Nvidia aims to cut AI factory deployment from 6 months to 14 days
Key Facts
Nvidia unveiled a reference design for AI factories at its GTC Berlin conference on 21 October 2026, saying it could cut deployment time at prepared sites from 6 months to a minimum of 14 days. Named "DSX Pod," the design is based on the company's "Vera Rubin" platform. It allows partners to build complete AI factories in modular containers, with systems integrated and tested before delivery to customers, according to Nvidia's statement. The announcement focuses on bringing computing capacity into production faster by moving assembly and integration work to a stage before equipment arrives on site. Jensen Huang said in the launch statement: "An AI factory creates value when it starts producing tokens, not when construction is complete."
The design initially targets new, large-scale AI infrastructure projects using liquid cooling, including those of Nvidia's cloud partners, according to the official product page. These projects still require prepared land and connections for power, facility water, cooling and networking before receiving the units. The company said "customers still need a prepared site," as field work, external connections, commissioning and verification after arrival remain part of implementation. In its technical blog, Nvidia explained that land, foundations, power and external connectivity can be prepared in parallel with container manufacturing and the integration of systems inside them. This arrangement reduces reliance on carrying out site work and unit manufacturing sequentially, providing the operational basis for faster deployment.
On the electrical side, the design uses a standardized 800-volt direct-current input, according to Nvidia's technical blog. Partners engineer generation, conversion and electrical solutions suited to each site's needs. Factory testing can include rack inventories and checks of firmware versions and approved system configurations. It also includes checks of cooling integrity, connectivity and system functions before shipment. Underscoring the role of systems integrators, Dell Technologies said its modular data center design using "DSX Pod" would combine computing, storage and networking in an integrated container. It added in its statement that the container would also include rack power and liquid cooling, bringing these components together within the AI factory it offers.
Alongside the modular design, Nvidia presented capabilities for managing electricity consumption and cooling within the broader "DSX" platform. The company said its "Max LPS" software can produce the same number of tokens using up to 30% less energy. This relies on dynamically allocating power during operation, according to its statement. For cooling, the broader platform's designs use liquid circulating in a closed loop to transfer heat. They use external dry coolers that discharge that heat into the air without evaporating water, according to Nvidia's description of the design.
The announcement comes after Nvidia reported revenue of $96.2 billion in the second quarter of its fiscal year 2027, according to its quarterly report. Revenue rose 106% from the same period a year earlier and 18% from the previous quarter. On infrastructure financing, Reuters reported in August 2026 that Nvidia had teamed up with 6 financial institutions to launch computing financing platforms. Those platforms aim to raise more than $500 billion in third-party capital for AI infrastructure. In a subsequent development during August, Reuters cited the Wall Street Journal as reporting that the company had suspended some deals under a financing initiative providing credit support to AI cloud companies in exchange for a share of revenue. During Thursday's session, Investing.com coverage linked volatility in AI stocks to concerns over heavy spending on debt-financed infrastructure. The coverage also pointed to rising borrowing costs as part of the broader pressure on the sector's shares.
Under the division of responsibilities for implementing the new design, Nvidia develops, tests and validates the design foundation, according to the official product page. Ecosystem partners handle engineering, manufacturing and integration, turning the reference design into customer solutions. Their responsibilities also include customizing and delivering solutions to meet project requirements. The partners' role extends to service and support after delivery under the implementation model presented by the company. Nvidia is working with these partners to target global deliveries of "DSX Pod" factories in early 2027.