Nvidia's DSX POD could cut AI data center setup from six months to as little as 14 days
decoder ·
Nvidia has unveiled the DSX POD, a modular data center solution designed to drastically reduce deployment timelines for AI infrastructure. Unveiled at GTC Berlin, this reference design utilizes Nvidia's Vera Rubin systems and is fully preassembled within shipping containers by partners including Dell, HPE, Lenovo, and Schneider Electric. The manufacturing process involves installing all necessary computing hardware, networking gear, cabling, power supplies, and liquid cooling systems at the factory. Consequently, on-site installation requires only connecting the units to existing power, cooling, and network infrastructure. This approach aims to cut setup time from six months down to as little as 14 days, allowing operators to generate revenue from AI models much sooner. Additionally, the broader DSX platform claims to use up to 100 percent less cooling water by employing closed-loop systems that release heat into the air rather than through evaporation. While some climates may still require extra cooling during extreme heat, the first units are scheduled to ship worldwide in early 2027.
decoder