NVIDIA introduces factory-built AI infrastructure, targeting 14-day site deployment

NVIDIA introduced DSX POD at GTC Berlin on October 21, 2026, giving partners a reference design for building complete Vera Rubin AI computing facilities inside modular containers. Deliveries are planned for early 2027. The aim is to get systems running sooner by assembling and testing them before they reach the customer’s site.

DSX POD is a design foundation, not a facility NVIDIA will manufacture and install itself. It specifies how computing racks, networking, storage, power, cooling and supporting equipment work together as a containerized system. Partners turn that design into customer-specific infrastructure, taking responsibility for manufacturing, regional engineering, deployment and support.

Assembly and testing move ahead of delivery

In NVIDIA’s description of a conventional build, teams prepare the site, install racks and cables, then validate the network and cluster in stages. DSX POD changes that sequence. Site crews can prepare land, foundations, power and external connections while factory teams assemble the containers and integrate the equipment in parallel.

The shift changes what gets tested before shipment. Rather than sending individual racks to be assembled into a cluster onsite, partners validate the installed computing environment as one deployment unit. Checks can cover equipment inventory, firmware, power-on behavior, network connections, sensors, cooling integrity and system-level functions. Much of the internal wiring, piping and cabling also happens at the factory.

Networking and storage are part of that coordinated design, rather than separate systems reconciled late in installation. The architecture covers connections within containers, between containers and to the customer’s external network. Storage equipment can vary with deployment size and the partner’s implementation, within defined performance and service requirements.

Power, cooling and transport still need engineering

The reference design sets an 800-volt direct-current power input for the core containers. Partners must engineer the equipment that supplies that input, adapting to local utility conditions, electrical codes and available equipment. The standard interface lets the container design stay repeatable without requiring every site to use the same upstream electrical system.

Cooling follows a similar division. Factory teams can assemble, flush and pressure-test internal plumbing, including the connections to racks. At the destination, crews still connect the container to prepared water and heat-removal infrastructure. They then commission the full cooling loop and check its performance under operating conditions.

Transport is also part of the engineering specification. Containers need suitable equipment restraints, weight distribution and lifting points for shipping installed rack-scale systems. Partners document configuration and test status before departure, then repeat defined checks after arrival to establish that delivery has not compromised the factory-validated system.

Partners turn the reference design into facilities

NVIDIA named a global delivery ecosystem that includes Dell Technologies, HPE, Lenovo, Foxconn’s Ingrasys, Schneider Electric, Vertiv, Crusoe and Anduril. The work spans several distinct responsibilities:

• Computing manufacturers integrate the systems; modular solution providers manufacture enclosures and bring the subsystems together.
• Mechanical, electrical and plumbing specialists adapt power and cooling, while site partners handle installation and support commissioning.

For expansion, NVIDIA describes containers as repeatable units with common interfaces and validation processes. Partners retain room to adapt implementations to local codes, customer requirements and regional supply chains. The company also keeps conventional construction and onsite integration as options: DSX POD is an additional deployment path, not a requirement for every AI facility.