The Rubin Era: NVIDIA’s Next-Gen AI Factory Chips Enter Mass Production

On 6 January 2026, NVIDIA announced at CES 2026 that its next-generation AI computing platform, codenamed “Rubin,” has entered full-scale mass production. The Rubin platform centres on six newly designed chips, notably the Rubin GPU and Vera CPU, with the Rubin GPU built on TSMC’s 3nm process and featuring a third-generation Transformer Engine, delivering up to 50 PFLOPS of NVFP4 AI inference performance and a 3.5× improvement in training performance compared with Blackwell.

The Vera CPU is Arm-based and optimised for AI inference, boasting 88 Olympus cores and up to 1.2 TB/s memory bandwidth, while NVLink 6 Switch provides up to 3.6 TB/s of bandwidth per GPU to support large-scale MoE models. Networking is handled by ConnectX-9 SuperNIC and Spectrum-6 Ethernet switches, offering end-to-end connectivity up to 800 Gb/s for AI factories.

In DGX configurations, the DGX Vera Rubin NVL72 racks eight systems with 576 Rubin GPUs and 36 Vera CPUs, interconnected by NVLink 6, and the DGX Rubin NVL8 offers a compact, liquid-cooled option. The platform also includes the Inference Context Memory Storage Platform based on the BlueField-4 DPU to share context memory across multiple GPUs, with industry support from Microsoft, AWS, Google Cloud and Oracle; Microsoft plans to deploy Vera Rubin NVL72 systems in its Fairwater AI super-factory.