NVIDIA DSX POD: AI factory setup in as little as 14 days

NVIDIA said on October 21 that a Vera Rubin AI factory can come online at a prepared site in as little as 14 days, down from six months, if the systems arrive already assembled and tested inside a container.

The design behind that figure is DSX POD, a reference design NVIDIA announced at GTC Berlin. Partners build the units to it: they put complete Vera Rubin AI factories inside modular containers, then integrate and validate them before delivery. The units are meant to connect out of the box to a site’s power, cooling and networking.

The big change is where the work happens. In a traditional build, compute, networking, cabling, rack-level power and liquid cooling are all integrated and installed onsite, which takes a lot of planning, construction, specialized labor and logistics. DSX POD moves that integration to the factory. Partners finish rack-to-rack cabling, wiring, piping and functional validation before the container ships. The DSX POD product page names the problem it targets: large AI factories can be slowed by integration complexity and by the onsite work of installing, cabling, connecting and testing before capacity goes live. NVIDIA also says factory-tested containers reduce exposure to component delays and to error-prone onsite work.

The factory work runs separately from site preparation. That means the 14 days only cover the period after the power, water and network hookups are ready, and “as little as” marks the figure as a best case.

Each standardized unit hooks into the prepared site and then scales into a larger AI factory. The design follows the philosophy of NVIDIA’s broader DSX infrastructure platform, but it has its own architecture built around purpose-built containers. One of the listed benefits is DSX MaxLPS. NVIDIA describes it as policy-driven power control, dynamic workload orchestration and high-fidelity telemetry, meant to get the most throughput within defined power limits.

Testing Catalog reported that NVIDIA also introduced an OpenShell Reference Architecture alongside DSX POD. OpenShell is NVIDIA’s runtime for fleets of autonomous AI agents. Each agent runs in its own sandbox under policies enforced at the kernel level, and the runtime is the core of NVIDIA’s Open Agent Safety Platform.

NVIDIA is working with Anduril, Armada, Bull, Crusoe, DayOne, Dell Technologies, Delta, Eaton, Foxconn (Ingrasys), GIGABYTE, HPE, Lenovo, Pegatron, Quanta Cloud Technology, Schneider Electric, Vertiv and Wistron to deliver DSX POD AI factories worldwide in early 2027.

NVIDIA connects deployment speed to revenue through time to first token, its term for the moment a new AI factory starts running models for production workloads. Jensen Huang, NVIDIA’s founder and CEO, said: “An AI factory creates value when it starts producing tokens — not when construction is complete.”