Nvidia has unveiled a plan to speed up datacenter deployments by shipping fully assembled compute containers, which it calls DSX pods. The modular units, announced at GTC Berlin, are built by partners including HPE, Dell, and Schneider Electric. Each 28-ton pod contains seven Vera Rubin NVL72 racks, packing 504 GPUs and 252 Vera CPUs, along with all necessary interconnects, electrical, and plumbing.
Nvidia's pitch is that customers can drop a pod onto a prepared site, connect power, networking, and cooling, and start running workloads in as little as two weeks—compared to the six months it can take to stand up a traditional datacenter. The company says factory validation of every fitting and cable will catch problems before deployment, rather than after customers have wired everything together.
The pods are customizable, with buyers able to choose among partner components, and additional containers can be added for more compute. Nvidia also used the event to detail new DSX specs for waterless datacenters, which rely on closed-loop coolants and dry coolers instead of water-thirsty cooling towers. First DSX pods are expected to roll off production lines early next year. The Register notes the effort reflects Nvidia's push to reduce supply chain friction as AI demand strains datacenter capacity. No other sources were provided, so the report stands alone.{