Tech

Nvidia claims Rubin generation liquid cooling cuts AI data centre water use to near zero

Nvidia asserts that its fully liquid-cooled architecture for the upcoming Rubin hardware can reduce water consumption by up to 100 per cent compared to traditional systems, a move it says all cloud providers are adopting.

Author
Owen Mercer
Markets and Finance Editor
Published
Draft
Source: The Verge · original
Nvidia says its AI data center design runs hotter to use a lot less water
The chipmaker’s new reference design operates servers at higher temperatures to eliminate reliance on conventional cooling towers, though critics note omissions regarding construction and power costs.

Nvidia has unveiled its Rubin generation reference design for artificial intelligence data centres, claiming the fully liquid-cooled architecture can reduce water consumption to near zero. The company states that by operating servers at higher temperatures, up to 45 degrees Celsius, the system eliminates the need for conventional cooling towers, which typically consume approximately 2.6 million gallons of water per megawatt per year.

According to Josh Parker, Nvidia’s head of sustainability, the design captures heat directly at the chip and transports it via liquid loops operating at these elevated temperatures. This process allows outdoor dry coolers to reject heat efficiently for much of the year, offering greater flexibility regarding ambient air temperature. Parker claims this approach achieves up to a 100 per cent reduction in water use compared to legacy systems.

The announcement comes amid growing public scrutiny over the substantial water and energy demands of AI infrastructure. Nvidia asserts that every cloud provider and data centre operator building for the Rubin generation is transitioning to this liquid-cooled model. The company highlights that the design has eliminated massive amounts of power usage alongside the significant reduction in water consumption.

However, the announcement has drawn criticism for omitting key contextual data. As noted by Gizmodo, Nvidia’s disclosure does not address the environmental impact of data centre construction, the power generation requirements for these facilities, or the comparative costs of building liquid-cooled infrastructure versus less efficient air-cooled alternatives.

Industry peers are also exploring efficiency gains through thermal management. Amazon has recently reported on increasing heat tolerances to improve efficiency in its mostly air-cooled data centres, reflecting a broader sector trend toward optimising thermal performance, albeit through different technical approaches.

Continue reading

More from Tech

Read next: Tech giants slash 21,000 jobs at Oracle as AI reshapes workforce strategy
Read next: OpenAI partners with Trail of Bits to bolster open-source security
Read next: Unsloth enables local deployment of Z.ai’s GLM-5.2 with aggressive quantisation