Nvidia says AI data centres can run with almost no water after Amazon revealed 9 billion litres of water usage

Nvidia has unveiled a liquid-cooled AI data centre design that it says can bring water use close to zero. The company says higher operating temperatures and closed-loop cooling could cut power demand, even as Amazon's figures show how water-intensive current data centres remain.

In Short

• Nvidia claims near-zero water use with fully liquid-cooled AI data centre
• Closed-loop system runs at higher temperatures to improve efficiency
• This comes as Amazon reports 9 billion litres of annual data centre water usage

Nvidia has introduced a new cooling approach for AI data centres that it says can bring water usage close to zero, as the industry struggles with rising energy demands from powerful AI systems. The company’s latest Rubin-generation infrastructure uses a fully liquid-cooled design that removes traditional air cooling and significantly cuts reliance on water-based cooling methods. The goal is to make high-performance AI computing more efficient without adding pressure on local water resources.

In this system, heat is removed directly at the chip level using liquid flowing through sealed cold plates. The coolant absorbs heat and circulates through a closed loop, meaning it is reused continuously instead of being consumed. Nvidia says this setup allows data centres to move away from large cooling towers and fan-heavy systems that typically drive up both energy use and water consumption.

A key change is the temperature at which the system operates. Nvidia says its cooling liquid can run safely at up to 45 degrees Celsius, or 113 degrees Fahrenheit, which is significantly higher than conventional data centre cooling environments. Instead of being a drawback, the company says this higher temperature improves efficiency by reducing the need for energy-intensive cooling equipment.

Explaining the design, Nvidia said, “Hot tubs sit at about 38 to 40 degrees Celsius, warm enough that most people can only soak for about 15 minutes. NVIDIA’s newest AI servers can run their cooling liquid even hotter — up to 45 degrees Celsius, or 113 degrees Fahrenheit.”

Nvidia says the DSX AI factory reference design is built around a fully closed-loop liquid cooling system that continuously recycles coolant without requiring fresh water input during normal operation. In favourable conditions, the company claims this could reduce water use to nearly zero, especially in regions where dry coolers can handle heat rejection for most of the year.

“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,” said Ali Heydari, director of data centre cooling and infrastructure at Nvidia. He added that only in limited cases, depending on climate conditions, would additional cooling systems be required for short periods.

Nvidia also points out that cooling has historically accounted for a large share of data centre electricity use, sometimes as much as 40 per cent. By moving to liquid cooling and allowing higher operating temperatures, the company says operators can reduce both energy consumption and long-term operating costs.

Amazon’s billions-of-gallons water use shows today’s reality

While Nvidia is outlining a near water-free future for AI infrastructure, current cloud operations still rely heavily on water. Amazon recently reported that its global data centres used 2.5 billion gallons, or about 9.46 billion litres, of water in a single year.

That volume is roughly equal to around 5 percent of the annual water consumption of metro Seattle, where Amazon is based. The company said its figures show efficiency improvements but also vary widely depending on climate, cooling method, and location of facilities.

AWS vice president Kerry Person pushed back on concerns around industry-wide water consumption, saying, “If you look at the press right now, the data center industry is apparently consuming all of the water in the world. When you actually look at the data and look at the details, nothing could be further from the truth.”

- Ends