Nvidia introduced DSX POD at GTC Berlin on October 21, a modular design for artificial intelligence data centers that the company says can cut deployment at a prepared site from six months to as little as 14 days. The announcement targets the time required to bring computing capacity into service.
The launch accompanied chief executive Jensen Huang's appeal for Europe to accelerate its AI industry. Associated Press reported that he addressed about 3,000 developers, analysts and journalists at Berlin's Tempodrom, highlighting European partners including Siemens, Mercedes-Benz and Volkswagen's Audi.
According to Nvidia, partners will assemble Vera Rubin computing systems inside containers and integrate networking, cabling, rack power and liquid cooling before shipment. Wiring, piping and functional checks take place at the factory, while preparation of the destination site proceeds separately.
The 14-day figure is a company claim tied to a prepared location, not a timetable for developing an entire site from scratch. Nvidia says the units must connect to existing power, cooling, networking and supporting infrastructure. The announcement therefore retains a requirement for site preparation alongside factory assembly.
Nvidia links the approach to earlier model training and production inference, when deployed models process requests. Huang said the economic value comes when facilities begin producing tokens, the units processed by AI models. The company's argument is that earlier operation gives customers an opportunity to generate revenue sooner.
Worldwide deliveries are planned for early 2027 through 17 partners, including Bull, Dell Technologies, HPE, Lenovo, Eaton, Schneider Electric and Vertiv, Nvidia said. The schedule places commercial rollout next year; the deployment savings remain the supplier's stated expectation ahead of those deliveries.
Comments