NVIDIA's Vera Rubin Platform Powers AI Factories with Advanced Optical Networking

NVIDIA has announced the full production ramp of its Vera Rubin platform, a highly anticipated system poised to power the next generation of agentic AI factories worldwide. This extensive POD-scale platform is engineered to function as a single, massive AI supercomputer, integrating a suite of NVIDIA technologies including Vera Rubin NVL72 systems, the NVIDIA Vera CPU, NVIDIA Groq 3 LPX, NVIDIA Vera BlueField-4 STX storage, and NVIDIA Spectrum-6 SPX Ethernet racks. The platform's design is a testament to NVIDIA's commitment to delivering unparalleled performance, efficiency, and security for complex AI workloads. A cornerstone of the Vera Rubin platform's advanced capabilities lies in its innovative cloud networking infrastructure. Specifically, the introduction of Spectrum-X Ethernet Photonics, built upon cutting-edge co-packaged optics (CPO) technology, represents a significant leap forward. This new generation of switching technology is designed to address the escalating demands of AI workloads by dramatically improving network performance and efficiency. NVIDIA highlights several key advantages of Spectrum-X Ethernet Photonics, including a remarkable five-fold increase in power efficiency, which is critical for managing the immense energy consumption of large-scale AI deployments. Furthermore, it boasts five times longer AI uptime, indicating enhanced reliability and stability for continuous AI operations. The technology also promises 1.3 times faster time to deployment compared to networks relying on traditional transceivers, streamlining the setup and scaling of AI factories. The adoption of co-packaged optics is a strategic move, as it allows for the integration of optical and electrical components into a single package, reducing signal loss and improving overall performance. By simplifying the network design and liberating more power that would otherwise be consumed by traditional networking components, NVIDIA's CPO networking provides a robust and high-bandwidth foundational fabric essential for building and operating million-GPU AI factories. This advanced networking capability is crucial for ensuring seamless data flow and communication between the vast number of GPUs required for intensive AI training and inference tasks. The Vera Rubin platform is specifically optimized for agentic AI workloads, which Jensen Huang, founder and CEO of NVIDIA, described as a new class of workload where "one prompt can launch a thousand-step journey of reasoning, retrieval, tool use and response generation." To support such demanding and multi-faceted AI processes, the platform delivers a 10x agent throughput at scale compared to the previous-generation NVIDIA Grace Blackwell platform. This significant performance improvement is directly facilitated by the high-speed, low-latency networking provided by Spectrum-X Ethernet Photonics. NVIDIA's extensive supply chain ecosystem, comprising hundreds of partners globally, is actively engaged in manufacturing Vera Rubin-based systems at scale. Major server makers and infrastructure providers, including Dell Technologies, HPE, Lenovo, and Supermicro, are in full production, ensuring that AI labs, cloud providers, and hyperscalers can readily access the technology needed to build tomorrow's intelligence. Cloud providers such as CoreWeave, Lambda, and Oracle Cloud Infrastructure are among the first ecosystem partners and adopters, underscoring the platform's immediate relevance and impact on the cloud AI landscape. The integration of NVIDIA DOCA™ software platform further enhances security across every layer of the AI factory, protecting data, agents, and inference through capabilities enforced directly in BlueField-4 silicon, ensuring a secure and reliable environment for these critical AI operations.