NVIDIA has introduced the Rubin generation of AI infrastructure, establishing a new industry benchmark with its first fully liquid-cooled server architecture. Operating with coolant temperatures up to 45 degrees Celsius, the platform eliminates traditional air-cooling dependencies and introduces the NVIDIA DSX AI factory reference design to standardize deployment practices. This architecture captures thermal loads directly at the processor and networking components using closed-loop liquid cold plates, routing elevated-temperature coolant to outdoor dry coolers in regions where ambient conditions can efficiently reject heat without mechanical chillers. The transition addresses the unsustainable power trajectories of modern AI workloads. Cooling infrastructure has historically consumed up to 40 percent of a data center’s electricity. By enabling higher operating coolant temperatures and removing evaporative water systems, the closed-loop methodology aligns with engineering principles that raise chiller plant temperatures to slash cooling energy consumption. For a 50-megawatt hyperscale facility, the shift to fully liquid-cooled infrastructure yields annual savings exceeding four million dollars in energy and water expenses. Water consumption drops to near zero, as the system continuously recirculates a 75 percent water and 25 percent propylene glycol mixture without requiring makeup water. The comprehensive thermal redesign delivers substantial spatial and operational advantages. By cooling peripheral circuitry that previously relied on finned heat sinks, Rubin servers eliminate internal fans, removing the need for cold aisle containment and drastically reducing acoustic output. The consolidated cooling architecture allows compute modules to occupy two rack units instead of six, enabling higher density deployments and streamlined facility footprints. Ali Heydari, NVIDIA’s director of data center cooling and infrastructure, noted that the reference design achieves zero water consumption in standard deployments, relying exclusively on dry coolers except during extreme climatic events. Industry executives view the platform as an unavoidable evolution as silicon power densities surpass the physical limits of air cooling. Richard Whitmore, chief executive of Schneider Electric’s Motivair division, emphasized that liquid cooling has become mandatory for next-generation AI hardware. Hardware and cooling infrastructure providers are now aligning supply chains to support the thermal paradigm. The architecture also enables waste heat recovery, allowing residual thermal energy to be redirected for municipal or commercial heating, further improving the sustainability profile of AI operations. As computational demand continues to outpace facility construction, NVIDIA’s 45-degree liquid cooling standard redefines thermal efficiency for hyperscale environments. By decoupling server performance from ambient air temperatures and eliminating mechanical refrigeration dependencies in favorable geographies, the industry gains a scalable pathway to sustain AI growth without proportional increases in energy consumption or water usage.