Nvidia DSX-Pod aims to cut AI data center setup time

Nvidia unveiled the DSX-Pod reference design at GTC Berlin to accelerate the installation of large AI data centers, aiming to reduce the timeline from an average of six months to as little as 14 days, though Jensen Huang acknowledged 'a few weeks' as a realistic target. The system involves pre-assembling Vera Rubin-series servers in shipping containers to speed up deployment. Nvidia also announced improvements to European supply chains and new techniques to lower water usage and manage power constraints, including automatic load shedding during peak demand and closed-loop cooling using water up to 45 degrees Celsius. The strategy addresses the high financial stakes of data center downtime, with estimates suggesting a gigawatt-scale center generates 100 to 200 million US dollars in revenue per day. Analysts note that while these measures improve efficiency, the overall resource consumption and demand for memory chips are expected to rise, potentially driving further price increases in the memory market next year. Nvidia argues that engineering solutions, rather than regulation, are the appropriate response to AI safety and infrastructure challenges.