NVIDIA Rubin 100% liquid-cooled, eliminates need for data center chillers

NVIDIA's Vera Rubin platform is the first AI infrastructure to achieve 100% liquid cooling, operating at 45°C (113°F) supply temperature instead of the previous 35°C standard. The higher temperature allows data centers to eliminate energy-intensive mechanical chillers and use dry coolers with ambient air instead, cutting cooling energy consumption by an estimated 4% for each degree increase in facility temperature.

At hyperscale (a 50-megawatt facility), the shift to 45°C liquid cooling saves over $4 million annually in cooling-related energy and water costs. In favorable climates, NVIDIA's design enables chiller-less operation with zero water consumption — reducing facility water use from roughly 2.6 million gallons per megawatt per year to near-zero. The coolant (75% water, 25% propylene glycol) enters cold plates at 45°C and exits around 55°C after absorbing heat, maintaining full chip performance within validated operating limits.

Rubin achieves this efficiency without sacrificing compute density: the platform delivers twice the power of Grace Blackwell while maintaining the same airflow rates and water temperature. The new architecture eliminates 43 cables and six water pipes per chassis, reducing assembly time from 2 hours to 5 minutes. Every cloud provider and data center operator building for Rubin is now making the transition to liquid-cooled infrastructure.

For operators planning AI infrastructure, the 45°C standard means a fundamental shift in facility design — moving from legacy air-cooled racks to purpose-built AI halls with liquid distribution units (CDUs) and custom cooling loops. Hyperscalers are already using digital twins to model chip-to-campus thermals and free-cooling hours before deployment. This becomes a key cost lever as capex in cooling infrastructure directly impacts the economics of multi-billion-dollar data center buildouts.