Nvidia unveils containerised DSX POD design for prepared AI data-center sites
DataNews Editorial Team
Nvidia unveiled DSX POD at GTC Berlin on 21 October, describing a pre-integrated, pre-tested containerised AI data-center module built around Vera Rubin NVL72 systems. The company says deployment at a prepared site can take as little as 14 days, versus a typical six months.
- DSX POD is a pre-assembled, pre-tested containerised AI data-center module announced by Nvidia at GTC Berlin.
- Each module is described as containing seven Vera Rubin NVL72 racks and 504 GPUs, alongside networking and liquid cooling.
- Nvidia’s up-to-14-day deployment claim applies only to sites where power, cooling and network connectivity are already in place.
- Factory integration could reduce on-site deployment work, but it does not address grid-interconnection timelines or the availability of electrical capacity.
Nvidia has introduced DSX POD, a containerised AI data-center design intended to shift much of the installation, cabling and testing work from the deployment site to a partner factory.
Each POD is described as a standard-format container incorporating seven Vera Rubin NVL72 racks, networking and liquid cooling. Nvidia states that this configuration contains 504 GPUs. A single module contains roughly 10 million components and weighs about 28 metric tonnes.
The company’s central deployment claim is that an AI factory can be brought online at a prepared site in as little as 14 days, compared with a usual six-month deployment period. This is not a timeline for developing or fully commissioning a new data-center site. It assumes that electrical supply, cooling infrastructure and network connections are already available.
For operators and customers building AI clusters, the approach could reduce the share of site-based work and lessen commissioning risk where power and supporting infrastructure have already been secured. Factory integration, if delivered as described, could therefore shorten the physical deployment cycle for compute infrastructure on resource-ready sites.
However, DSX POD does not substitute for grid interconnection or solve delays in obtaining power capacity. The competitiveness of data-center operators and AI-infrastructure customers would still depend on pre-reserved electrical capacity, cooling readiness and network availability. No module power draw, cooling load, commercial availability, price, deployed-unit count, customers or operating sites were provided.
The design points to greater factory integration of AI compute infrastructure, which could reduce on-site assembly, cabling and commissioning work for clusters deployed at resource-ready locations. It does not remove the principal infrastructure constraint: operators still need power, cooling and network connectivity in place before the module can be deployed.
- — / October 24, 2026 · EnglishSource →
This article was prepared by the DataNews editorial team based on the sources listed above.