On October 21, Nvidia announced DSX POD at its GTC Berlin event — a reference design for a complete AI data center built inside a modular, shipping-container-sized unit at a partner factory, then delivered ready to connect. The company says the approach can reduce on-site deployment time at a prepared site from six months to as little as 14 days.
Each DSX POD contains seven Vera Rubin server racks along with factory-installed cabling and liquid cooling. Nvidia will not manufacture the units itself; instead, it has named build partners including Dell Technologies, HPE, Lenovo, Foxconn, Gigabyte, Quanta, and Pegatron. Schneider Electric and Vertiv are among the power and cooling suppliers. First deliveries are expected in early 2027. Nvidia says the dry-cooler design can eliminate cooling-water use by up to 100 percent compared with conventional cooling-tower approaches, and that its DSX MaxLPS software can deliver the same token output with up to 30 percent less power.
Two caveats worth noting for planning purposes. The 14-day figure applies only to sites where power, networking, and cooling infrastructure are already in place — site preparation is not included. And the DSX POD is positioned as a way to run AI workloads while permanent facilities are still being built, not as a permanent replacement for traditional data centers.
Organizations evaluating AI compute capacity or planning data center expansions may want to track partner pricing and availability as the early-2027 delivery window approaches. Facilities managers should note that while water consumption may drop, the dry-cooler design still requires adequate outdoor heat-rejection capacity and electrical infrastructure before a unit can go live.
Does this affect your business?
Our advisors can help you figure out the right move, at no cost.