Nvidia on Tuesday unveiled DSX POD, a reference design aimed at sharply reducing the time needed to build AI data centres, at GTC Berlin in Berlin, Germany.

DSX POD has partners complete and supply an Nvidia Vera Rubin-based AI factory inside a modular container. Integration and installation of computing, networking, cabling, rack-level power and liquid cooling are completed at a manufacturing plant before shipment. Customers only need to connect the container at a site equipped with power, cooling and network. Nvidia said this can cut on-site build time from 6 months to as little as 14 days.

Nvidia said containers pre-tested at the factory reduce the risk of parts delays and on-site construction errors. Partners finish rack-to-rack cabling, wiring, piping and functional verification before shipment. The work proceeds separately from site preparation, further shortening the overall schedule. Nvidia defined the point when a new AI factory starts running models for actual services as “time to first token,” and said faster timing brings forward revenue generation.

Nvidia founder and CEO Jensen Huang (젠슨 황) said, “AI factories create value not when construction is finished, but when they start producing tokens.” He added, “With DSX POD, we will bring factory-level precision and speed to AI infrastructure construction and reduce deployment time from months to weeks.”

Nvidia plans to supply DSX POD AI factories worldwide from early 2027 in cooperation with Anduril, Armada, Bull, Crusoe, DayOne, Dell Technologies, Delta, Eaton, Foxconn (Ingrasys), Gigabyte, HPE, Lenovo, Pegatron, Quanta Cloud Technology (QCT), Schneider Electric, Vertiv and Wistron.