Nvidia Develops Breakthrough Cooling Tech for AI Machines

Nvidia's Vera Rubin AI servers achieve 100% liquid cooling at 45°C supply temperatures, eliminating the need for mechanical water chillers in most climates. Announced at CES 2026, the breakthrough cuts data center capex and water usage; HVAC stocks sold off 11-21% on the news. Vera Rubin available to customers H2 2026.

NVDA Nvidia's Vera Rubin AI server platform achieves 100% liquid cooling — every chip and every networking component cooled entirely by liquid in a closed loop, with zero fans in the system. The key technical breakthrough is a 45°C (113°F) coolant supply temperature, using a water-propylene glycol mix flowing through cold plates directly on processors. Jensen Huang framed it at CES 2026: "No water chillers are necessary. We are cooling this supercomputer with hot water."

The innovation matters because at 45°C, most climates can reject heat using ambient air alone, eliminating the need for mechanical water chillers. Prior Blackwell systems required 35°C supply temperatures, which still needed energy-intensive chillers. The shift to chillerless cooling cuts both capital cost (no chiller hardware) and ongoing water consumption — addressing the "AI water challenge" that has drawn scrutiny from municipalities and regulators as data center density scales. The announcement triggered immediate market reaction: HVAC stocks sold off sharply, with Modine Manufacturing falling ~21% and Trane Technologies ~11% on the day of the reveal.

Vera Rubin chips are available to customers beginning H2 2026, positioning NVDA to lock in major hyperscaler orders before the next product cycle. The structural implication for AI infrastructure capex is significant: operators can now deploy higher-density racks without the cooling plant limitations that currently constrain many data center facilities, potentially compressing the timeline for scaling from hundreds to thousands of megawatts of AI compute per campus.