Nvidia introduces the DSX POD, a new reference design for AI factories that drastically speeds up infrastructure rollout. By using modular containers, installation time can be cut from months to just fourteen days.
At the GTC conference in Berlin, Nvidia announced DSX POD. The solution tackles the logistical challenges of building large-scale compute for AI models. Nvidia supplies modular ‘containers’ based on the Vera Rubin architecture, which make an AI factory operational in just 14 days.
Modular containers
The core of the DSX POD is moving the installation work to the factory. Instead of carrying out complex tasks at the final site, the systems are built, integrated and validated in modular containers before they are shipped. This includes the cabling, rack-level power supply and even liquid cooling.
Nvidia chief Jensen Huang stresses that an AI factory only creates value when it produces tokens. By speeding up the construction of the infrastructure, customers can scale up their capacity faster. This reduces the dependence on specialised labour and complex logistics at the construction site itself.
The use of factory-ready containers also reduces the risk of errors and delays caused by missing components. As a result, the road to model training and production inference is considerably shorter for organisations that want to grow quickly.
Broad ecosystem of partners
The rollout of this technology relies on a broad network of partners. Nvidia is working with various players from the industry, including Dell Technologies, HPE, Lenovo, Schneider Electric and Vertiv. This collaboration ensures that the modular AI factories can be delivered worldwide. Other parties involved in this ecosystem include Bull, Foxconn, Gigabyte and Wistron.
By combining this diverse expertise, the DSX POD infrastructure can be produced at scale. The design follows the philosophy of the broader Nvidia DSX platform, which emphasises reproducibility. The first deliveries of these modular AI factories are expected in early 2027.