Nvidia Launches Vera Rubin (www.tomshardware.com)

Nvidia has unveiled its next-generation AI data center architecture, Vera Rubin, during CES 2026. This architecture features a suite of six specialized chips, including the Vera CPU and Rubin GPU, integrated via advanced NVLink 6 technology to support unprecedented inference and training capabilities. Each Vera Rubin NVL72 rack is designed to deliver an impressive 3.6 exaFLOPS of inference performance and 2.5 exaFLOPS for training, marking substantial enhancements over Nvidia's previous Blackwell generation. The architectural design focuses on optimizing AI compute efficiency, utilizing memory enhancements and high-bandwidth interconnects to facilitate more effective scaling of sophisticated large language models, particularly those employing mixture-of-experts architectures. The significance of Vera Rubin lies in its potential to address the escalating demands for AI compute in various sectors, from robotics to autonomous vehicles. With its modular, cable-free design and improved reliability mechanisms, the architecture promises not just enhanced performance but also more power-efficient operations and reduced costs for AI modeling. Additionally, the introduction of the Inference Context Memory Storage Platform aims to alleviate bottlenecks in key-value cache operations, further bolstering throughput. As Nvidia anticipates a ramp-up into production in late 2026, Vera Rubin is positioned to play a crucial role in shaping the future landscape of AI infrastructure, allowing partners to better manage their investments in AI capacity.