Nvidia Unveils Modular DSX POD Containers to Shrink AI Factory Deployment to 14 Days

A 28,000-kilogram shipping container housing seven liquid-cooled racks of upcoming Vera Rubin silicon. That is the physical footprint of the DSX POD, a modular infrastructure design Nvidia unveiled at the GTC Berlin conference. The system aims to compress the timeline for bringing new artificial intelligence capacity online.[1][2][5]

The containerized reference design shifts the most complex integration work from the construction site to the manufacturing floor. Nvidia claims the approach can cut deployment time at a prepared site from six months to as little as 14 days, allowing operators to begin training models much sooner.[1][4]

"An AI factory creates value when it starts producing tokens — not when construction is complete," Nvidia CEO Jensen Huang said in a statement. Huang told the Berlin audience that slow data center start-ups can eat up billions of dollars in potential revenue while expensive hardware sits idle.[1][2]

The DSX POD is built around the company's next-generation Vera Rubin architecture. A single container houses 504 Rubin GPUs spread across seven Vera Rubin NVL72 racks, connected through approximately 17 kilometers of Spectrum-X cabling. The unit is designed to be transported by truck on ordinary roads.[1][2][3]

Inside the manufacturing facility, partners complete the rack-to-rack cabling, wiring, piping, and functional validation before the unit ever ships. The container arrives with its internal liquid-cooling loop finished and tested, reducing the risk of component delays and error-prone onsite assembly that plague traditional builds.[3][4]

The limits of a 14-day deployment

However, the heavily marketed 14-day deployment window is a vendor target that only applies after the destination is fully prepared. The timeline assumes that supporting utilities, civil work, and external network connectivity are already finished and waiting for the container's arrival.[1][4]

The site must engineer massive upstream power generation and conversion to feed the container's 800-volt DC requirement. While the DSX POD handles the internal plumbing from the racks to its cooling interface, the facility itself must supply the water and the external heat rejection systems.[3]

The modular design does not solve the primary bottleneck constraining AI expansion in regions like Europe: access to grid power. A containerized factory can be delivered in weeks, but securing the utility agreements and electrical infrastructure to power it often takes years.[3]

Nvidia is not manufacturing the containers directly. The DSX POD is a reference design that will be built and shipped by a vast ecosystem of hardware and facility partners. The roster includes Dell Technologies, HPE, Lenovo, Foxconn, Gigabyte, Schneider Electric, and Vertiv.[1][2][4]

Capital costs and neocloud pressures

Deliveries of the containerized systems are expected to begin in early 2027. Nvidia has not disclosed pricing for the DSX POD, but the capital requirements are immense. Analysts estimate that a single Vera Rubin rack can cost more than $7 million.[2][4]

At those prices, a fully loaded container represents tens of millions of dollars in capital expenditure. That financial reality makes idle installation time punishing for infrastructure operators, particularly the emerging class of "neocloud" providers that lease AI computing capacity to other businesses.[2]

Many of those specialized cloud providers have watched their shares fall recently as investors worry about the heavy debt loads required to purchase Nvidia hardware. Neocloud stocks, including CoreWeave and Nebius, sold off during the Berlin conference despite the promise of faster deployment times.[2]

"The demand of the capital is much greater than the availability of the capital," Huang acknowledged to reporters, though he maintained that underlying demand for the infrastructure remains off the charts. The DSX POD is pitched directly at improving those project economics.[2]

Expanding the manufacturing supply chain

To support the massive scale of these upcoming deployments, Nvidia is simultaneously funding expansions across its global supply chain. The company plans to increase substrate capacity for its products by two and a half times over the next three years.

Nvidia has committed a multibillion-dollar capacity reservation advance to IBIDEN to expand a plant in Japan. It is also directing a $1.1 billion advance to King Yuan Electronics to fund new testing facilities and equipment in the United States.

The shift toward containerized delivery marks Nvidia's continued climb up the product stack. The company began by selling individual graphics processing units, moved to multi-GPU servers, advanced to complete racks, and is now defining the architecture for entire prefabricated data center modules.[2]

By standardizing the electrical, cooling, and logistical packages around its silicon, Nvidia is attempting to turn the unpredictable process of data center construction into a repeatable supply chain product. The success of the DSX POD will depend on whether utility providers can keep pace.[3][4]

Efficiency claims and software controls

Alongside the physical containers, Nvidia introduced software controls designed to optimize the power and cooling efficiency of the DSX platform. The company claims its reference designs, which pair closed-loop liquid cooling with dry coolers, can bring facility cooling water use close to zero.[3]

According to Nvidia's projections, a two-gigawatt deployment utilizing these designs could save 54 million liters of water a day compared to traditional cooling methods. However, local climates will ultimately dictate whether supplemental mechanical cooling is required during the hottest periods of the year.

The company also claims its DSX MaxLPS software can produce the same number of tokens while consuming up to 30 percent less power. Like the 14-day deployment timeline, these efficiency gains will require independent verification under real-world operating conditions once the containers begin shipping.