Nvidia claims its Rubin generation reference design for fully liquid-cooled data centers eliminates what it calls 'massive amounts of power usage and pretty much all water usage,' a direct response to growing public opposition over AI infrastructure's resource consumption.
The design works by running hotter and routing liquid cooling directly to components, cutting the evaporative water loss that traditional air-cooled facilities require at scale. But Nvidia's own blog post omits construction impact, upstream power generation requirements, and, as Gizmodo flagged, any cost comparison between liquid-cooled and conventional air-cooled builds.
Read the original for the specific efficiency claims Nvidia is making for 'every cloud' deployment, and to judge whether this is a genuine infrastructure shift or a targeted reframe of the data center criticism problem.
[READ ORIGINAL →]