NVIDIA unveils DSX POD reference design for Europe AI factories
NVIDIA said on Wednesday it has introduced DSX POD, a reference design for complete Vera Rubin AI factories built by partners inside modular containers, with the company saying the system can reduce deployment at a prepared site from six months to as little as 14 days.
Evidence and context
The announcement was published by NVIDIA Newsroom on Oct. 21, 2026, and was framed around faster infrastructure deployment for Europe. NVIDIA said the units are integrated and validated before delivery, then arrive ready to connect to prepared power, cooling, networking and other site infrastructure. The company said the approach shifts integration work such as rack-to-rack cabling, wiring, piping and functional validation from the construction site to the manufacturing facility.
NVIDIA said the aim is to shorten the path to model training and production inference, allowing customers to begin generating production tokens and revenue sooner. Founder and Chief Executive Jensen Huang said, in remarks carried by the company, that "an AI factory creates value when it starts producing tokens — not when construction is complete" and that DSX POD is intended to reduce deployment time from months to weeks.
The company said it is working with a broad partner ecosystem that includes Anduril, Armada, Bull, Crusoe, DayOne, Dell Technologies, Delta, Eaton, Foxconn (Ingrasys), GIGABYTE, HPE, Lenovo, Pegatron, Quanta Cloud Technology (QCT), Schneider Electric, Vertiv and Wistron. NVIDIA said deliveries of DSX POD AI factories worldwide are planned for early 2027.
Technical details and limitations
According to NVIDIA's technical blog published the same day, DSX POD is built around a standard 800 VDC input power design and combines compute, storage, scale-out networking and management networks inside factory-installed containers. The blog said the design also incorporates liquid-cooling plumbing with manifolds, isolation, instrumentation and leak detection, along with factory validation covering firmware baselines, power-on checks, fabric connectivity, sensors, cooling integrity and system functionality.
At the same time, the public technical material does not provide fixed DSX POD GPU counts, container dimensions, total megawatt capacity, storage capacity or cooling flow rates. NVIDIA also said sites still require prepared power, cooling, networking and final commissioning, so the 14-day figure is a vendor claim for deployment at an already prepared site rather than an independently verified benchmark for total data center construction.