Nvidia Unveils Modular AI Factories DSX POD on Vera Rubin

Nvidia has unveiled DSX POD, a turnkey solution for AI factories built on the Vera Rubin platform. The system is assembled in modular containers directly at the factory, fundamentally changing the approach to data center deployment.

Accelerating Deployment

Deploying ready-made infrastructure on a prepared site takes just 14 days, whereas traditional construction requires about six months. The process involves parallel preparation: racks are installed, cables are laid, and systems are tested at the factory, while foundations are poured and power is supplied at the site.

Ready-made containers are delivered to the location, connected to each other, and linked to external power, cooling, and communication networks. This assembly-line approach allows for faster token generation and revenue realization.

Energy Efficiency and Resource Savings

The platform uses constant 800 V voltage and liquid cooling, which is pressure-tested during production. Nvidia engineers claim that the architecture completely eliminates water evaporation by using dry coolers.

For a campus with a capacity of 2 GW, this ensures savings of up to 54 million liters of water per day compared to traditional cooling towers. In extreme heat, additional cooling may be required, but the main loop remains effective.

Manufacturers and Delivery Timelines

Assembly of modules will be carried out by giants such as Dell, HPE, Lenovo, Foxconn, Schneider Electric, and Vertiv, as well as defense company Anduril. Deliveries of the first complexes are scheduled for early 2027.

• Dell, HPE, Lenovo, Foxconn
• Schneider Electric, Vertiv
• Anduril (defense sector)

"An AI factory becomes beneficial when it starts generating tokens, not when construction is finished," notes Nvidia CEO Jensen Huang. The company aims to transform the market by selling complete industrial solutions rather than individual chips.