NVIDIA announced its DSX Pod AI Factory on Wednesday, October 21, introducing a system designed to be transported by truck and speed up AI data center deployment. The container-sized unit integrates seven Vera Rubin racks, factory-installed cabling and liquid cooling to reduce setup work at deployment sites. NVIDIA says the design could help new facilities begin serving AI tokens within 14 days, compared with roughly six months for traditional deployments.
The system targets operators building large-scale AI infrastructure, where installing and connecting computing equipment can delay service launches. NVIDIA designed the DSX Pod as a modular unit that travels on standard roads and connects with site infrastructure after delivery. Manufacturing partners are expected to produce the system, with initial deliveries planned for early 2027.
Factory Assembly Cuts On-Site Work
The DSX Pod measures roughly the size of a standard shipping container and weighs around 28,000 kilograms. Its seven Vera Rubin racks arrive with key cabling and liquid-cooling components already assembled at the factory. After delivery, operators must position the unit, connect it to power and other infrastructure, and then complete validation and testing.
This approach shifts much of the integration process away from construction sites and into controlled factory environments. Traditional deployments often require extensive coordination across servers, networking, cooling and power systems. NVIDIA’s design aims to make AI computing capacity easier to install as demand for model training and inference grows.
Manufacturing Partners Prepare for 2027
NVIDIA will offer the DSX Pod as a reference design rather than relying on a single manufacturer. The company named Foxconn, Gigabyte, Lenovo, HPE, Quanta and Pegatron among its production partners. Deliveries should begin in early 2027, although actual deployment timelines will depend on site readiness and infrastructure requirements.
The announcement also reflects NVIDIA’s broader push beyond individual processors toward integrated AI systems. By combining computing racks, cooling and cabling in a transportable package, the company aims to simplify infrastructure expansion. The proposed 14-day timeline remains NVIDIA’s target, not a guarantee for every data center.
Also Read: NVIDIA's Vera CPUs Attract Modal as Chip Rivalry Heats Up