AI’s Hidden Toll: Data Centers Gulping Water and Gas as Tech Giants Chase Billions

Eleven gas-powered data centers alone could spew more than 129 million tons of greenhouse gases yearly. That’s like Morocco’s total output. AI firms race ahead anyway. Profits first.

Gizmodo laid it bare: over 3,000 new U.S. data centers proposed or building, many tied to fossil-fueled plants that dodge the grid. Behind-the-meter setups. Quick natural gas turbines. xAI’s Colossus in Tennessee fired up methane generators without permits, per the Southern Environmental Law Center. Natural gas hit records in 2025, says the U.S. Energy Information Administration. America now tops global new gas power development, thanks to data centers, according to Global Energy Monitor.

A Microsoft project in Texas? Permitted for 11.5 million tons of CO2-equivalent a year. Stargate across Texas and New Mexico? Over 24 million tons. These are caps, not actuals. Centers run hot.

Water Vanishes into Server Racks

Thirstier still. Data centers could guzzle 5 billion cubic meters of water yearly by 2027. That’s Europe’s estimate from the European Commission. Google slurped 6.4 billion gallons globally in 2023. One Iowa site hit 1 billion gallons in 2024—enough for all Iowa homes five days. Microsoft’s forecast? 18 billion liters by 2030, up 150% from 2020, reports The New York Times. Doesn’t count $50 billion in new deals.

North American centers used nearly 1 trillion liters in 2025, matching New York City’s yearly draw, per Mordor Intelligence data in Finance & Commerce. Phoenix area? 385 million gallons annually, says Consumer Reports. A single facility? Up to 5 million gallons daily, rivaling 16,000 households, per EPA via the same report. Indirect use—from power plants—dwarfs it. 90-93% of footprint there, notes Milwaukee Journal Sentinel.

Investors hound Amazon, Microsoft, Google for site-level disclosure. Water use jumped 51% from 2020-2024 to 5,637 megaliters. And AI accelerates it.

Google’s emissions climbed 48% since 2019. Microsoft’s 29% since 2020. Both blame AI data centers, per their reports cited in NPR. Scope 3 indirects rose 150% average 2020-2023 for the big four, UN’s ITU finds. Amazon up 182%.

But. Numbers vary wildly. Agriculture swallows 80-90% of U.S. water. One pound of beef? 1,847 gallons. 21.7 million ChatGPT prompts match a burger. Centers: 0.3% nationally. Local hits hardest. Drought zones. Iowa. Phoenix.

Backlash Builds, Fixes Falter

Communities revolt. New York Times tracks opposition stalling builds: local fights. New Brunswick, N.J., ditched data centers for parks. San Marcos, Texas, council voted no after 100 comments. Ireland eyes moratoriums. UK court challenge over approvals. Maine’s first statewide AI data center pause, per recent X posts.

Companies pledge net-zero by 2030. Water-positive too. Microsoft eyes zero-water cooling designs. Cold plates cut emissions 15%, water 30-50% lifetime, their Nature study claims. Google measures Gemini prompt: 0.24 watt-hours, 0.26 ml water. Five drops. Immersion cooling. Waste heat for purification? Europe’s pitching it.

Nuclear revives. Microsoft restarts Three Mile Island. Amazon buys Talen nuclear. Meta hunts sites. Gas lingers, though. GE Vernova’s turbines backlog to 2029. Vistra, Constellation surge on deals.

Projections alarm. U.S. centers to 150 GW by 2028, Spain-sized, Bloom Energy says. AI servers: 165-326 TWh by then. Global data centers: 1,050 TWh by 2026, IEA via AIMultiple. By 2030, 3-4% world electricity. Or more.

Greenpeace calls greenwashing: 74% AI climate claims unproven. No big emissions cuts from ChatGPT et al. Yet AI chugs on. Hyperscalers spent $320 billion on infra last year. Projections: $700 billion capex cycle. Nvidia thrives. But grids strain. Bills rise. EESI links centers to hikes.

Trump’s order pushed private power. No strings. Gas renaissance. xAI skirts rules. Emissions permitted high. Actuals? Opaque. No AI-specific metrics from firms, notes science papers.

Fragment. Trade-offs stark.

And so the race continues. Billions in revenue. Trillions in capex. Planet pays. Investors push transparency. Regulators eye moratoriums. Tech innovates cooling, hunts nukes. But demand explodes. No off switch.

One truth. AI’s footprint grows with models. GPT-3 training? 700,000 liters evaporated. Inference now 80-90% of compute. Queries multiply. 760 billion liters in 2026? Digiconomist warns 80 million tons CO2 yearly—New York scale.

Local pain real. Global math messy. Beef burgers thirstier per use. But scale shifts. AI’s not pausing for fixes.