Nvidia's Liquid-Cooled AI Servers Target Data Center Water Use
The chip giant's new infrastructure runs at higher temperatures and eliminates air-cooling fans that consume water, potentially saving millions annually.

Nvidia unveiled a server infrastructure design on Monday aimed squarely at one of AI's growing environmental controversies: data center water consumption. The company's newest AI servers will rely entirely on liquid cooling, removing the need for traditional air-cooling fans that draw on water resources.
The system circulates a closed-loop coolant mixture of water and propylene glycol to dissipate heat, with no requirement to pull in fresh water during operation. According to Ali Heydari, Nvidia's director of data center cooling and infrastructure, the design "eliminated massive amounts of power usage and pretty much all water usage."
A higher temperature threshold
The technical advance centers on operating temperature. Nvidia's coolant remains functional at up to 45°C (113°F), significantly higher than the industry standard of 30°C. That difference matters because warmer cooling systems can exhaust heat to the outside environment without running energy-intensive air conditioning units when ambient temperatures allow.
"The thing that's exciting about what Nvidia announced is it shows really what's possible in terms of pushing up this liquid input temperature to 45°C," said Andrew A. Chien, a computer science professor at the University of Chicago who directs the CERES Center for Unstoppable Computing. "Because if it's cool enough outside, you don't need to" run HVAC systems.
Chien's center has spent a decade researching data center efficiency. He noted the approach contradicts conventional wisdom but delivers meaningful environmental benefits, though he cautioned that truly zero water use remains unrealistic.
Why it matters
Data center water consumption has emerged as a flashpoint in communities hosting AI infrastructure. Residents near facilities have reported contaminated water supplies, reduced pressure, and unauthorized water diversion. The United Nations projected earlier this month that AI-related water use could match the annual needs of 1.3 billion people by decade's end. Nvidia's design shift—and similar moves by Microsoft, which announced water-free cooling for new data centers in August 2024—signals the industry recognizes regulatory and social pressure mounting around resource consumption.
Economics and adoption
Nvidia estimates a 50-megawatt hyperscale facility could save more than $4 million annually in cooling-related energy and water costs by switching to liquid-cooled infrastructure. The company did not respond to questions about system costs or plans to retrofit existing data centers.
The expense of liquid cooling systems remains a barrier. Chien acknowledged the technology costs more upfront but said the direction reduces total power consumption for large facilities—a trade-off more operators should consider as environmental scrutiny intensifies.
Microsoft's 2024 announcement indicated its water-free cooling approach would save over 125 million liters per data center annually, demonstrating the scale of impact possible with alternative designs.
These details were first reported by Fortune.
This is an original analysis by the Omega editorial team. Source reporting: AI Watch.
Want systems like this working for your business?
Book a Call
