NVIDIA moves DSX POD data-center assembly into the factory
NVIDIA introduced DSX POD, a modular design that assembles computing, networking, power distribution and liquid cooling inside containers before shipment. Manufacturers will test the combined equipment while customers prepare the destination site. Partners plan deliveries in early 2027. Power, cooling and network connections still need to be ready when the containers arrive.
Artificial Intelligence··Night
Computing and cooling are assembled before shipment
NVIDIA introduced DSX POD at its GTC event in Berlin on 21 October. The modular reference design places Vera Rubin computing systems and the infrastructure needed to operate them inside purpose-built containers. Partner manufacturers will prepare the equipment before it travels to the customer. The work includes servers, networking, electrical distribution and liquid cooling within the same container-based design.[1], [2]
Manufacturers will install the rack-to-rack cables, electrical wiring and liquid-cooling pipes. Functional validation will also take place at the manufacturing facility. NVIDIA describes factory integration and testing as a way to reduce exposure to component delays and the need for assembly at the destination. The system is intended to support both training AI models and running them in production, where they generate responses for users.[1], [2]
The destination still needs prepared connections
Site preparation will proceed separately from manufacturing. When the containers arrive, power, cooling and network connections must already be available, along with the location's other supporting infrastructure. The factory handles internal integration; the destination provides the external services needed to operate the installed systems. NVIDIA contrasts the approach with onsite projects that bring planning, construction, specialist labor and logistics together at the final location.[1], [2]
Partners plan the first deliveries for early 2027
The reference design carries forward the broader DSX platform's approach to AI infrastructure while using a distinct architecture for container integration and validation before transport. The equipment assembled inside the containers will connect to the services prepared outside them. NVIDIA's announcement describes an infrastructure design and future partner deliveries, without identifying a completed customer deployment. Electricity and other external site services remain requirements for each installation.[1]
Worldwide deliveries are planned for early 2027. The participating companies include Bull, Dell Technologies, HPE, Lenovo, GIGABYTE, Schneider Electric and Vertiv, alongside other computing and infrastructure partners. Their announced role covers the equipment and physical systems that make up the modular installation. NVIDIA has not disclosed container dimensions, maximum connected power or prices. The timetable concerns the first systems that partners plan to deliver under the reference design.[1], [2]
