NVIDIA Unveils Rubin: Six New Chips Power Next-Gen AI Supercomputer

Technology firm NVIDIA unveiled the Rubin platform at CES 2026, introducing a unified AI supercomputer constructed from six new chips that collectively ship 5 instances the coaching compute of the Blackwell sequence. The platform is designed to set a brand new benchmark for setting up, deploying, and securing large-scale AI methods whereas minimizing prices to help wider adoption of AI know-how.

The Rubin platform achieves its efficiency via tight codesign throughout its six elements: the NVIDIA Vera CPU, NVIDIA Rubin GPU, NVIDIA NVLink 6 Switch, NVIDIA ConnectX-9 SuperNIC, NVIDIA BlueField-4 DPU, and NVIDIA Spectrum-6 Ethernet Switch. This built-in method reduces coaching instances and lowers inference token prices.

Rubin introduces 5 key improvements, together with next-generation NVLink interconnects, the Transformer Engine, Confidential Computing, the RAS Engine, and the NVIDIA Vera CPU. These developments allow agentic AI, superior reasoning, and large-scale mixture-of-experts (MoE) mannequin inference at as much as ten instances decrease price per token in contrast with the Blackwell platform. The system can prepare MoE fashions with 4 instances fewer GPUs, additional accelerating AI adoption.

The platform is predicted to be deployed by main AI labs, cloud service suppliers, {hardware} producers, and startups, together with Amazon Web Services (AWS), Anthropic, Black Forest Labs, Cisco, Cohere, CoreWeave, Cursor, Dell Technologies, Google, Harvey, HPE, Lambda, Lenovo, Meta, Microsoft, Mistral AI, Nebius, Nscale, OpenAI, OpenEvidence, Oracle Cloud Infrastructure (OCI), Perplexity, Runway, Supermicro, Thinking Machines Lab, and xAI.

Unveiling AI-Native Storage And Software-Defined Infra For AI Factories

NVIDIA Rubin introduces an AI-native storage and safe, software-defined infrastructure designed to help large-scale AI workloads effectively. The NVIDIA Inference Context Memory Storage Platform, powered by BlueField-4, permits for fast sharing and reuse of key-value cache knowledge, bettering throughput and power effectivity for multi-turn agentic AI purposes. BlueField-4 additionally consists of the Advanced Secure Trusted Resource Architecture (ASTRA), offering a single, trusted management level for safe provisioning and isolation in bare-metal and multi-tenant AI environments.

The platform is obtainable in a number of configurations, together with the Vera Rubin NVL72, which integrates 72 Rubin GPUs, 36 Vera CPUs, NVLink 6, ConnectX-9 SuperNICs, and BlueField-4 DPUs, and the HGX Rubin NVL8, linking eight Rubin GPUs to help x86-based generative AI platforms. NVIDIA DGX SuperPOD serves as a reference structure for deploying Rubin methods at scale, combining compute, networking, and administration software program.

Next-generation networking and storage are supplied by the Spectrum-6 and Spectrum-X Ethernet platforms, that includes co-packaged optics, AI-optimized materials, and high-speed 200G SerDes communication. These improvements ship enhanced reliability, power effectivity, and scalability, enabling Rubin-based AI factories to function throughout a number of websites as unified environments and supporting future million-GPU infrastructures.

NVIDIA Rubin In Full Production, Ready For Deployment Across Cloud And AI Labs

The new platform has entered full manufacturing, with Rubin-based methods anticipated to turn into accessible via companions within the second half of 2026. Major cloud suppliers, together with AWS, Google Cloud, Microsoft, and OCI, in addition to NVIDIA Cloud Partners equivalent to CoreWeave, Lambda, Nebius, and Nscale, are among the many first to deploy Rubin-powered cases. Microsoft plans to combine the NVIDIA Vera Rubin NVL72 rack-scale methods into next-generation AI knowledge facilities, together with future Fairwater AI superfactory websites, forming the inspiration for superior coaching and inference workloads throughout enterprise, analysis, and client purposes.

CoreWeave will incorporate Rubin methods into its AI cloud platform, supporting a number of architectures and enabling optimized efficiency for coaching, inference, and agentic AI workloads. Cisco, Dell, HPE, Lenovo, and Supermicro are additionally anticipated to supply servers constructed on Rubin merchandise. Leading AI labs, together with Anthropic, Cohere, OpenAI, Meta, and others, are adopting Rubin to coach bigger, extra succesful fashions and help long-context, multimodal methods with diminished latency and value in comparison with earlier GPU generations.

Infrastructure and storage companions, equivalent to IBM, NetApp, Nutanix, Pure Storage, SUSE, and VAST Data, are collaborating with NVIDIA to design next-generation Rubin platforms. Rubin represents NVIDIA’s third-generation rack-scale structure, supported by greater than 80 MGX ecosystem companions. Red Hat has introduced an expanded collaboration to offer an entire AI stack optimized for Rubin, leveraging its hybrid cloud portfolio together with Red Hat Enterprise Linux, OpenShift, and Red Hat AI, extensively used throughout Fortune Global 500 corporations.

The publish NVIDIA Unveils Rubin: Six New Chips Power Next-Gen AI Supercomputer appeared first on Metaverse Post.