Nvidia on Oct. 21 unveiled a reference design for factory-assembled Vera Rubin AI systems, saying operators at prepared data center sites could begin serving workloads in as little as 14 days.
The company compares its target with roughly six months for a conventional deployment. The 14-day figure is a projection, not a reported result from an operating site, and depends on supporting infrastructure being ready before the equipment arrives.
Under the DSX POD design, partners would assemble and test racks inside containers before shipment. That factory work includes internal power wiring and liquid-cooling equipment. Once the containers reach a site, crews would still need to connect them to power, cooling and networks, then complete final testing and commissioning.
Nvidia says moving more installation work into factories could make deployment more predictable and allow operators to serve customers sooner. Its technical blog notes that prefabricated data centers already exist. The proposed change is to assemble and test the interconnected equipment used for AI workloads before delivery.
DSX POD is a reference design for partners to build and deliver, rather than a system Nvidia says is already in service. The company named Dell Technologies, Foxconn (Ingrasys), HPE and Lenovo among the partners involved. Nvidia expects partner deliveries worldwide to begin in early 2027.
