NVIDIA has unveiled the Rubin platform, a major advancement in AI computing designed to power the next generation of large-scale AI systems. Introduced at CES in Las Vegas, the Rubin platform features six new components: the NVIDIA Vera CPU, Rubin GPU, NVLink 6 Switch, ConnectX-9 SuperNIC, BlueField-4 DPU, and Spectrum-6 Ethernet Switch. Built through an advanced codesign approach, the platform is engineered to accelerate agentic AI, mixture-of-experts models, and long-context reasoning while dramatically reducing inference costs. At the heart of this innovation is the NVIDIA DGX SuperPOD, which remains the foundational architecture for deploying Rubin-based systems across enterprises and research institutions. The DGX SuperPOD integrates the full stack of NVIDIA technology—from computing and networking to software—into a unified system, eliminating complex infrastructure integration and enabling teams to focus on AI innovation. The Rubin platform introduces five key advancements that collectively deliver up to a 10x reduction in inference token cost compared to the previous generation. This efficiency leap is critical as AI models grow in size, complexity, and reasoning depth. Two new system configurations will be available in the second half of the year. The DGX Vera Rubin NVL72 unifies eight DGX Vera Rubin NVL72 systems, delivering 28.8 exaflops of FP4 performance and 600TB of high-speed memory. Each NVL72 system combines 36 Vera CPUs, 72 Rubin GPUs, and 18 BlueField-4 DPUs, creating a single, coherent memory and compute space across the rack. With 260TB/s of aggregate NVLink throughput, the system eliminates the need for model partitioning, allowing the entire rack to function as one unified AI engine. The DGX Rubin NVL8 system brings Rubin performance into a liquid-cooled, x86-based form factor, offering an efficient on-ramp for organizations to adopt the Rubin platform. Each NVL8 system features eight Rubin GPUs and sixth-generation NVLink, delivering 5.5x more NVFP4 FLOPS than previous Blackwell-based systems. Networking is a cornerstone of the Rubin platform. The next-generation 800Gb/s end-to-end suite, powered by Spectrum-6 Ethernet switches, Quantum-X800 InfiniBand switches, BlueField-4 DPUs, and ConnectX-9 SuperNICs, is designed to handle the most demanding AI workloads. This infrastructure ensures high performance, reliability, and scalability, removing traditional bottlenecks in large-scale AI clusters. NVIDIA Mission Control, the AI data center operations and orchestration software, will be extended to Rubin-based systems. It enables automated management of deployments, cluster operations, and infrastructure resiliency, including rapid leak detection and autonomous recovery. The software also enhances power and cooling efficiency, critical for managing the energy demands of gigawatt-scale AI factories. The NVIDIA AI Enterprise software platform, including NIM microservices and support for models like the Nemotron-3 family, will be available on Rubin systems, streamlining development and deployment. With the Rubin platform, NVIDIA is setting the stage for industrial-scale AI. The DGX SuperPOD with Rubin-based systems will serve as the foundation for building frontier models, multimodal AI, and agentic systems capable of reasoning across thousands of steps—delivering unprecedented intelligence at lower cost.