Growing opposition to data centers has increasingly focused on how much electricity and water they consume. Nvidia is now promoting its Rubin-generation reference design for a fully liquid-cooled data center, saying it can cut “massive amounts” of power use while reducing water consumption to almost nothing. The claim, however, does not resolve every concern tied to large AI facilities, including the environmental impact of construction and the major power generation needed to run them. As Gizmodo noted, Nvidia’s blog post also does not detail how the cost of building this kind of liquid-cooled data center compares with a less efficient air-cooled facility, though the company says “every cloud provider and data center operator building for [Rubin] is making the transition.”
Part of the efficiency improvement comes from allowing AI servers to operate at higher temperatures, reaching up to 113 degrees Fahrenheit, or 45 degrees Celsius. Amazon recently made a similar point in a report, saying higher heat tolerance is helping improve efficiency in its data centers, which are largely cooled by air.
In Nvidia’s approach, heat is pulled directly from the chip and moved through liquid loops that run at much higher temperatures. That setup allows outdoor dry coolers to release heat more efficiently throughout much of the year, while offering greater flexibility across different ambient air conditions.
Josh Parker, Nvidia’s head of sustainability, said the reference design can reduce water use from about 2.6 million gallons per megawatt each year in traditional cooling-tower-based systems to nearly zero, describing it as a potential reduction of up to 100 percent.
