NVIDIA's Rubin AI platform ships with 100% liquid cooling across every chip and networking component—a full departure from the hybrid setups that have dominated data center thermal management for the past decade.

Published on 22 June 2026, NVIDIA's specifications confirm that Rubin operates at a 45°C warm-water cooling temperature. To understand why that matters: older liquid cooling systems required much colder water (typically 18–25°C), which meant data centers had to run expensive mechanical chilling equipment year-round. Rubin's 45°C threshold lets facilities reject heat into cooling towers or dry coolers at ambient temperatures that previously made chilling mandatory, cutting both upfront capital costs and the ongoing energy burden of refrigeration machinery.

From Hybrid to Homogeneous

Earlier GPU generations—Hopper, Blackwell—used liquid cooling selectively. The main processor dies and some memory got cooled with liquid, but supporting components like networking chips, power distribution boards, and optical transceivers stayed on air. That created a thermal bottleneck: even if the compute heart ran cool, the air-cooled periphery limited how densely operators could pack racks.

Rubin closes that gap entirely. NVIDIA's blog post confirms the platform extends liquid loops to every chip and every networking component in the rack, eliminating the mixed-mode thermal planning that has constrained hyperscale deployments.

The progression from Blackwell is concrete. NVIDIA's Blackwell platform, detailed in an April 2025 post, achieved a 300x improvement in water efficiency compared with earlier GPU generations and delivered a 12% increase in compute throughput while reducing energy consumption through its liquid cooling design. Rubin takes those gains as a baseline.

The Water Efficiency Picture

Here is a distinction that matters. NVIDIA's cooling architecture reduces the water circulating through on-premises liquid loops—the coolant flowing through pipes and manifolds inside the building. That is different from the total water footprint of an AI facility. As TechCrunch noted on 22 June 2026, the broader water problem also involves upstream consumption at power plants that generate electricity—a factor determined by where the power comes from and local geography, not by chip-level cooling design. Rubin's architecture handles the on-site side. The power-generation side remains a grid and location problem.

This is not a shortcoming of the engineering. It is a clarification for anyone scoping sustainability claims around these specs.

Production Status and Timing

Rubin entered production at CES 2026 in January, where Data Center Frontier reported the 45°C warm-water capability alongside the production announcement. The June 2026 blog post from NVIDIA fills in the full technical picture and confirms the all-liquid scope that January coverage had only partially covered. For infrastructure teams planning Rubin deployments, the June documentation is the authoritative reference.

What This Changes for Operators

At 45°C supply temperature, Rubin-compatible facilities can often operate without mechanical chillers in temperate climates, running in economizer-only mode for much of the year. That changes power usage calculations (measured as PUE, or Power Usage Effectiveness), cooling tower sizing, and—for colocation providers selling AI-optimized space—the cost of retrofitting older buildings.

The all-liquid mandate also reshapes the deployment conversation. Operators who have deferred liquid cooling infrastructure investments while running hybrid configurations now lose the middle path with Rubin. The platform requires full cooling distribution units and manifold infrastructure from day one. That is a harder sell for enterprise buyers with existing air-cooled buildings, but a cleaner fit for new hyperscale facilities designed around liquid from the ground up.

The move from selective liquid coverage on Blackwell to total elimination of in-rack air cooling on Rubin, paired with warm-water operation that removes chiller dependencies, suggests NVIDIA is treating thermal architecture as a core design constraint rather than a bolt-on solution. Given where GPU power consumption is heading, that makes practical sense. The open question for the industry is whether facility standards bodies and colocation providers can redesign their infrastructure requirements to match what the silicon roadmap now demands.