Nvidia Unveils Rubin Vera Rubin Superchip Platform for Data Centers

Nvidia on Tuesday introduced Rubin, a next-generation AI data-center platform centered on the Vera Rubin superchip, during CEO Jensen Huang’s keynote at CES in Las Vegas. Company materials described Rubin as a six-component architecture that pairs an Arm-based Vera CPU with Rubin GPUs and a suite of networking and infrastructure accelerators, and Nvidia said the chips are in "full production" and have been sampled with partners.

The Rubin platform comprises the Vera CPU, the Rubin GPU, an NVLink 6 Switch, the ConnectX-9 SuperNIC, the BlueField-4 DPU and the Spectrum-6 Ethernet Switch. Nvidia detailed a Vera Rubin superchip design that packages one Vera CPU with two Rubin GPUs in a single processor, a configuration the company presented as a building block for very large AI systems.

At CES Nvidia framed Rubin as "the next generation of AI" and a platform that "sets a new standard" for constructing and deploying models at scale. The company highlighted five headline innovations: a latest-generation NVLink interconnect, a next-generation Transformer Engine, Confidential Computing support, a RAS reliability, availability and serviceability Engine, and the NVIDIA Vera CPU. Nvidia said these elements together enable acceleration of what it described as agentic AI, advanced reasoning and massive-scale mixture-of-experts model inference, and it claimed "up to 10x lower cost per token versus the NVIDIA Blackwell platform."

Promotional material accompanying the announcement included sweeping performance claims and system-level illustrations. A company video called Vera Rubin the "world’s most powerful AI supercomputer" and cited figures such as 220 trillion transistors and 1,152 GPUs in large Rubin-based clusters, along with a reference to an MVFP4 architecture and broad workload support, from token generation to reinforcement learning and post-training scaling. Those figures and architectural claims appear in Nvidia’s public videos and product briefings.

Nvidia also announced expanded software and platform collaborations intended to speed enterprise adoption. The company said it will work with Red Hat to deliver a Rubin-optimized stack using Red Hat Enterprise Linux, Red Hat OpenShift and Red Hat AI for customers seeking turnkey deployments. Nvidia positioned the Rubin family across compute, networking and DPUs as designed to lower costs and accelerate mainstream AI use for large enterprises and cloud operators.

Questions remain about timing and deliverability. Nvidia emphasized production readiness and partner sampling at CES, but industry reporting continues to place general availability in the second half of 2026. That window suggests customers should expect hardware deliveries and broader availability later this year rather than immediate widespread deployment.

Financial commentary at the show underscored Nvidia’s dominant market position but displayed an anomalous market-cap discrepancy in some coverage; one report listed roughly $4.6 billion while other coverage treated Nvidia’s market value in the trillions, a mismatch that appears to reflect a typographical error rather than a change in valuation. Analysts and customers will be watching benchmarks, software ecosystem maturity and real-world cost metrics closely as Rubin moves from samples to volume systems.