Nvidia Unveils Design to Cut AI Data Center Deployment to 14 Days
Key Facts
Nvidia unveiled an AI factory design on Wednesday, October 21, coinciding with Jensen Huang's keynote at its Berlin conference, that it said could cut deployment time at a prepared site from 6 months to as little as 14 days. Called "DSX Pod," the design aims to accelerate the transition of infrastructure to running revenue-generating workloads. Huang said in the company's statement: "An AI factory creates value when it starts producing tokens." The modules are designed for the "Vera Rubin" platform, with computing and supporting infrastructure integrated at the factory before delivery. In an announcement accompanying the conference, Huang summed up the scaling challenge: "Demand is extraordinary, and production capacity is the constraint."
The design initially focuses on large-scale new AI infrastructure projects using liquid cooling, according to Nvidia's product page. Customers remain responsible for providing the prepared site, power connections, facility water, cooling and networking. The shortened deployment timeline therefore begins after site preparation and does not include building the site and its utilities from scratch. The company explains in its technical blog that container manufacturing and component integration can proceed in parallel with site grading and foundation work. Power and external telecommunications work can also continue at the project site while the modules are manufactured, rather than waiting for that work to be completed before equipment integration begins.
The integrated infrastructure includes storage and networking alongside computing equipment, according to engineering details published by Nvidia. The company says coordinating these components aims to reduce the work needed to reconcile separate systems after equipment arrives on site. Before the containers are shipped, factory testing can include checks of network connectivity, sensor operation and cooling integrity. The design includes a standard 800-volt direct-current power input, with partners tasked with designing power generation, conversion and delivery solutions according to site requirements. The partner ecosystem includes Vertiv, a specialist in data center infrastructure, among those participating in the design ecosystem.
In a separate announcement about the broader "DSX" platform, Nvidia said "Max LPS" software helps operators produce the same number of tokens using up to 30% less energy. The company linked additional cooling requirements to local climate conditions, explaining that the hottest periods may require supplemental mechanical cooling. The European announcement comes in a market that holds about 5% of global AI computing capacity, compared with 75% in the United States, according to the Associated Press. Huang said Nvidia invested $10 billion in European startups and partners in the year preceding his keynote, according to the agency. The company's announced investments thus cover startups and partners, while the new factory design aims to accelerate the deployment of computing infrastructure.
In the market, Nvidia shares rose 2.1% in morning trading as Huang delivered his Berlin keynote, according to Investing.com. That was an intraday move during Wednesday's session, rather than its closing result. The gain followed a decline in the stock during the session on Thursday, October 15, according to an earlier report by the website. The stock fell 1.84% in that session to close at $217.23. The morning gains coincided with the company's presentation of plans to accelerate AI data center installation and expand supply chain production capacity.
On supply, Nvidia plans to increase supply capacity for the chip substrates used in its products to 2.5 times its current level over the next 3 years, according to its statement accompanying the Berlin conference. The plan is part of the supply chain expansion the company announced to meet demand for AI infrastructure. Nvidia and its partners aim to deliver "DSX Pod" factories globally in early 2027. That timing puts the start of deliveries after the announcement of the design tailored to the "Vera Rubin" platform. Sessions at the Berlin conference continue through Thursday, October 22, according to the conference schedule published by Nvidia.