Nvidia has unveiled its latest innovation for AI data centers: a groundbreaking liquid cooling system for its Rubin AI servers. This new approach, revealed on June 23, 2026, allows servers to operate efficiently with cooling liquid as warm as 45 degrees Celsius, signaling a significant shift towards greater efficiency and sustainability in data center operations.
Revolutionizing Data Center Cooling
The Rubin AI infrastructure marks Nvidia's first generation to incorporate 100% liquid cooling. Unlike traditional setups, every chip and networking component within these servers is cooled in a closed-loop system, entirely eliminating the need for fans. This design, detailed in Nvidia's DSX AI factory reference architecture, promises substantial reductions in power and water usage, particularly at hyperscale.
Ali Heydari, Nvidia's director of data center cooling and infrastructure, emphasized the system's environmental benefits. "The NVIDIA DSX reference design for AI factories has zero water consumption — we have eliminated massive amounts of power usage and pretty much all water usage," Heydari stated, though he noted that some climates might still require chillers for a minimal portion of the year.
Significant Cost and Resource Savings
Cooling has historically been a major energy drain, accounting for up to 40% of a data center's total electricity consumption. Industry data suggests that even a one-degree Celsius increase in chiller plant temperature can reduce cooling energy costs by approximately 4%. Nvidia projects that a 50-megawatt hyperscale facility could realize savings exceeding $4 million annually in combined energy and water expenses.
In regions with favorable climates, this 45-degree cooling system can nearly eliminate the need for cooling water, reducing consumption from roughly 2.6 million gallons per megawatt per year—typical for conventional cooling tower systems—to almost zero.
Technical Details and Design Advancements
Traditional data centers rely on complex air-cooled systems, featuring hot and cold aisles and generating fan noise upwards of 85 decibels. The Rubin systems, by contrast, employ a coolant mixture of 75% water and 25% propylene glycol. This mix circulates through cold plates attached to processors, entering at 45 degrees Celsius and exiting at around 55 degrees after absorbing heat, all while processors maintain peak performance.
Richard Whitmore, president and CEO of Motivair, Schneider Electric’s advanced cooling division, highlighted the necessity of such advancements. "Once the watts per chip crossed a certain level, liquid cooling became mandatory," Whitmore commented, underscoring the increasing power density of modern AI hardware.
The shift to full liquid cooling necessitated a complete redesign of components previously air-cooled. Nvidia's new tray-level architecture features a single inlet and outlet, sealed front panels, and increased rack density, allowing systems that once occupied six rack units to now fit into just two. This compact design, combined with the ability to use outdoor dry coolers without refrigeration equipment in suitable locations, offers unprecedented flexibility for data center operators. Cloud providers and data center builders adopting the Rubin platform are already transitioning to this new standard.
Looking ahead, Nvidia also envisions the potential for waste heat recovery from these highly efficient data centers, which could be repurposed to heat nearby buildings as AI workloads and data center construction continue their rapid expansion.