AI’s thirst for water sparks crisis as data centres strain global resources: Moody’s

High-performance computing workloads, such as artificial intelligence (AI) model training, generate significantly more heat than conventional computing tasks, increasing cooling needs and, potentially, water consumption, according to Moody’s Ratings.

Data centres cool their servers to prevent them from overheating. Most existing data centres use air-based cooling. But roughly half of much larger hyperscale or AI data centres rely on water-based cooling that requires less electricity than air-cooling systems but depends on local water supplies.

As cloud-computing and internet giants, or hyperscalers, build or lease new data centres to meet escalating computing demand, their water consumption may strain resources in some markets, cautioned Moody’s. Data centre developers that fail to use water-efficient or waterless cooling systems may face heightened scrutiny from regulators and investors, it said.

The proliferation of data centres has intensified competition for water resources, particularly in regions with high water stress, said Moody’s. This situation is creating tensions among local communities, policymakers, water utilities and industry stakeholders, underscoring the need for data centre developers to implement sustainable water management plans.

Governments in water-stressed regions are likely to adopt policies aimed at harmonising economic growth with environmental protection efforts, the credit rating agency said.

“Governments and utilities are implementing stricter regulations to promote sustainable water use and conservation, pressuring data centre developers and operators to adopt more water-efficient or waterless practices,” it said, adding that local governments are increasingly requiring detailed water impact assessments before approving new data centre projects.

“Some jurisdictions are introducing mandatory water-use reduction targets or requiring the use of air-based or new waterless cooling systems, which are being adopted by a growing number of AI-focused hyperscale data centres. But these new systems’ lack of long-term operational track records raises concerns about reliability and maintenance and hinders widespread adoption,” said Moody’s.

Data centres are among the most resource-intensive facilities in modern infrastructure. They require vast amounts of electricity to power the escalating computing needs of cloud-services providers and new AI applications, as well as growing volumes of water to manage the heat generated by their server components. Hyperscale data centres reportedly consume an average of 2.1 million litres of water per day, comparable to the daily usage of a small city.

Managing rising water consumption is a key challenge for the data centre industry, particularly in regions prone to droughts and water scarcity. High-performance computing workloads, such as AI model training, generate significantly more heat than conventional computing tasks, which increases cooling needs and water consumption. As groundwater shortages become more prevalent around the world, increasing public scrutiny of data centre water use is emerging as a credit risk.

Hyperscalers like Alphabet Inc.’s Google business, Microsoft Corporation, Amazon.com Inc.’s Amazon Web Services (AWS) unit and Facebook parent Meta Platforms Inc. have made strides in lowering average water use per data centre and have committed to being “water positive” by 2030 by replenishing more water than they consume. But the industry’s overall consumption of water continues to climb amid the sustained surge in hyperscale data centre capacity. Much of this capacity is leased from data centre developers that use the most efficient existing cooling technologies, which are typically water-based.