Nvidia unveils DSX POD modular AI factory built around Vera Rubin
Nvidia unveiled DSX POD, a modular AI data center using Vera Rubin, claiming 14-day deployment and lower resource use.
The containers are assembled and tested in manufacturing facilities before shipment, with internal cabling, power connections, cooling pipes and networking installed. At deployment, operators connect each unit to prepared infrastructure at their chosen location. Nvidia CEO Jensen Huang linked faster deployment to earlier revenue from AI infrastructure. “An AI factory creates value when it starts producing tokens — not when construction is complete,” he said. “With NVIDIA DSX POD, we’re bringing factory-built precision and speed to AI infrastructure construction…so customers can scale capacity and bring revenue-generating production online faster.”
The reference design starts with Vera Rubin NVL72 systems. Huang described each system as comprising seven liquid-cooled computing racks connected through about 17 kilometers of Spectrum-X cabling. Nvidia says reducing installation time can allow operators to begin training models and running production workloads earlier. The 14-day estimate depends on site readiness: power availability, cooling capacity, networking and other supporting systems must already be prepared.
Nvidia also claims resource-efficiency gains, though it says savings depend on facility design and operating conditions. Its cooling approach uses closed-loop liquid cooling with outdoor dry coolers to release heat, and Nvidia says it can eliminate water evaporation during normal heat rejection. Nvidia estimates that a 2-gigawatt deployment could save 54 million liters of water a day compared with conventional systems relying on cooling towers. Extreme temperatures may require additional mechanical cooling, potentially reducing water savings depending on local climate and operating conditions.
Nvidia claims its DSX MaxLPS technology can deliver equivalent token output using 30% less power through dynamic allocation within an AI facility's available electricity budget. Another component, DSX Flex, is designed to adjust data center power consumption in response to electricity prices, grid conditions and utility requirements.
Nvidia is collaborating with Dell Technologies, HPE, Lenovo, Schneider Electric and Vertiv to manufacture and supply DSX POD systems worldwide. Deliveries are expected to begin in early 2027. Nvidia's projected efficiency gains still require independent verification under real-world operating conditions, TechRadar reported.
Editor's Summary
Nvidia's DSX POD packages Vera Rubin computing, networking, power and liquid cooling into factory-built containers, with a claimed deployment time of 14 days if sites are ready. Nvidia also claims lower water and power use, but those figures depend on operating conditions and need independent verification. Deliveries are expected in early 2027 through partners including Dell, HPE, Lenovo, Schneider Electric and Vertiv.