The AI Thirst Problem: How the Global Data Centre Boom Is Draining Communities From Georgia to Gujarat
The AI Thirst Problem: How the Global Data Centre Boom Is Draining Communities From Georgia to Gujarat
A scandal over 29 million gallons of unmetered water in suburban Atlanta has cast a harsh light on a systemic resource conflict—one that is now being exported, at $210 billion scale, to some of India's most water-stressed states.
When residents of Annelise Park, a leafy subdivision 20 miles south of Atlanta, noticed their water pressure dropping last year, local authorities issued a conservation advisory and asked households to stop watering their lawns. The culprit, it turned out, was not a drought—though Georgia was in the midst of one—but a 615-acre data centre campus codenamed Project Excalibur.
The facility, developed by Quality Technology Services (QTS), a company owned by Blackstone, the New York-based private equity giant, had quietly drawn nearly 29 million gallons of water through two connections that Fayette County utility officials did not know existed. One had been installed without the county's knowledge; the other had never been linked to QTS's billing account. A public records request by a local resident eventually surfaced a letter from the county demanding $147,474 in retroactive charges—the cost of 44 Olympic swimming pools of water.
The company paid the bill. It was not fined.
"When residents complained of low water pressure in Georgia or dust control efforts in Arizona, they unknowingly tipped off regulators in areas fraught with depleting water supplies."
The episode—now one of the most widely cited examples in a growing national debate over data centre accountability—is far from isolated. It is, rather, a symptom of a structural collision between the AI infrastructure boom and the physical limits of freshwater supply. And as investment capital chases that boom eastward, the collision is being set up, at vastly larger scale, across India.
The numbers behind data centre water consumption are staggering and accelerating. US data centres directly consumed 17.4 billion gallons of water in 2023, a figure the Environmental Protection Agency projects will rise to between 38 and 73 billion gallons by 2028. In Texas alone, a study by the Houston Advanced Research Center estimated data centres would consume 49 billion gallons in 2025 and as much as 399 billion gallons by 2030—equivalent to lowering Lake Mead, America's largest reservoir, by more than 16 feet in a single year.
Water is used primarily to cool the vast banks of servers that power artificial intelligence. In evaporative cooling systems—among the most prevalent—hot air is passed through towers of water, which absorbs heat and is then lost to evaporation. Up to 85 per cent of water drawn this way does not return to the supply. A medium-sized data centre can consume up to 110 million gallons per year; a larger hyperscale facility can consume five million gallons per day, the equivalent of a town of up to 50,000 people.
The Brookings Institution notes that training a single advanced AI model can directly evaporate 700,000 litres of clean freshwater. Researchers at the University of California, Riverside, estimated that every 100-word AI prompt consumes roughly one 500ml bottle of water. With billions of users issuing prompts every day, aggregate consumption reaches industrial scale almost invisibly.
About two-thirds of data centres built since 2022 are located in areas already experiencing water stress, according to a Bloomberg News investigation. Georgia's Public Service Commission froze Georgia Power base rates through 2028 specifically to prevent data centres from shifting electricity costs to residential customers. Fayetteville city council has since banned new data centres in every zoning district under Ordinance 26-0-12.
In February 2026, Prime Minister Narendra Modi convened India's AI Impact Summit in New Delhi, a five-day event that drew OpenAI chief executive Sam Altman, Alphabet chief executive Sundar Pichai, and a roster of global technology leaders. The summit produced headline-grabbing commitments. Reliance Industries announced roughly $110 billion in AI and digital infrastructure investment; Adani Group pledged $100 billion to build renewable energy-powered hyperscale data centres by 2035. Together, the two conglomerates committed $210 billion—a sum that, Adani argued, would catalyse a $250 billion AI infrastructure ecosystem in India over the decade.
The Adani push builds on AdaniConneX, a joint venture with US-based EdgeConneX, and on a strategic partnership with Alphabet's Google, which had announced a $6 billion data centre in India in mid-2025—its largest in Asia. A separate Adani-Google campus is planned in Visakhapatnam; Microsoft is developing campuses in Hyderabad and Pune. The summit produced an additional $50 billion commitment from Microsoft for AI access across the Global South.
The ambition is sovereign and strategic: Adani Group executive director Jeet Adani warned that India must build its own AI infrastructure or become dependent on imports. The vision is framed in terms of energy security, job creation, and the desire to move up the global value chain.
What goes largely unmentioned in the investment presentations is water.
"India's data centre water consumption is projected to rise from 150 billion litres in 2025 to 358 billion litres by 2030—more than doubling within five years."
