Nvidia unveils boxed AI factories it says go live in 14 days

Nvidia wants AI data centers to arrive built and tested in modular containers. The chipmaker on Wednesday introduced DSX POD, a reference design for complete Vera Rubin AI factories that partners build inside modular containers. The units are integrated and tested before delivery. Nvidia says they cut deployment at a prepared site from six months to as little as 14 days.

The pitch is about time. A data center earns nothing while it is under construction. Nvidia says moving the build into a factory lets customers start running models, and earning revenue, sooner. The company announced the design at its GTC Berlin event.

What’s in the box: DSX POD moves the integration of computing, networking, cabling, rack-level power and liquid cooling to the manufacturing plant, according to the release. Partners finish the rack-to-rack cabling, wiring, piping and functional checks before shipment, Nvidia says. The units then connect to a site’s power, cooling and network.

Parallel work: That factory work happens while the customer prepares the site. Nvidia says this cuts the planning, construction, specialized labor and logistics of a traditional onsite build. It also says factory-tested containers reduce exposure to component delays and error-prone field work.

The partners: Nvidia lists 17 companies working on the units. They include server makers Dell Technologies, HPE and Lenovo, power and cooling suppliers Schneider Electric, Vertiv and Eaton, and Bull. Others are Anduril, Armada, Crusoe, DayOne, Delta, Foxconn’s Ingrasys, Gigabyte, Pegatron, Quanta Cloud Technology and Wistron.

The pitch: “An AI factory creates value when it starts producing tokens — not when construction is complete,” Nvidia CEO Jensen Huang said in the release. He said the design brings “factory-built precision and speed” to AI construction.

What Nvidia didn’t say: The release gives no price, no size for a single unit and no power rating. It names no customers. The 14-day figure applies only to a site that already has power, cooling and networking in place.

Our read: Nvidia is selling speed of construction, not a faster chip, and the claim deserves care. The 14 days counts from delivery to a prepared site. It does not include winning a grid connection, permits or cooling supply, which the release leaves to the customer. Nvidia gives no figure for those steps. So the design can shorten one stage of a build without shortening the whole. The partner list matters more than the headline number. With Dell, HPE, Lenovo, Schneider Electric and Vertiv on board, the design spans most of the gear in a modern AI hall. If the partners deliver, that could make Vera Rubin capacity easier to order as a standard product. The Berlin launch and the release’s Europe headline point to buyers there, but Nvidia gives no regional delivery plan. Until a customer runs one of these units on Nvidia’s timeline, the 14 days is a target, not a track record.

What’s next: Nvidia says partners will deliver DSX POD AI factories worldwide in early 2027.

Sources: Nvidia news release