Six months to 14 days: Nvidia (NVDA.US) launches DSX POD, AI factory delivery ushered in “containerization”

The Zhitong Finance App learned that Nvidia (NVDA.US) released the DSX POD at the GTC conference in Berlin. This is a set of reference designs for the complete Vera Rubin AI factory. The partner built it in a modular container and integrated and verified before delivery.

Once on site, these units can be connected to the site's power, cooling, and network infrastructure, which is expected to reduce deployment time from six months to as little as 14 days.

The design moved the integration of computing systems, networking, cabling, rack-level power supply, and liquid cooling from the construction site to the factory. Nvidia said this approach can reduce construction volume, labor requirements, logistics challenges and delays, and enable customers to launch AI training and inference systems faster and generate revenue earlier.

Nvidia CEO Hwang In-hoon said, “AI factories create value when starting token production — not when construction is completed.” “With Nvidia DSX POD, we bring the precision and speed of factory manufacturing into AI infrastructure construction — shortening deployment time from months to weeks, so that customers can expand production capacity and launch income-generating production systems faster.”

According to Nvidia, its DSX reference design uses closed-loop liquid cooling and dry coolers to achieve close to zero cooling water for the facility. For a 2-gigawatt deployment, the potential water savings could reach 54 million litres per day.

Additionally, DSX MaxLPS software can help operators reduce power consumption by up to 30% when generating the same number of tokens. CoreWeave (CRWV.US), Firmus, Google Cloud (GOOGL.US), IREN (IREN.US), Lambda, Nebius (NBIS.US), Nscale, and Sharon AI all plan to adopt this technology.

The AI company is collaborating with Anduril, Armada, Bull, Crusoe, DayOne, Dell Technologies (DELL.US), Delta, Eaton (ETN.US), Foxconn's Ingrasys, Gigabyte, HPE (HPE.US), Lenovo, Asus, Guangda Cloud Technology, Schneider Electric (SBGSF.US), and Vertiv (VRT.US) to deliver DSX POD AI plants globally, which is expected to begin deployment in early 2027.