Article

NVIDIA has unveiled the Rubin platform, featuring six new chips designed to enhance AI supercomputing capabilities. The platform offers up to 10x reduction in inference token costs and requires 4x fewer GPUs for training mixture-of-experts (MoE) models compared to the Blackwell platform.

Key components include the NVIDIA Vera CPU, Rubin GPU, NVLink 6 Switch, ConnectX-9 SuperNIC, BlueField-4 DPU, and Spectrum-6 Ethernet Switch. The platform supports massive-scale AI factories and will be integrated into next-generation data centers by partners like Microsoft, AWS, and Google Cloud, with availability expected in the second half of 2026. Significant collaborations with companies such as Red Hat and CoreWeave aim to optimize the AI stack for the Rubin platform, enhancing both performance and efficiency for AI workloads.