Nvidia used its GTC conference in Berlin to present a cluster of infrastructure announcements, according to Network World. The most relevant for data-center builders is DSX POD, a factory-built modular approach that packages compute and supporting infrastructure into integrated systems instead of assembling each site from scratch.
Nvidia is framing DSX POD as a turnkey option and says it can shorten deployment timelines from months to weeks. CEO Jensen Huang tied the pitch to revenue: in his view, an AI factory creates value once it begins producing tokens, not when construction ends. The timeline is the company's claim, and the source gives no customer deployments or site data to test it against.
The DSX reference architecture combines closed-loop liquid cooling, dry coolers and power-management software. Nvidia says its DSX MaxLPS software can produce the same volume of AI tokens with up to 30% less power. The company also estimates that a 2-gigawatt deployment using its dry-cooling approach could save 54 million liters of cooling water per day compared with conventional cooling-tower systems. Both numbers are vendor figures, and the 30% is an upper bound rather than a typical result.
Separately, Nvidia announced the OpenShell Reference Architecture. It combines Vera Rubin compute, Vera CPUs, BlueField-4 DPUs and security software to build, monitor and constrain AI agents. It extends the existing OpenShell secure runtime into full-stack infrastructure and is designed to enforce security policies while agents carry out tasks across enterprise and edge systems. Nvidia says it is available now.
The event also covered robotics and telecom. Nvidia expanded its Isaac platform, and Agile Robots introduced a humanoid reference platform due in early 2027. In telecom, T-Mobile reported the first AI-RAN call on a commercial network, while BT, Deutsche Telekom and Elisa announced pilots with Nokia software accelerated by Nvidia technology.
Why it matters: Prefabricated designs, lower power draw and reduced water use address constraints operators already face, but every headline figure here originates with Nvidia. Operators and investors should look for independent deployment data before treating the weeks-not-months timeline or the efficiency gains as settled.
TwinStack automated analysis, based on reporting by Network World. Check the original for full details.