[DigitalToday reporter Chi-gyu Hwang] AI computing technology company Nvidia unveiled the DGX SuperPOD, a supercomputer based on its next-generation AI computing platform Rubin, at the CES 2026 IT and consumer electronics show in Las Vegas, the company said on Jan. 7. It said it also laid out an AI infrastructure blueprint for agentic AI and large-scale inference.

Nvidia DGX SuperPOD is based on the Nvidia Rubin platform, expanding scalable large system deployments and accelerating the next leap in AI computing.

Nvidia DGX SuperPOD is based on DGX Vera Rubin NVL72. It integrates 8 DGX Vera Rubin NVL72 systems and includes 576 Rubin GPUs, delivering 28.8 exaflops of FP4 performance and 600 terabytes (TB) of high-speed memory. Nvidia DGX SuperPOD is based on DGX Rubin NVL8 systems, includes 512 Rubin GPUs, and consists of 64 DGX Rubin NVL8 systems in total.

Nvidia unveiled the Rubin platform at CES 2026 in Las Vegas. The platform consists of 6 new chips for building a single AI supercomputer and was developed with a focus on accelerating agentic AI, mixture-of-experts (MoE) models and long-context inference.

Nvidia founder and CEO Jensen Huang said, "The Rubin platform was unveiled at exactly the right time, when demand for AI computing for training and inference is surging."