Nvidia promises DSX POD AI factories ready in 14 days

At GTC Berlin on Wednesday, Nvidia introduced DSX POD, a blueprint for complete AI data centers that its partners assemble in shipping containers and deliver pre-integrated. The company says deployment drops from six months to "as little as 14 days," as long as the site is already prepared.

That figure comes from the manufacturer and describes its best case. There is no independent verification yet.

The idea is to move work from the construction site to the factory. Cabling between racks, piping, power for each rack, liquid cooling and functional tests are done before shipping. On site, all that is left is connecting power, cooling and the network.

The units use Vera Rubin systems, Nvidia's generation of AI computing. Partners that will build them include Dell Technologies, HPE, Lenovo, Foxconn (Ingrasys), Schneider Electric, Vertiv and Crusoe, among 10 other names on the list. Jensen Huang, the company's founder and CEO, summed up the bet: value arrives "when it starts producing tokens," not when construction ends. A token is the unit of text that AI models process.

Nvidia calls a data center dedicated to producing those tokens an "AI factory." The approach recalls the modular container data centers that already exist. What is new is that the design is Nvidia's and several suppliers build it at once.

For anyone who does not buy infrastructure, the effect is indirect. An analysis by Constellation Research locates the problem in how slowly AI capacity gets built and sets as a requirement a site that already has power. If the promise holds, the pressure would shift from equipment to electricity and available sites.

The release gives no price or capacity per container. Partners would deliver DSX POD around the world from early 2027.