Every time someone asks an AI chatbot to draft an email, summarize a document, or generate an image, a vast and mostly invisible infrastructure springs into action: racks of specialized processors running at near-capacity, drawing power from electrical grids, and requiring enormous quantities of water to stay cool.
The environmental ledger of this activity is only now being tallied — and the numbers are striking.
The Scale of Consumption
Data centers running large language models consumed an estimated 460 terawatt-hours of electricity globally in 2025, according to the International Energy Agency — roughly equivalent to the annual power consumption of France. Projections for 2030 suggest this could double or triple.
Water usage tells a similarly sobering story. Cooling a single large AI data center can require millions of gallons of water daily, drawn from local aquifers and river systems.
Who Bears the Cost
The environmental burden is not distributed equally. Many of the largest data center clusters are located in regions already experiencing water stress — Arizona, Texas, and parts of Western Europe among them. Communities near these facilities often see impacts on local water supplies while deriving limited economic benefit.
“We’ve essentially outsourced the environmental costs of AI to communities that often have the least say in whether these facilities are built,” says Dr. Sasha Rivers, an environmental justice researcher at MIT.
What the Industry Says
Technology companies argue they are investing heavily in renewable energy procurement and efficiency improvements. Several have pledged to operate on 100% clean energy by 2030.
Critics contend that accounting practices obscure the full picture — that renewable energy certificates purchased in one region don’t necessarily offset real-time fossil fuel consumption in another.
The debate will only intensify as AI adoption accelerates.