NVIDIA introduced its modular artificial intelligence infrastructure design called DSX POD at the GTC event held in Berlin on October 21. Based on the Vera Rubin platform, the solution brings together computing systems and the components to run them in factory-prepared containers. According to the company’s statement, commissioning time on a pre-prepared site can be reduced from six months to 14 days. This period does not cover the entirety of building a data center from scratch.
DSX POD’s approach is to move a significant portion of the work carried out in the field to the production facility. Partners deploy servers, establish connections between cabinets, and prepare electrical wiring and liquid cooling piping. It then tests the systems before shipping. Modules reaching the customer must be connected to ready-made electricity, cooling and network infrastructure.
The scalability advantage here comes from being able to increase capacity with new modules using the same design and ports. NVIDIA’s product page states that the initial focus is on large-scale, start-up, liquid-cooled AI infrastructures. Therefore, the DSX POD is not offered as a uniform box that can be directly added to every existing data center.
The 14-day period is not a delivery guarantee for all projects, but the shortest commissioning target announced by NVIDIA. Land, energy connection and facility preparation must be completed in advance. This distinction is important: Faster module installation does not automatically introduce an electrical connection or cooling system that is not readily available.
DSX POD moves installation work to the factory
NVIDIA’s technical description shows that the design does not consist only of processor and server layout. Networking, storage, management systems, power distribution and cooling are handled together. While a standard 800 V direct current power input is defined at the container border, business partners organize the energy conversion and local electricity solution before this according to the field.
The parallel progress of factory production and field preparation is another source of gains in the calendar. While one team prepares the terrain and external connections, the other team can assemble and test the containers. After delivery, the modules need to be placed, connected to each other, checked after transportation and put into operation. Modular data centers already exist; NVIDIA provides a standardized design and verification process for Vera Rubin systems here.
NVIDIA is working with partners including Dell Technologies, HPE, Lenovo, Schneider Electric and Vertiv to deliver DSX POD-based systems globally in early 2027. Today’s announcement does not mean that systems will start being delivered immediately. The schedule announced by the company marks the next stage in which this modular approach will be carried into commercial projects.