In an official blog post on June 21, NVIDIA detailed the comprehensive liquid cooling technology used in the Vera Rubin platform. Rubin is the world’s first AI compute infrastructure platform to achieve 100% liquid cooling — every chip and every network component in the system is cooled by liquid, with no fans at all. The coolant is a mixture of 75% water and 25% propylene glycol, operating at temperatures up to 45°C (above the upper limit of a typical hot tub, which ranges from 38 to 40°C). This counterintuitive high temperature is the key to efficiency gains: for every 1°C increase in temperature, cooling energy consumption drops by about 4%. For a 50-megawatt hyperscale data center, migrating to a liquid cooling architecture can save more than $4 million annually in cooling-related energy and water costs. In suitable climates, this architecture can operate without chillers, reducing water consumption from roughly 2.6 million gallons per megawatt per year for traditional cooling towers to nearly zero. Ali Heydari, Director of Data Center Cooling at NVIDIA, stated that the DSX AI factory reference design has already eliminated a significant amount of power consumption and nearly all water usage.
At the structural level, the new architecture captures heat directly at the chip surface and conducts it via liquid loops to outdoor dry coolers, eliminating the need for hot-aisle/cold-isle air management. Previous liquid-cooled servers used a hybrid architecture, where only the GPU/CPU were cooled by cold plates and the remaining components still relied on air cooling. With Rubin, the cooling approach for all components has been redesigned: the server front panel has switched from a ventilated grille to a sealed panel, rack density has increased, systems that once occupied 6U of space have been compressed into 2U, and noise contributions have been reduced to extremely low levels. Richard Whitmore, President of Motivair (a Schneider Electric company), noted that as single-chip power consumption crosses the critical threshold that air cooling can handle, “liquid cooling is no longer an option — it’s a necessity,” and the entire ecosystem is now fully adapting. Citing an analysis by Essence Securities, the STAR Market Daily reported that the bottleneck for AI computing power expansion has shifted from chip supply to power and thermal management supply, and the main investment theme for AIDC is evolving from “increased volume of server components” to “systemic restructuring of power infrastructure.”