Nvidia's Rubin design aims to slash AI data center water use by running servers hotter

Nvidia has unveiled a reference design for its next-generation Rubin AI data centers that it says can nearly eliminate water consumption by running servers at higher temperatures. The fully liquid-cooled approach captures heat directly at the chip and uses liquid loops operating at up to 113°F (45°C), allowing outdoor dry coolers to reject heat without evaporative water loss, according to briefly.co.

"By capturing heat directly at the chip and using liquid loops at higher temperatures, the system enables outdoor dry coolers to reject heat efficiently, reducing water usage from roughly 2.6 million gallons per megawatt per year to near zero."

That represents a dramatic reduction from the roughly 2.6 million gallons of water per megawatt per year that conventional data centers consume. The Verge reported that Nvidia claims the design also significantly cuts power usage, though the company has not provided detailed cost comparisons with less efficient air cooling systems.

The VergeNvidia claims liquid-cooled Rubin data center design cuts energy use and nearly eliminates water consumption

However, both outlets noted that the design does not address broader environmental concerns. Briefly.co reported that the Rubin blueprint does not account for construction impacts, power generation requirements, or cost comparisons with air cooling. The Verge added that critics point to omissions in Nvidia's blog post about the technology's limitations.

"This doesn't address broader concerns about AI data centers, including environmental impacts during construction and the massive power generation requirements."

As AI workloads continue to grow, the energy and water demands of data centers have drawn increasing scrutiny from regulators and environmental groups. Nvidia's Rubin design offers a potential path to reducing one of those impacts, but the full environmental cost of powering and building these facilities remains unresolved.