NVIDIA introduces Rubin as a six-chip computing platform

NVIDIA introduced the Rubin platform on 5 January, combining a new GPU and Vera CPU with networking, switching and data-processing chips. The CES announcement presented six components designed together for training and serving AI models.

Its Vera Rubin NVL72 system links 72 Rubin GPUs and 36 Vera CPUs. NVIDIA also described an eight-GPU server board for x86 systems. The range targets different installations rather than a single accelerator card that can simply replace an older product.

NVIDIA said Rubin was in full production, while products from partners were scheduled for the second half of 2026. The announcement listed cloud and server companies planning to offer the platform. Its claims of lower token costs and faster training remain NVIDIA’s comparisons. For infrastructure buyers, the consequential change was a coordinated platform spanning compute, networking and context storage, with deployment still dependent on a partner’s system and delivery schedule. Capacity planning therefore remains a system decision.