Companies
Nvidia Introduces DSX POD Server Racks to Accelerate AI Factory Development
Nvidia has launched the DSX POD, a new reference design aimed at revolutionizing the construction of data centers. Announced during the GTC Berlin event by CEO Jensen Huang, the DSX POD is a compact unit that integrates seven liquid-cooled server racks into a single container-sized design.
The DSX POD is engineered for speed, allowing prepared sites to connect to existing power and cooling systems within 14 days, a significant reduction from the previous six-month timeline. Each unit houses seven Vera Rubin NVL72 racks, totaling 504 Rubin GPUs and 252 Vera CPUs, all contained within an enclosure comparable to a standard shipping container, weighing around 28,000 kg.
Inside the DSX POD, Nvidia has implemented 17 km of Spectrum-X networking cable and advanced liquid cooling technology. The Spectrum-X platform facilitates rapid communication between thousands of chips, enabling them to function as a cohesive system. Deliveries of the DSX POD are anticipated to commence in early 2027, with Nvidia collaborating with 17 ecosystem partners, including Dell Technologies, HPE, and Lenovo.
Huang emphasized the economic potential of AI factories, citing that a 1 GW facility could cost approximately $50 billion to construct and generate around $40 billion in annual revenue. However, he cautioned that delays in deployment could lead to significant revenue losses, estimating a three-month delay could result in a $10 billion opportunity cost.
Nvidia also highlighted the efficiency of the DSX POD, claiming it can reduce power consumption by up to 30% through dynamic power allocation and nearly eliminate water usage for cooling via closed-loop systems.
Huang's focus on Europe during the keynote was strategic, as many European governments are advocating for AI sovereignty, seeking to establish local control over computing infrastructure rather than relying on foreign providers. While the 14-day installation claim is contingent on having adequate infrastructure in place, the rollout of the DSX POD could significantly impact the AI infrastructure landscape.
FAQ
What is the DSX POD introduced by Nvidia?
The DSX POD is a new reference design by Nvidia that integrates seven liquid-cooled server racks into a compact, container-sized unit, aimed at accelerating the construction of data centers for AI development.
How quickly can the DSX POD be installed?
The DSX POD can be connected to existing power and cooling systems within 14 days, significantly reducing the previous installation timeline of six months.
What components are housed within the DSX POD?
Each DSX POD unit contains seven Vera Rubin NVL72 racks, totaling 504 Rubin GPUs and 252 Vera CPUs, along with 17 km of Spectrum-X networking cable and advanced liquid cooling technology.
What are the economic implications of the DSX POD for AI factories?
A 1 GW AI factory could cost around $50 billion to build and generate approximately $40 billion in annual revenue, but delays in deployment could lead to significant revenue losses, with a three-month delay potentially costing $10 billion.
When are deliveries of the DSX POD expected to begin?
Deliveries of the DSX POD are anticipated to commence in early 2027, with Nvidia collaborating with 17 ecosystem partners, including Dell Technologies, HPE, and Lenovo.