NVIDIA DSX POD Turns AI Factory Deployment From Six Months Into 14 Days

NVIDIA used its GTC Berlin keynote on Tuesday to unveil DSX POD, a reference design for complete Vera Rubin AI factories that partners build inside modular containers, fully integrated and validated before they ever leave the factory floor. According to NVIDIA, once a site has power, cooling and networking ready, deployment drops from roughly six months to as little as 14 days.

The idea is to move the slowest parts of data center construction — rack-to-rack cabling, wiring, piping, rack-level power distribution and liquid cooling — into a manufacturing facility, where they are completed and functionally tested before shipment. NVIDIA argues this reduces exposure to component delays and the error-prone onsite work that traditionally stretches AI factory builds out for months.

"An AI factory creates value when it starts producing tokens — not when construction is complete," said founder and CEO Jensen Huang, framing the product as a supply-chain rethink that changes how data centers get installed and built.

The delivery ecosystem is broad: Anduril, Armada, Bull, Crusoe, DayOne, Dell Technologies, Delta, Eaton, Foxconn's Ingrasys unit, GIGABYTE, HPE, Lenovo, Pegatron, Quanta Cloud Technology, Schneider Electric, Vertiv and Wistron are all lined up to manufacture and ship DSX POD systems, with global deliveries beginning in early 2027.

NVIDIA also repeated two efficiency claims for the wider DSX platform: cooling water use can be cut by up to 100 percent, and the same AI output can be produced with up to 30 percent less power. Both figures are company-reported and have not been independently verified. Huang said NVIDIA plans to invest nearly 10 billion dollars in Europe's technology ecosystem and works with more than 12,000 regional partners, alongside AI collaborations with companies including L'Oreal, Spotify, Vodafone, Porsche, Nestle and HSBC.

Time to first token is the economic hinge of the pitch. NVIDIA explicitly ties factory-built containers to earlier production revenue: a container that only needs to be bolted to prepared utilities starts serving inference weeks before a traditional build would. For cloud operators burning capital on land, substations and cooling plants, compressing the gap between site readiness and billable capacity is one of the few levers that directly improves return on the largest infrastructure buildout in the industry's history.

The containerized approach also signals how standardized AI infrastructure has become. When the reference design for a multi-hundred-million-dollar AI factory can be stamped out by 17 different partners, capacity itself starts to look like a fungible commodity — which is precisely the market NVIDIA wants to create as Vera Rubin ramps through 2027.