DatacenterDynamics reported that Nvidia launched DSX POD at its GTC Berlin event this week. It is a modular, container-based version of Nvidia's design for AI data centers powered by Vera Rubin. Nvidia said the containers are built and tested before delivery. They then connect to a prepared site's power, cooling and network. Nvidia said this cuts deployment from six months to 14 days. It named 17 partners, including Dell, HPE, Schneider Electric and Vertiv, to deliver DSX POD sites worldwide in early 2027.
Nvidia also made claims about the wider DSX design. It said the design can cut water used for cooling by up to 100 percent. It also said the design can cut power use by up to 30 percent for the same output. At 2GW scale, Nvidia claims a saving of 54 million liters of water a day against cooling towers. It added that hot climates may need extra mechanical cooling during extreme heat. Nvidia also introduced DSX Flex, which it says lets data centers adjust power draw to grid needs. All figures come from Nvidia. DatacenterDynamics gives no outside tests, price, unit size or customer names.
For anyone planning AI capacity, Nvidia is offering a way to shorten build time, if the site's power and cooling are already in place. The 14-day figure rests on that prepared site, and the first deliveries are not due until early 2027.