Nvidia CEO Jensen Huang unveiled the DSX POD modular AI factory reference design at the GTC conference in Berlin. The system is built around the Vera Rubin platform and aims to streamline the deployment of AI infrastructure by pre-integrating computing, networking, cabling, power, and liquid cooling components into modular units.
By shifting assembly and verification to the factory, the DSX POD reduces on-site construction complexity and minimizes risks related to installation errors or logistics delays. Nvidia states that if the site infrastructure is ready, the deployment cycle can be reduced from approximately six months to as little as 14 days.
The initiative focuses on accelerating the time-to-value for operators by enabling them to start running production-level AI workloads and generating tokens sooner. Nvidia emphasizes that an AI factory only creates value once it begins processing data, rather than upon completion of construction.
Nvidia is collaborating with a wide range of partners to handle the manufacturing, integration, and delivery of these systems. The list includes Dell Technologies, HPE, Lenovo, Gigabyte, Foxconn Industrial Internet, Pegatron, QCT, Wistron, Schneider Electric, and Vertiv. Global delivery is scheduled to begin in early 2027.
