Nvidia announces plug and play Vera Rubin racks that can be shipped on a truck

The DSX Pod AI Factory is Nvidia’s containerized reference design for AI computing. It combines several Vera Rubin racks into one transportable unit. This matters because customers can receive a larger, integrated AI system instead of assembling every rack and connection at the data-center site.

Each pod contains seven Vera Rubin AI racks. Nvidia says the pod is about the size of a standard shipping container and is designed for normal road transport. Its cabling and liquid-cooling systems are assembled before shipment, so the main infrastructure arrives already integrated.

The pod still needs to be positioned, connected to other containers, and validated through final burning and testing. Nvidia says manufacturing partners will ship the design worldwide, with deliveries expected to begin in early 2027. The system could help customers, including those in Europe, overcome deployment delays.

A single DSX Pod contains seven Vera Rubin AI racks. Nvidia designed the pod as a compact, containerized building block rather than as a loose collection of separate servers. This gives customers a defined unit of AI infrastructure that can be manufactured, shipped, and installed together.

The pod is roughly the size of a standard shipping container. It weighs about 28,000 kilograms, or nearly 62,000 pounds. Nvidia says it is designed to travel on a truck using normal roads, despite holding multiple high-performance AI racks and their supporting systems.

The scale shows how Nvidia’s products have expanded beyond individual chips and servers. The company now describes a progression from GPUs, to multi-GPU servers, to racks, and then to pods containing several racks. Nvidia did not announce a pod price, while analysts estimate one Vera Rubin rack could cost more than $7 million.

The factory completes much of the difficult integration work before the DSX Pod leaves for its customer. Nvidia says the pod’s cabling and liquid-cooling systems are assembled at the factory. This means the delivered unit is more than a box of separate racks; key internal connections and cooling equipment are already installed.

When the pod reaches its destination, it must be placed in its planned location. It then has to be connected to the rest of the containers in the wider installation. Nvidia executive Ian Buck said the final required step is validation through burning and testing, which checks the system before it serves AI workloads.

This approach shifts labor from the construction site to a controlled manufacturing setting. Nvidia says the pod is designed for normal truck transport and worldwide customers. Manufacturing partners, including Foxconn, Lenovo, HPE, and others, are expected to build and ship the reference design.

A new AI data center normally needs substantial installation and integration work before it can serve tokens. Nvidia says the DSX Pod could reduce that timeline from six months to 14 days. The shorter schedule matters because customers can begin using AI computing capacity much sooner after the physical site is ready.

The mechanism is factory preassembly. Nvidia installs the pod’s Vera Rubin racks, cabling, and liquid cooling before shipment. The pod can then travel by truck as a largely integrated unit. At the destination, workers position it, connect it with other containers, and perform final burning and testing instead of building the complete system from separate parts.

The 14-day figure is Nvidia’s stated reduction, not an independently verified result in the article. The company intends the pod for customers worldwide and says it could address deployment challenges for European customers. Deliveries from manufacturing partners are expected to begin in early 2027.

A GPU chip performs much of the mathematical work behind AI computing, but a useful system needs more than chips. Multiple GPUs must be installed in servers and racks, connected so they can share work, supplied with data from storage, and kept within safe operating temperatures. Networking, storage, and cooling therefore support the computing hardware.

The DSX Pod reflects that system-level design. It contains seven Vera Rubin racks, with cabling and liquid cooling assembled at the factory. Nvidia also says it now offers networking, storage, and other data-center needs. These pieces help turn separate processors into an integrated installation that can run AI workloads and serve tokens.

The article presents this as Nvidia’s product evolution: from individual GPUs, to servers with four or eight GPUs, then racks, and now containerized pods. The source does not detail each component’s technical performance, but it shows why Nvidia is packaging more of the surrounding infrastructure with its computing hardware.

Nvidia’s move reflects the growing complexity of AI infrastructure. Customers need computing hardware plus networking, storage, racks, cabling, and cooling. By offering these pieces as integrated systems, Nvidia can address more of the work required to build an AI data center instead of selling only the processors inside it.

The DSX Pod shows this change clearly. It combines seven Vera Rubin racks in a containerized unit, with cabling and liquid cooling assembled at the factory. Nvidia’s systems have progressed from individual GPUs to multi-GPU servers, full racks, and now pods that can be shipped and connected as building blocks.

This changes Nvidia’s role from chip supplier toward complete data-center infrastructure provider. The company says the value of what it offers per gigawatt has risen from about $18 billion to more than $40 billion. The article links that increase to the broader amount of computing and infrastructure Nvidia supplies, though it does not give a detailed revenue breakdown.

Gigawatts refer to the size of data-center infrastructure by how much power it uses. In the article, a one-gigawatt facility is therefore a way to describe the scale of the installation’s electrical demand. It is not a count of servers or racks by itself, although larger computing systems generally require more supporting infrastructure.

Nvidia says its value supplied into one gigawatt rose from about $18 billion, to $20-something billion, and then to more than $40 billion. The company’s systems have expanded from GPUs to servers, racks, networking, storage, cooling, and integrated pods. More of those components can be included in the infrastructure serving the same power-scale measure.

The article presents this as a continuing trend. CEO Jensen Huang said his sense is that the value will keep rising. The source does not provide a formula for the figures or separate the value of each component, so the numbers describe Nvidia’s overall offering rather than a detailed cost breakdown.