Nvidia says its AI data center design runs hotter to use a lot less water
TL;DR
Nvidia is promoting a Rubin-generation reference design for AI data centers using 100 percent liquid cooling and higher operating temperatures. The servers can reportedly run as hot as 113 degrees Fahrenheit or 45 degrees Celsius, with heat captured at the chip and moved through liquid loops to dry coolers. Nvidia claims water use can fall from about 2.6 million gallons per megawatt per year in conventional cooling-tower systems to near zero.
Nauti's Take
This is a useful technical move, not a free pass for ever-larger AI factories. Nvidia is addressing a real acceptance problem: lower water use is much easier to defend publicly than more efficient megawatt machines.
The deciding factor will be whether operators publish hard numbers on total cost, energy profile, and local impact. Without that, this remains a strong infrastructure pitch with a green edge, not proof of sustainable AI data centers.
Briefingshow
Water has become one of the most visible flashpoints around AI infrastructure, especially in resource-constrained regions. If liquid cooling with dry coolers scales as claimed, it could reduce one local source of resistance. But it pushes the harder debate toward power generation, grid capacity, construction impact, and the real cost of this architecture.
Sources