Nvidia introduced the DSX POD modular AI factory reference design at the GTC Berlin conference on October 21. The system integrates Vera Rubin hardware, networking, power, and liquid cooling into pre-validated modular containers.
By shifting assembly and testing to the factory, Nvidia claims deployment time on a prepared site can be reduced to as little as 14 days, compared to the traditional six-month timeline.
The process involves partners handling cabling, electrical wiring, and plumbing before shipment. Clients remain responsible for preparing site-specific power, cooling, and network infrastructure.
CEO Jensen Huang stated that faster deployment allows customers to begin model training and inference services sooner, potentially accelerating revenue generation.
Nvidia plans to deliver these modular AI factories globally by early 2027 in collaboration with partners including Dell Technologies, HPE, Lenovo, Bull, Schneider Electric, and Vertiv.
