Article

NVIDIA has launched the Rubin platform, featuring six new chips aimed at advancing AI supercomputing, with a notable 10x reduction in inference token costs and a requirement of 4x fewer GPUs for training mixture-of-experts models compared to its predecessor. Key components include the NVIDIA Vera CPU and Rubin GPU, and the platform is set to support large-scale AI factories in next-generation data centers, with partnerships including Microsoft, AWS, and Google Cloud. Collaborations with Red Hat and CoreWeave will further optimize the AI stack for enhanced performance and efficiency.