Nvidia says DSX POD cuts deployment to 14 days

Nvidia introduced DSX POD, a containerized version of its DSX reference design for AI data centers powered by its Vera Rubin platform, and said the units can reduce deployment time from six months to 14 days. Details were released at the GTC Berlin event this week; Nvidia first unveiled the broader DSX AI Factory Platform at GTC San Jose earlier this year.

Seventeen partners are due to deliver DSX POD facilities worldwide in early 2027: Anduril, Armada, Bull, Crusoe, DayOne, Dell Technologies, Delta, Eaton, Foxconn, Gigabyte, HPE, Lenovo, Pegatron, Quanta Cloud Technology, Schneider Electric, Vertiv and Wistron. The units ship as containers already integrated and validated, and connect to a prepared site's power, cooling, networking and other supporting infrastructure.

Deployment time is the gap between committed capital and a facility earning revenue, so the six-months-to-14-days claim is the number allocators financing data center capacity will test first. The partner roster is the other detail worth reading: it shows which equipment and integration firms sit in the delivery chain, though the coverage does not break out each one's contribution. Nvidia's own caveat — that certain climates may require additional mechanical cooling during extreme heat — is a siting question for any facility pitched around dry coolers.

On water, Nvidia says the DSX platform pairs closed-loop liquid cooling with outdoor dry coolers that release absorbed heat into the air without evaporating water. At 2GW of scale, the company claims the approach saves 54 million liters of water a day compared with conventional cooling-tower-based systems.

The same launch added two power products. Nvidia says DSX MaxLPS dynamically allocates power within a facility's provisioned budget, producing the same token output with 30 percent less power, and that DSX Flex will let data centers adapt workload consumption to meet utility needs. Ian Buck, speaking to media ahead of the announcement, said that when there is an energy spike and "the grid needs an extra gigawatt, our data centers can be a resource to respond to that need."

Every efficiency figure here — the claimed up-to-100 percent cut in cooling water, 54 million liters a day, 30 percent less power — is Nvidia's, and the coverage reports no independent measurement. The questions that remain for an allocator underwriting a data center or a power thesis are site-level: whether a given climate supports dry coolers without supplemental mechanical cooling, and whether grid-flexibility terms can be contracted firmly enough to carry weight in a power thesis.