Behind Every AI Prompt Lies a Water Bill—Who’s Paying It? AI’s Water Consumption Now Exceeds Global Bottled Water Use: Study

A new peer-reviewed study has highlighted a striking but often overlooked consequence of the rapid expansion of artificial intelligence: its demand for water. The research suggests that the annual water consumption of AI systems could now exceed the total amount of bottled water consumed around the world.

The study led by Dutch researcher Alex de Vries-Gao, set out to quantify the environmental footprint of AI workloads, focusing on the sprawling network of data centres that power modern AI applications. However, the analysis faces a key challenge: most major technology companies do not separately disclose the energy and water use specifically to AI versus other computing tasks. To work around this, the research used industry environmental reports and estimated AI’s share of data centre activity to arrive at its figures.

Water Use: A Surprising Metric

Based on the modelling in the paper, AI-related data centre operations may draw between 312.5 billion and 764.6 billion litres of water per year. To put that in perspective, this range surpasses the volume of bottled water consumed worldwide annually. While data centres require water directly for cooling equipment, a significant amount is also embedded indirectly through the generation of electricity used to power and cool AI infrastructure.

Carbon Footprint on Par With a Major City

The environmental impact extends beyond water. The study estimates that AI systems could produce between 32.6 million and 79.7 million tonnes of CO₂ emissions in 2025—a footprint comparable to that of a large urban area like New York City. These figures highlight that AI’s environmental costs are not limited to energy alone but span multiple resource categories.

Inference Drives the Demand

Contrary to the common focus on the energy-intensive training phase of large AI models, the research points to inference —the ongoing computation that happens each time a user interacts with an AI system—as a key contributor to environmental impact. Daily usage, from answering text queries to generating images and videos, involves constant computation that cumulatively drives higher energy and water use than training alone.

Calls for Greater Transparency

The study recommends for transparency. Because major AI companies generally include AI workloads within broader data centre reporting, policymakers and researchers lack clear, up-to-date data on the actual environmental costs of AI. The study highlights that better disclosure of resource consumption specific to AI could inform more effective regulation and sustainability strategies.

As AI becomes a permanent layer of the global digital economy, its environmental footprint can no longer be treated as an afterthought. The question now is not whether AI will continue to grow, but whether its growth can be aligned with realistic limits on water, energy, and transparency.

Share