Nvidia’s Rubin-powered DGX SuperPOD challenges Huawei’s AI dominance with fewer GPUs while delivering unmatched Exaflops performance at industrial scale

Nvidia has introduced its latest Rubin-powered DGX SuperPOD, a groundbreaking AI supercomputing platform that delivers an extraordinary 28.8 Exaflops of performance using only 576 GPUs. This achievement marks a significant leap in efficiency and power, positioning Nvidia as a formidable competitor against Huawei’s established AI dominance. The DGX SuperPOD leverages advanced integration of cutting-edge hardware components, combining high compute density with optimized data throughput to achieve unprecedented performance at an industrial scale.

At the core of each NVL72 system within the DGX SuperPOD are 36 Vera CPUs, 72 Rubin GPUs, and 18 Data Processing Units (DPUs). This heterogeneous architecture enables a balanced and efficient workload distribution, enhancing both compute and data handling capabilities. The Rubin GPUs are specifically designed to accelerate AI workloads, while the Vera CPUs provide robust general-purpose processing. DPUs manage data movement and security, reducing bottlenecks and improving overall system efficiency.

A key factor in the DGX SuperPOD’s performance is its exceptional interconnect bandwidth. Each DGX rack achieves an aggregate NVLink throughput of 260TB/s, facilitating rapid data exchange between GPUs and CPUs. This high bandwidth is critical for large-scale AI training and inference tasks, where data locality and communication speed directly impact performance. Additionally, the DGX racks integrate 600TB of fast memory and NVMe storage per system, ensuring ample capacity and swift access to datasets necessary for complex AI models.

The design philosophy behind the DGX SuperPOD emphasizes not only raw computational power but also scalability and efficiency. By using fewer GPUs than comparable systems, Nvidia reduces power consumption, cooling requirements, and physical footprint, making the platform more accessible for industrial deployment. This approach contrasts with competitors who often rely on sheer GPU count to achieve performance, highlighting Nvidia’s focus on architectural innovation and system-level optimization.

Unveiled at CES 2026, the Rubin-powered DGX SuperPOD represents Nvidia’s commitment to pushing the boundaries of AI supercomputing. Its ability to deliver unmatched Exaflops performance with a streamlined GPU count sets a new standard for the industry. This advancement is expected to accelerate AI research and deployment across various sectors, including scientific computing, autonomous systems, and large-scale data analytics. Nvidia’s strategic integration of CPUs, GPUs, and DPUs in a cohesive system underscores the evolving landscape of AI hardware, where heterogeneous computing and high-speed interconnects are paramount.

In summary, Nvidia’s Rubin-powered DGX SuperPOD offers a compelling alternative to existing AI supercomputers by combining fewer GPUs with superior architecture and connectivity. Its industrial-scale performance and efficiency could redefine competitive dynamics in AI hardware, challenging established players like Huawei and enabling broader adoption of AI technologies worldwide.