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

In Brief

Nvidia unveiled the Rubin platform at CES 2026, combining six new chips into a unified AI supercomputer that delivers 5x the training compute of its Blackwell line.

Technology company NVIDIA unveiled the Rubin platform at CES 2026, introducing a unified AI supercomputer built from six new chips that collectively deliver five times the training compute of the Blackwell series. The platform is designed to set a new benchmark for constructing, deploying, and securing large-scale AI systems while minimizing costs to support wider adoption of AI technology.

The Rubin platform achieves its performance through tight codesign across its six components: 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 integrated approach reduces training times and lowers inference token costs.

Rubin introduces five key innovations, including next-generation NVLink interconnects, the Transformer Engine, Confidential Computing, the RAS Engine, and the NVIDIA Vera CPU. These advancements enable agentic AI, advanced reasoning, and large-scale mixture-of-experts (MoE) model inference at up to ten times lower cost per token compared with the Blackwell platform. The system can train MoE models with four times fewer GPUs, further accelerating AI adoption.

The platform is expected to be deployed by leading AI labs, cloud service providers, hardware manufacturers, and startups, including 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 secure, software-defined infrastructure designed to support large-scale AI workloads efficiently. The NVIDIA Inference Context Memory Storage Platform, powered by BlueField-4, allows for rapid sharing and reuse of key-value cache data, improving throughput and energy efficiency for multi-turn agentic AI applications. BlueField-4 also includes the Advanced Secure Trusted Resource Architecture (ASTRA), providing a single, trusted control point for secure provisioning and isolation in bare-metal and multi-tenant AI environments.

The platform is offered in multiple configurations, including 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 support x86-based generative AI platforms. NVIDIA DGX SuperPOD serves as a reference architecture for deploying Rubin systems at scale, combining compute, networking, and management software.

Next-generation networking and storage are provided by the Spectrum-6 and Spectrum-X Ethernet platforms, featuring co-packaged optics, AI-optimized fabrics, and high-speed 200G SerDes communication. These innovations deliver enhanced reliability, energy efficiency, and scalability, enabling Rubin-based AI factories to operate across multiple sites 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 production, with Rubin-based systems expected to become available through partners in the second half of 2026. Major cloud providers, including AWS, Google Cloud, Microsoft, and OCI, as well as NVIDIA Cloud Partners such as CoreWeave, Lambda, Nebius, and Nscale, are among the first to deploy Rubin-powered instances. Microsoft plans to integrate the NVIDIA Vera Rubin NVL72 rack-scale systems into next-generation AI data centers, including future Fairwater AI superfactory sites, forming the foundation for advanced training and inference workloads across enterprise, research, and consumer applications.

CoreWeave will incorporate Rubin systems into its AI cloud platform, supporting multiple architectures and enabling optimized performance for training, inference, and agentic AI workloads. Cisco, Dell, HPE, Lenovo, and Supermicro are also expected to offer servers built on Rubin products. Leading AI labs, including Anthropic, Cohere, OpenAI, Meta, and others, are adopting Rubin to train larger, more capable models and support long-context, multimodal systems with reduced latency and cost compared to previous GPU generations.

Infrastructure and storage partners, such as 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 architecture, supported by more than 80 MGX ecosystem partners. Red Hat has announced an expanded collaboration to provide a complete AI stack optimized for Rubin, leveraging its hybrid cloud portfolio including Red Hat Enterprise Linux, OpenShift, and Red Hat AI, widely used across Fortune Global 500 companies.