Nvidia's Factory-Built AI Containers Promise 14-Day Data Center Deployment
At GTC Berlin, Nvidia unveiled DSX POD, a prefabricated shipping container design that arrives fully assembled with GPUs and cooling systems, slashing typical data center setup timelines from months to roughly two weeks.
Nvidia introduced DSX POD at its GTC Berlin conference this week, a reference architecture for shipping containers that manufacturing partners will equip with Vera Rubin racks, networking infrastructure, and liquid cooling systems before shipment leaves the factory. The company contends that conventional data center rack installation and wiring typically demands months of labor, whereas a DSX POD unit can begin generating tokens within as few as 14 days, provided the receiving facility already possesses adequate electrical supply, thermal management, and network infrastructure in place.
Nvidia founder and CEO Jensen Huang articulated the underlying rationale in a statement: "An AI factory creates value when it starts producing tokens — not when construction is complete."
Container specifications and contents
Seven Vera Rubin NVL72 racks fit within each container, delivering a combined 504 Rubin GPUs. According to Ian Buck, Nvidia's vice president of hyperscale and high-performance computing, each unit incorporates approximately 10 million individual components and weighs roughly 28 metric tons.
The containers arrive with roughly 20 kilometers of copper cabling and 17 kilometers of scale-out networking already installed and validated. Upon arrival at a data center, installation crews position the container, establish connections to adjacent units, and execute final burn-in testing procedures, Buck explained during a press briefing preceding the formal announcement.
Cooling and water efficiency
Nvidia engineered the system to eliminate water consumption for thermal management. Vera Rubin operates effectively with coolant temperatures reaching 45 degrees Celsius, permitting the use of dry coolers—devices functioning similarly to automobile radiators—rather than the evaporative cooling towers that remain prevalent in many data center operations today.
In supplementary DSX documentation, Nvidia notes that operators in certain geographic regions may require chiller systems during peak temperature periods. For a 2-gigawatt deployment, the company projects that dry cooling can conserve approximately 54 million liters of water daily in comparison to traditional cooling tower systems.
Containerized data centers: A recurring concept
Containerized infrastructure has surfaced repeatedly throughout technology industry history. Sun Microsystems debuted Project Blackbox in 2006, a complete data center housed within a 20-foot container. In 2016, then-AWS CEO Andy Jassy concluded his re:Invent keynote by driving Snowmobile, a 45-foot container capable of transporting 100 petabytes of data, onto the stage. Microsoft and AWS have subsequently constructed containerized facilities for emergency response and military applications.
The DSX POD approach diverges fundamentally from these earlier implementations. Rather than functioning as standalone, self-contained data centers, DSX PODs operate as modular components that integrate into existing data center infrastructure. Buck characterized them as foundational building blocks for constructing new AI factories.
Manufacturing partners and timeline
Seventeen partners are manufacturing DSX POD systems, including Armada, which currently produces containerized edge data centers, alongside Bull, Dell, HPE, Schneider Electric, and Vertiv. Nvidia anticipates that partners will commence worldwide deliveries of DSX POD systems during early 2027.