India is already the world's largest extractor of groundwater, accounting for 25 per cent of global extraction. A 2024 satellite-based assessment found that North India lost nearly 450 cubic kilometres of groundwater between 2002 and 2021, with water tables dropping roughly 1.5 centimetres per year. The Central Ground Water Board's 2025 compilation classifies 26 per cent of India's groundwater blocks as "over-exploited," "critical," or "semi-critical." In Punjab, extraction exceeds annual recharge by 156 per cent; in Rajasthan by 147 per cent; in Haryana by 137 per cent—precisely the states that host or border the data centre corridors now being developed.
A 2025 analysis by UK non-profit Planet Tracker found that China, India, and Japan have the highest concentration of data centres in Asia; India ranks second only to Indonesia in the number of centres located in "extremely high" water-stressed regions, with 50 facilities already in such zones. A report by Earth Journalism Network identified Bengaluru's Devanahalli district—one of the fastest-growing data centre corridors in the country—as having no perennial source of water at all, relying entirely on groundwater.
In India, the order of scale and the absence of commensurate oversight is striking. A January 2026 United Nations report coined the term "global water bankruptcy," arguing that the familiar language of water stress no longer captures the structural imbalance between human demand and ecological limits. India's case, the report suggested, illustrates that imbalance with particular force: demographic pressure, climate variability, intensive agriculture, and now, a digital infrastructure push, are converging on aquifers that cannot replenish themselves at the rate of extraction.
India's Union Budget has incentivised data centre construction but lacks equivalent measures for cloud infrastructure, algorithms, or human capital—the other limbs of the AI value chain. Critics argue that India risks expending water and power to generate economic returns that flow primarily abroad, to the hyperscalers whose services the data centres will host.
The World Economic Forum's 2025 Global Risks Report named water supply shortages as India's single largest environmental threat for 2025–2027. The number of countries globally listing water shortages as a top-five national risk jumped from seven in 2024 to 27 in 2025.
Fayetteville, Georgia became the first US city to ban new data centres across all zoning districts. Georgia's legislature debated a moratorium on approvals statewide. Senator Bernie Sanders proposed a national pause on data centre construction. The Clean Cloud Act of 2025 would require facilities to report water and energy consumption publicly, monitored by the EPA and the Energy Information Administration.
QTS, for its part, argues that its Fayetteville campus uses a closed-loop cooling system that does not consume water for operational cooling. The water drawn unmetered, the company states, was consumed during construction—concrete mixing, dust suppression, site preparation—and will not recur once the campus is fully operational. The county accepted this explanation and did not impose any penalty beyond the retroactive charge.
Whether that explanation fully reassures local residents is another matter. Georgia remains in moderate-to-severe drought. The campus, at 615 acres, is one of the largest in the United States and continues to expand; full build-out may take several more years of construction-phase water use.
The conflict over AI and water is, at its core, a conflict over who bears the costs of infrastructure that benefits a global minority. In Georgia, the costs were borne by Annelise Park residents who were told to stop watering their lawns. In Haryana and Rajasthan, where farmers already face failing crops and sinking water tables, the costs of a $210 billion data centre buildout have yet to be fully calculated—let alone disclosed.
Upcoming United Nations Water Conferences in 2026 and 2028 are intended to reset global water governance. India's own Jal Shakti ministry has initiated programmes including the Atal Bhujal Yojana for community-led water budgeting. Whether those frameworks can keep pace with the speed and scale of private investment in AI infrastructure is the central question neither the summit communiqués nor the investment prospectuses address.
The water beneath Georgia and the water beneath Rajasthan are separated by thousands of miles. The capital that is drawing down both is rather more concentrated.
Politico (May 2026) — QTS / Fayette County water letters via public records request
Fortune (May 2026) — "America's data centers are thirsty. Rural towns are paying the price"
US Environmental Protection Agency (2025) — Data centre water consumption estimates
Brookings Institution (April 2026) — "Global energy demands within the AI regulatory landscape"
CNBC / TechCrunch / Millennium Post (February 2026) — Adani and Reliance AI Impact Summit announcements
Earth Journalism Network (November 2025) — "India's tech boom collides with deepening water crisis in Bengaluru"
SANDRP (March 2026) — "Groundwater 2025: depletion and contamination rising"
ORF (March 2026) — "Arresting India's groundwater depletion to avert water bankruptcy"
Planet Tracker / World Resources Institute Aqueduct Atlas (2025)
Central Ground Water Board, India — 2025 national groundwater compilation
United Nations (January 2026) — "Global Water Bankruptcy: Living Beyond Our Hydrological Means"
World Economic Forum — Global Risks Report 2025
Tom's Hardware / Benzinga — Georgia data centre moratorium reporting (2026)
This article was compiled from open-source and authoritative news reporting for analytical purposes. All data points are sourced from named publications and governmental or intergovernmental bodies. No proprietary financial data has been reproduced.
Comments
Post a Comment