell me the time in French”? Can software developers prioritize compact, low-energy models for everyday tasks? If every user demands efficiency in AI design, it shapes the infrastructure’s footprint.
As communities, we can insist on siting data centres where cooling doesn’t compete with coastal aquifers. As policymakers, we can demand that “innovation” not be shorthand for unaccountable extraction.
A Future That Doesn’t Drown in Progress
Abundant intelligence is a noble dream—where ideas flow, creativity blooms, productivity thrives. But it costs something material. When the tap runs dry, no Google Cloud or GPT-4o can refill it. Intelligence without life is not liberation—it is absence.
We stand at a choice. We can let the servers drink, and our community’s thirst. Or we can build AI into a future that protects the water we all need—for reading, for cooking, for growing. Where the promise of AI flows alongside the certainty of clean water, not instead of it.
That is the boundary of judged progress—progress that doesn’t carve away human dignity. Because nothing smarter—no matter how capable—matters more than the daily miracle of a child drinking from a tap without worry.
