NVIDIA wants AI factories to start producing results before construction projects drag on for months. On Oct. 21, 2026, the company introduced DSX POD, a modular design that could reduce deployment time from six months to as little as 14 days.
The idea is direct: build complete NVIDIA Vera Rubin AI factories inside modular containers, test those units in a factory, then connect them at a prepared site. Instead of assembling every major part onsite, customers receive units ready to connect to power, cooling, networking and other supporting infrastructure.
A New Route From Construction to Production
DSX POD is a reference design for complete NVIDIA Vera Rubin AI factories built by partners inside modular containers. The units arrive ready for connection, giving the project a defined path from delivery to operation without requiring every stage of construction to happen at the final site.
That shift targets one of the biggest delays in creating new AI capacity: the time between planning an AI factory and running models for production workloads. NVIDIA defines “time to first token” as the moment a new AI factory begins running models to serve those workloads. With factory-built units, that milestone could arrive in as little as 14 days instead of six months.
“An AI factory creates value when it starts producing tokens — not when construction is complete,” Jensen Huang, founder and CEO of NVIDIA, said.
That sentence captures the goal behind DSX POD. The project is not measured only by when a site is finished, but by when the infrastructure starts producing tokens for production use.
Factory-Built Units Target Delays and Onsite Risk
Factory testing gives DSX POD another important role. NVIDIA says factory-tested containers reduce exposure to component delays and costly, error-prone onsite work. Components can arrive as part of a prepared unit, while the final site focuses on connecting the container to the infrastructure already prepared for it.
The approach also creates a repeatable model for deploying AI factories in different locations. Power, cooling, networking and other supporting infrastructure remain essential, but the AI factory arrives as a modular system designed for those connections.
That combination of standardization and speed could help customers move from infrastructure construction to production workloads with fewer onsite steps. The claimed change is substantial: deployment time falls from six months to as little as 14 days when factory-built units are used.
“With NVIDIA DSX POD, we’re bringing factory-built precision and speed to AI infrastructure construction — reducing deployment time from months to weeks so customers can scale capacity and bring revenue-generating production online faster,” Huang said.
A Global Partner Network Is Preparing for 2027
NVIDIA is working with a global ecosystem of computing and infrastructure partners to deliver modular AI factories worldwide in early 2027. The partner list includes Anduril, Armada, Bull, Crusoe, DayOne, Dell Technologies, Delta, Eaton, Foxconn (Ingrasys), GIGABYTE, HPE, Lenovo, Pegatron, Quanta Cloud Technology (QCT), Schneider Electric, Vertiv and Wistron.
That lineup brings together companies connected to computing and infrastructure delivery, giving DSX POD a partner network for building and delivering the modular units. NVIDIA says these partners are working with the company on DSX POD AI factories worldwide in early 2027.
- DSX POD is a reference design for complete NVIDIA Vera Rubin AI factories.
- The factories are built by partners inside modular containers.
- Units arrive ready to connect to prepared power, cooling and networking infrastructure.
- Factory-built deployment can reduce the timeline from six months to as little as 14 days.
- Factory testing reduces exposure to component delays and costly, error-prone onsite work.
- Global delivery is planned for early 2027.
The next milestone is clear: early 2027, when NVIDIA and its partners plan to deliver these modular AI factories worldwide. If DSX POD reaches its stated deployment timeline, the race to build AI capacity could focus less on completing long construction projects and more on reaching the first production token.
Based on
