NVIDIA Data Center Hardware Is Being Cooled With Water 'Hotter Than A Hot Tub'
NVIDIA says its newest AI servers can run on 45C liquid, cutting data center energy, water use, and noise. The company frames the shift as a more efficient alternative to air cooling.
Intelligence analysis by GPT-5.4 Mini
The article argues that data centers have been wasting enormous amounts of energy by overchilling air, when hotter liquid can still pull heat away from chips efficiently. NVIDIA’s Rubin systems are presented as a cleaner, quieter, and more space-efficient model for future AI infrastructure.
NVIDIA is trying to cool computers with warm water instead of icy air. It is like using a bathtub that is not cold at all to pull heat away from a phone charger that is getting too hot.
Analysis
Hotter Coolant, Cooler Operations
NVIDIA’s pitch is intentionally counterintuitive: if the coolant is warm enough, outdoor dry coolers can dump heat without relying on chillers. That matters because the traditional data center model spends a lot of power turning rooms into near-freezers just so the hardware stays within safe limits.
The article frames this as a systems-level rethink rather than a small efficiency tweak. Instead of fighting the heat with more refrigeration, the design lets the heat move through a closed liquid loop and out of the building more directly.
The Real Bill Is Not Just Electricity
The most striking claim in the piece is that cooling can account for as much as 40 percent of a data center’s electricity bill. NVIDIA’s cited estimate that every one-degree increase in chiller temperature can reduce cooling energy by about 4 percent helps explain why the company is emphasizing thermal strategy as much as chip performance.
Water use is the other major pressure point. The article says conventional cooling towers can consume around 2.6 million gallons per megawatt per year, while NVIDIA’s architecture can bring that figure near zero by avoiding evaporative cooling and water towers altogether.
Why Rubin Could Influence The Rest Of The Industry
The story is not claiming that every facility can adopt this model immediately. Geography still matters, and the article notes that the same temperature expectations do not make sense everywhere, especially in hotter climates.
Even so, the broader implication is clear: if liquid cooling at higher temperatures works at scale, it could make future data centers quieter, denser, and cheaper to run. The article also hints that this does not fix the huge installed base of existing facilities, but it could set a new baseline for the next generation of AI infrastructure.
Key points
- NVIDIA says its Rubin AI servers use 45C liquid cooling, which is hotter than a hot tub but still effective.
- The company argues that warmer coolant can reduce the need for energy-hungry chillers.
- The article says cooling can consume up to 40 percent of a data center’s electricity bill.
- NVIDIA claims the new design can sharply cut water use by avoiding evaporative cooling towers.
- The architecture is also described as quieter and more space-efficient than older server setups.
If the approach works broadly, data centers could use less electricity and far less water while staying just as powerful. The article also suggests quieter rooms and smaller server racks, which could make future AI facilities easier to build and operate.
The design may not translate well to every climate, since the article says temperature management expectations differ by geography. It also does nothing to improve older data centers already built around air cooling, so the biggest gains may take time to reach the wider industry.
