Nvidia introduced DSX POD at GTC Berlin on Oct. 21, 2026, outlining modular AI factories that partners would integrate and validate before shipment. The company plans worldwide partner deployments in early 2027, shifting substantial installation work from customer sites into manufacturing facilities.
The company announcement describes container-based reference designs built around its Vera Rubin systems. Partners would complete work including connections between racks, electrical installation, piping and functional validation before delivery.
A prepared site remains essential
Nvidia says units would connect to power, cooling, networking and other supporting infrastructure at a prepared location. Its claim that deployment could fall from six months to as little as 14 days should not be read as a timetable for acquiring land, securing electricity and building an entire facility from scratch.
The company names Dell Technologies, HPE, Lenovo, Schneider Electric and Vertiv among the partners working on the rollout. The announcement describes a deployment plan, not completed customer installations.
Infrastructure becomes part of the product
In a separate Berlin release, Nvidia described the broader DSX platform as coordinating computing, networking, storage, software, power and cooling.
The operational consequence is a change in where integration happens. Buyers evaluating the modular approach still need to establish responsibility for site readiness, delivery and acceptance testing. Faster factory integration cannot by itself establish that a particular location has the infrastructure needed to run the equipment.