Nvidia's Rubin AI servers use liquid cooling to cut data‑center water use to near zero

Nvidia unveiled its next‑generation Rubin AI servers, which rely on a closed‑loop liquid‑cooling system that circulates a 75% water, 25% glycol mixture directly through cold plates mounted on CPUs and GPUs. The coolant operates at temperatures up to 45 °C, allowing dry coolers to reject heat without the need for traditional evaporative towers. This design removes all fans, virtually eliminates water evaporation and can lower facility‑level water consumption to near zero in suitable climates.

Independent testing shows the water‑cooled H100 GPUs deliver roughly 17% higher performance than air‑cooled equivalents, while the higher‑temperature coolant improves power usage effectiveness and enables far greater rack density—two Rubin racks can replace six conventional ones. Nvidia highlighted these efficiency gains during London Climate Week and noted that manufacturing partners such as Supermicro are already preparing production. The technology aims to reduce both water and electricity use in AI data centers, offering a potential procurement advantage as the industry seeks lower total cost of ownership.