Inside NVIDIA's answer to the AI data center water crisis

A report by the United Nations University Institute for Water, Environment and Health (UNU-INWEH) found that the water used by artificial intelligence is expected to equal the basic annual water needs of 1.3 billion people by 2030. However, NVIDIA might have just found a way to address that.

On Monday at London Climate Week, NVIDIA announced that its latest infrastructure can be fully cooled with liquid warm enough to reduce—and in most climates, eliminate—the need for additional chilling equipment.

The announcement lands at a moment when data center water use is under growing scrutiny. The infrastructure choices teams make in the next few years will determine whether AI deepens that strain or helps relieve it.

What it does

NVIDIA's Rubin-generation platform is the world's first AI server architecture to achieve 100% closed-loop liquid cooling. Instead of relying on evaporative cooling towers that continuously consume water, it circulates a sealed mixture of approximately 75% water and 25% propylene glycol directly over the chips.

While previous liquid-cooled deployments typically targeted supply water in the high‑20s to low‑30s Celsius, Rubin raises that ceiling to 45°C—high enough to reject heat via dry coolers using ambient air alone, with no chillers and no evaporation. 

“The NVIDIA DSX reference design for AI factories has zero water consumption—we have eliminated massive amounts of power usage and pretty much all water usage,” says Ali Heydari, director of data center cooling and infrastructure at NVIDIA. “With dry-cooler-based designs, it’s a closed-loop system with no evaporative water cooling—outside of maybe 1% of the year when we might need chillers in some climates.”

John Parker, NVIDIA’s chief sustainability officer, went so far as to claim that “the water consumption challenge for data centers is largely solved” in an interview prior to London Climate Week. 

Why it matters

The average data center requires a vast amount of water to function, which rapidly adds up. A medium-sized data center can consume up to roughly 110 million gallons of water per year for cooling, which is equivalent to the annual water usage of approximately 1,000 households. 

Large data centers, meanwhile, can consume up to 5 million gallons per day or about 1.8 billion annually, comparable to the annual usage of a town of 10,000 to 50,000 people. 

Many are understandably concerned about the impact that AI could have on the environment. In a 2025 poll by the University of Chicago and The Associated Press–NORC Center for Public Affairs Research, 72% of Americans reported they were at least somewhat concerned about the environmental impact of AI, while 41% were at least very concerned.

Fortunately, through liquid cooling, NVIDIA can reduce facility cooling water consumption from roughly 2.6 million gallons per megawatt per year for conventional cooling-tower-based systems to near zero, making for a far more efficient and far less resource-intensive system. 

NVIDIA’s new cooling system doesn’t just benefit the environment, either. By raising chiller plant temperatures by a single degree, data centers could save 4% to 5% in energy costs—which, for a 50-megawatt hyperscale facility, could save millions annually in cooling-related energy and water costs.

Industry adoption and challenges

NVIDIA is not the only one moving in this direction. Microsoft has already announced that its next-generation data centers will use closed-loop, chip-level cooling designed to consume zero water for evaporation and began rolling that design into new builds in late 2024. 

Even so, getting there is not as simple as dropping Rubin-class racks into an existing hall. Most current facilities were designed around air cooling or legacy water-cooling plants, with tower and chiller capacity sized for much colder water and very different load profiles. In many cases, adopting something like NVIDIA’s DSX reference design will mean major plant changes or greenfield builds rather than incremental retrofits.

What this means for the future of AI infrastructure

NVIDIA's announcement represents a structural shift in how AI infrastructure is built. Instead of treating cooling and water use as an afterthought, Rubin brings them into the core design of high‑density AI systems, with DSX providing a concrete blueprint for how those systems can be deployed at scale.

If these Rubin‑class systems and similar designs perform as expected, that shift is unlikely to stay limited to NVIDIA. Water is already one of the main issues communities and regulators consider when they review new data center projects, and that scrutiny will influence which regions can keep adding AI capacity and which run into hard limits first.

For most enterprises, these changes will show up indirectly. You might not decide how a data center is cooled, but you do choose cloud regions, colocation partners, and NVIDIA‑based platforms—and those choices determine the performance, cost, and environmental profile of the AI capacity you rely on. As providers standardize on designs like Rubin and DSX, questions about water use are likely to join latency, reliability, and price in the list of things worth asking about.

As a partner in the NVIDIA Partner Network, FullStack helps teams make those decisions with clearer trade‑offs. We design AI backbones on NVIDIA’s latest stack—from Rubin‑class infrastructure patterns to NVIDIA AI Enterprise and agentic toolkits—and model how different options affect cost, risk, and the constraints your organization cares about, including security, sovereignty, and sustainability.

Contact us today if you’re interested in learning more about what we can offer your business.