NVIDIA has introduced DSX POD, a reference design for complete Vera Rubin-based AI factories that partners will build and test inside modular containers before shipping them to the site. The company announced it on October 21, 2026 at GTC Berlin and says that, on a site that’s already prepped, deployment can drop from around six months to just 14 days. First deliveries are planned for early 2027.
NVIDIA DSX POD in 30 seconds
- DSX POD is NVIDIA’s new reference design for Vera Rubin AI factories built into modular containers.
- Compute, networking, cabling, rack-level power, and liquid cooling are assembled and tested at the factory.
- NVIDIA talks about going from around six months to just 14 days, always on an already-prepared site.
- Bull, Dell, HPE, Lenovo, Schneider Electric, and Vertiv are among the 17 partners, with deliveries starting in early 2027.
The idea is to take out of the construction site much of the work that’s done today at the data center itself. According to NVIDIA, assembling the compute hardware, networking, cabling, rack-level power distribution, and liquid cooling moves to the partner’s factory. There, the cabling between racks, electrical connections, piping, and functional testing are also completed, while the site is prepared in parallel. When the module arrives, it’s connected to power, cooling, networking, and the rest of the systems already waiting on-site.
NVIDIA founder and CEO Jensen Huang summed up the approach in the announcement, saying an AI factory starts generating value when it produces tokens, not when construction finishes. He added that DSX POD brings the precision and speed of industrial manufacturing to building AI infrastructure and cuts deployment from months to weeks, so customers can start the revenue-generating production sooner.
What changes compared to a traditional deployment
The 14-day figure comes with an important caveat: it refers to a site that’s already prepared. The announcement doesn’t include permitting, grid connection, or building the site’s own power and cooling infrastructure within that timeframe — in many data center projects, those are the longest phases. What DSX POD shortens is the installation and startup of the compute once all of that is ready.
NVIDIA also doesn’t detail each module’s capacity. The announcement doesn’t give megawatts, GPU count, or racks per container, nor pricing. It does present it as part of its DSX platform, through which the company wants the power grid, the building, and the compute to be designed as a single system. Revista Cloud explained earlier how utilities are joining NVIDIA DSX to tackle the power bottleneck, and a few weeks ago how Lenovo and NVIDIA are preparing AI factories ready to deploy in weeks.
What happens at the factory and what still happens on-site
NVIDIA’s technical blog details the split of tasks. At the partner’s factory, racks are installed, internal cabling and piping are completed, control systems are integrated, base configurations are loaded, and functional tests are run. Pre-shipment validation can cover rack inventory, firmware, power-on, network connectivity, sensors, and the integrity of the cooling loop. That liquid cooling loop, which Vera Rubin systems require, is cleaned and pressure-tested before leaving, and the container includes leak detection.
In parallel, the site prepares the land, foundations, power supply, and outside connectivity. When the containers arrive, they need to be placed, connected to each other and to the site’s power, cooling, and network, checked for any transport damage, and commissioned. The design starts from a standard 800V DC electrical input at the main containers, an architecture the industry has been working on for months, as we reported when covering the Infineon and NVIDIA deal for 800V data centers. Upstream generation, conversion, and power supply remain with each project’s own partners.
NVIDIA itself qualifies the scope. Its product page states that DSX POD moves work to the factory but doesn’t eliminate what’s done on-site, and that its initial focus is new, large-scale, liquid-cooled projects aimed at large AI factory operators such as its NVIDIA Cloud Partners. The technical blog also clarifies that the container isn’t mandatory and that conventional construction remains an option. Neither document gives power figures, GPUs per container, dimensions, weight, or price, nor measured time-savings data: the benefits are presented as design goals.
Seventeen partners, deliveries in 2027
The list of companies working with NVIDIA to deliver DSX POD factories includes Anduril, Armada, Bull, Crusoe, DayOne, Dell Technologies, Delta, Eaton, Foxconn (through Ingrasys), GIGABYTE, HPE, Lenovo, Pegatron, Quanta Cloud Technology (QCT), Schneider Electric, Vertiv, and Wistron. The announcement doesn’t explain each one’s role, though the list mixes server makers, power and cooling specialists, AI cloud operators, and defense and edge companies such as Anduril and Armada.
According to NVIDIA, partners will deliver the first DSX POD factories worldwide in early 2027. Although the announcement’s headline talks about accelerating infrastructure deployment in Europe, the text doesn’t name specific European sites, customers, or countries. Bull is the only European server maker on the list, and among the power and cooling specialists are France’s Schneider Electric, Vertiv, Eaton, and Delta.
On the Vera Rubin platform underlying these modules, Jensen Huang described it in his GTC Berlin keynote as a product already in full production. Revista Cloud already reported that NVIDIA Vera Rubin NVL72 multiplies GB300’s performance by up to 3.7 times, also according to the company’s own figures.
Frequently asked questions
What is NVIDIA DSX POD?
It’s an NVIDIA reference design for complete Vera Rubin AI factories that partners assemble, wire, and test inside modular containers before shipping them.
How long does it take to deploy a DSX POD factory?
NVIDIA says that, on a site that’s already prepared, deployment can drop from around six months to just 14 days.
Does DSX POD eliminate on-site work at the data center?
No. According to NVIDIA, the land, power, cooling, and network must already be ready, and on arrival the containers still need to be connected and commissioned.
When will DSX POD be available?
Partners will deliver the first DSX POD factories in early 2027, according to NVIDIA.
Sources:
- NVIDIA Newsroom: NVIDIA Introduces DSX POD AI Factories to Accelerate Europe’s Infrastructure Deployment
- NVIDIA Technical Blog: Achieve Productive AI Capacity Faster and More Predictably with NVIDIA DSX POD
- NVIDIA: DSX POD Reference Design for AI Factory Infrastructure
- NVIDIA Blog: GTC Berlin 2026 keynote by Jensen Huang
