Nvidia wants the slowest part of building an AI data center to happen somewhere other than the data center. At GTC Berlin on Oct. 21, 2026, the company introduced DSX POD. It is a reference design for complete Vera Rubin AI factories that partners build inside modular containers, then integrate and validate before the containers ship. According to Nvidia's announcement, this approach cuts deployment at a prepared site from six months to as little as 14 days.

The 14-day figure is Nvidia's own claim. Independent coverage repeats it as Nvidia's claim and does not confirm it. Constellation Research also points out the catch: the clock starts only once a prepared site exists, and a prepared site also needs power.

What DSX POD Actually Is

DSX POD is not a single product that Nvidia builds and sells. The company calls it a reference design, meaning a standardized blueprint that manufacturing and infrastructure partners build to. Nvidia's technical blog describes it as a factory-built design that pre-integrates rack-scale infrastructure in containers so the system can be validated before shipment. The design is built around the Vera Rubin platform.

The main change is where integration work happens. Nvidia's release says integration of compute, networking, cabling, rack-level power and liquid cooling moves from the construction site to the factory. In a conventional build, those systems are assembled and commissioned on site. Under the DSX POD model, the technical blog says, site teams prepare foundations, power and external connectivity in parallel while the containerized systems are built and tested elsewhere. Partners handle manufacturing, engineering, deployment and support.

Nvidia CEO Jensen Huang summed up the reasoning in the company's release: "An AI factory creates value when it starts producing tokens — not when construction is complete."

Who Is Building It

Nvidia named 17 partners for DSX POD: Anduril, Armada, Bull, Crusoe, DayOne, Dell, Delta, Eaton, Foxconn (through Ingrasys), GIGABYTE, HPE, Lenovo, Pegatron, QCT, Schneider Electric, Vertiv and Wistron. The list spans server makers, power and cooling equipment suppliers, and infrastructure operators. A Seeking Alpha summary published on TradingView lists the same partners except Wistron, which it omits.

Nvidia presented the launch as a way to speed up Europe's infrastructure deployment. Its release did not say how many units any partner has committed to build or where the first containers will go.

The Efficiency Claims in Nvidia's Companion Release

On the same day, Nvidia published a second release about its broader DSX platform. It included several water and power figures. All of them come from Nvidia and its partners, and none has been independently verified.

Cooling water. Nvidia says closed-loop designs using dry coolers reduce facility cooling water use to near zero. It says Vera Rubin infrastructure supports 100% liquid cooling with coolant temperatures of up to 45 degrees Celsius, and that dry coolers release heat into the air without evaporating water. Nvidia says that at 2-gigawatt scale, DSX can save 54 million liters of water a day compared with conventional cooling-tower-based systems, equivalent to the daily water use of 180,000 households. The release adds that local climate determines whether supplemental mechanical cooling is needed during the hottest periods.

Power per token. Nvidia says its DSX MaxLPS software, which dynamically allocates power within a facility's provisioned budget, can help operators produce the same number of tokens with up to 30% less power. The release cites two partner tests:

  • In a proof of concept across HGX B200 systems, Nvidia says Lambda demonstrated 24% more cluster token throughput and 19% more GPU capacity within a fixed power budget.
  • In an evaluation at an Nscale AI factory in Iceland, using four GB300 NVL72 racks within a 264-kilowatt provisioned power budget, Nvidia says Nscale demonstrated 49% higher inference throughput and 37% more managed GPU capacity.

Grid flexibility. Nvidia also described a DSX Flex trial with Emerald AI, EPRI, Nebius and the U.K. National Grid. According to Nvidia, the Blackwell Ultra GPU cluster cut its power draw by up to 40% in response to simulated grid events while critical workloads kept running. Nvidia says DSX Flex is meant to help AI factories operate as flexible power loads that support grid reliability and enable faster grid connection. The release describes an early trial with simulated events, not a response to an actual grid emergency.

The companion release also lists power and cooling companies in the wider DSX ecosystem: ABB, Eaton, GE Vernova, Hitachi Energy, Schneider Electric, Siemens Energy, Trane Technologies and Vertiv.

What 14 Days Does and Does Not Cover

For developers, utilities and local officials, the details behind the headline number matter. The 14-day figure, as Nvidia states it, covers deployment at a prepared site. It does not include the work that usually takes longest on a large data center project:

  • Power. Constellation Research notes that a prepared site needs power. The sources do not say that DSX POD changes how power is secured.
  • Foundations and connectivity. Nvidia's technical blog places foundations, power and external network connections with site teams, working in parallel with factory integration. That work still has to be finished before the containers arrive.
  • Permitting and siting. None of the sources says DSX POD affects zoning, environmental review or local approvals.

The model takes the integration and commissioning of IT, networking and cooling out of the on-site schedule. Whether that shortens a project's total timeline depends on whether site preparation and power delivery can keep up with factory output. The sources do not give data on that.

The 54-million-liter figure is an estimate for a 2 GW deployment compared with conventional cooling towers. It is not a measurement from any operating site, and the release does not describe how dry-cooler designs affect power use, noise or land footprint.

What Happens Next

Nvidia says its partners will deliver DSX POD AI factories worldwide in early 2027. Constellation Research reports that Huang said DSX POD will ship globally in 2027. The sources do not give specific customers, sites or deployment dates for the first units. Nvidia's efficiency figures for water, power per token and grid flexibility come from company releases and partner proofs of concept. They have not yet been reported from production deployments.

Sources