Nvidia DSX POD Ships a Vera Rubin AI Data Center in a Shipping Container

What is inside the box

Nvidia announced the Nvidia DSX POD at GTC Berlin on October 21: a modular reference design for complete Vera Rubin AI data centers, shipped in containers. Each pod holds seven liquid-cooled Vera Rubin NVL72 racks: 504 GPUs and 252 Vera CPUs, linked by 17 kilometers of Spectrum-X cabling. The container is roughly the size of a standard shipping container, weighs about 28,000 kg, and is designed to travel by truck on normal roads.

Seventeen partners are signed on: Anduril, Armada, Bull, Crusoe, DayOne, Dell Technologies, Delta, Eaton, Foxconn's Ingrasys, GIGABYTE, HPE, Lenovo, Pegatron, Quanta Cloud Technology, Schneider Electric, Vertiv, and Wistron. Deliveries are planned for early 2027, worldwide, with European sovereign AI projects explicitly in mind.

No price was disclosed. Analysts estimate a single Vera Rubin rack can cost more than $7M, which puts a seven-rack pod well into the tens of millions before you pour a single foundation.

14 days instead of six months

The headline number is deployment time: from six months to as little as 14 days at a prepared site. The trick is moving work upstream. Partners complete rack-to-rack cabling, wiring, piping, and functional testing at the factory, in parallel with site preparation, instead of doing it all sequentially on site.

Ian Buck, Nvidia's VP of hyperscale computing, put it plainly: "The only thing that has to happen when it gets to the site is it's placed in its location, connected to the rest of the other containers, and validated through final burning and testing." Jensen Huang's version: "AI factories create value not when construction is finished, but when they start producing tokens."

Huang also noted that a Vera Rubin node takes under five minutes to assemble, versus about two hours for a Grace Blackwell node. Vera Rubin is in full production and fully liquid-cooled, which is part of what makes the container approach viable: closed-loop liquid cooling with dry coolers rejecting heat to air, which Nvidia says cuts cooling water use by up to 100% across the broader DSX platform.

Nvidia DSX POD: reading the fine print

Three caveats, because the marketing is doing a lot of lifting here.

First, the 14-day figure assumes an already-prepared site. Nvidia's own FAQ says the design moves on-site work upstream rather than removing it. Permits, power hookups, and foundations still take as long as they take. What DSX POD eliminates is the sequential integration bottleneck, not the civil engineering.

Second, the efficiency claims describe the DSX AI factory platform, not the container alone. The "up to 30% less power for the same AI output" and water savings are company figures about the platform architecture. Real-world numbers will depend on site conditions and workloads.

Third, this is distinct from but rhymes with the modular data center trend we have been tracking. VAST Data and Hyperfusion shrank an AI data center to a container earlier this month, and DayOne crossed 1 gigawatt of billings building AI power hubs in Johor. The industry has converged on the same answer: build it in a factory, ship it in a box.

Why containers, why now

The context is the grid. Datacenter electricity use rose 17% in 2025 and is set to double by 2030. Grid hookups, not chips, are now the key bottleneck on new AI capacity. When power is the constraint, the winners are whoever can turn a prepared site with a power connection into revenue fastest. Six months of on-site integration is six months of a very expensive power allocation sitting idle.

There is also a sovereignty angle. European governments want AI capacity inside their borders, built fast, without a five-year construction saga. A containerized reference design with 17 partners is Nvidia selling exactly that: AI factories as a product, not a project.

The honest summary: DSX POD does not make data centers cheap or simple. It makes them repeatable. In a market where cloud switching just became free in the UK but nobody is switching because moving is still painful, repeatability is the whole game. The cloud providers figured out years ago that standardization beats craftsmanship at scale. Nvidia is now selling that lesson, one 28-ton container at a time.