NVIDIA has taken advantage of the CES framework to begin the deployment of NVIDIA Rubin, its new computing platform for AI. According to the green giant, it is “a new standard for build, deploy and secure the world’s largest and most advanced AI systems at the lowest cost to accelerate the widespread adoption of artificial intelligence ».
NVIDIA has been at the Las Vegas fair with several new features, such as the launch of DLSS 4.5 Super Resolution and multiple frame generation. It also announced full, native support for GeForce Now for Linux, both important announcements for PC gaming. Of course, taking into account that its hardware solutions for AI have led to multimillion-dollar revenue and profit figures, there has been no shortage of news in this section and the main one is the one we are going to comment on.
NVIDIA Rubin, deployment
The Rubin platform uses extreme code design in the six chips that make it up (the NVIDIA Vera CPU, the NVIDIA Rubin GPU, the NVIDIA NVLink 6 switch, the NVIDIA ConnectX-9 SuperNIC, the NVIDIA BlueField-4 DPU and the NVIDIA Spectrum 6 Ethernet switch) to drastically reduce training time and inference token costs as explained by Jensen Huang, founder and CEO of NVIDIA:
“Rubin arrives at just the right time, as demand for AI computing, both for training and inference, is on the rise… With our annual delivery rate of a new generation of AI supercomputers, and extreme code design on six new chips, Rubin takes a big step towards the next frontier of AI”.
The Rubin platform, which named after Vera Florence Cooper Rubinthe pioneering American astronomer whose discoveries transformed humanity’s understanding of the universe, includes the NVIDIA Vera Rubin NVL72 rack-scale solution and the NVIDIA HGX Rubin NVL8 system.
The Rubin platform features innovations including the latest generations of NVIDIA NVLink interconnect technology, Transformer Engine, Confidential Computing and RAS Engine, as well as the NVIDIA Vera CPU. These advancements will accelerate agentic AI, advanced reasoning, and large-scale Mixture of Experts (MoE) model inference at a cost per token up to 10x lower than the NVIDIA Blackwell platform. Compared to its predecessor, the NVIDIA Rubin platform trains MoE models with 4x fewer GPUs to accelerate AI adoption, according to company data.
Scalable intelligence
Agential AI and reasoning models, along with cutting-edge video generation workloads, are redefining the boundaries of computing. Multi-step problem solving requires models to process, reason and act on long sequences of tokens, NVIDIA describes. Designed to meet the demands of complex AI workloads, the five innovative technologies of the Rubin platform include:
– NVIDIA NVLink 6th generation: Provides the fast and fluid communication between GPUs necessary for today’s massive MoE models. Each GPU offers 3.6 TB/s of bandwidth, while the Vera Rubin NVL72 rack provides 260 TB/s, more bandwidth than the entire Internet. With network-integrated computing to accelerate collective operations, as well as new features to improve serviceability and resiliency, the NVLink 6 switch enables faster, more efficient AI training and inference at scale.
– CPU NVIDIA Vera: Designed for agent reasoning, NVIDIA Vera is the most energy-efficient CPU for large-scale AI factories. The NVIDIA CPU is built with 88 NVIDIA custom Olympus cores, full ARMv9.2 support, and ultra-fast NVLink-C2C connectivity. Vera promises high performance, bandwidth and industry-leading efficiency to support a wide range of workloads in modern data centers.
– GPU NVIDIA Rubin– Powered by a third-generation Transformer engine with hardware-accelerated adaptive compression, the Rubin GPU delivers 50 petaflops of NVFP4 processing for AI inference.
– NVIDIA 3rd Generation Confidential Computing: Vera Rubin NVL72 is the first rack-scale platform to offer NVIDIA Confidential Computing, which maintains data security across the CPU, GPU and NVLink domains, protecting the world’s largest proprietary model, training and inference workloads.
– Second generation RAS engine: Spanning GPU, CPU, and NVLink, the Rubin platform offers real-time health checks, fault tolerance, and proactive maintenance to maximize system productivity. The rack’s modular cable-free tray design allows for up to 18 times faster assembly and maintenance than Blackwell.
Broad ecosystem support
The platform will be widely supported and the world’s leading AI labs, cloud service providers, computer manufacturers and startups expected to adopt NVIDIA Rubin include 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.
NVIDIA Rubin is in production and products based on the platform will be available through the company’s broad ecosystem of partners. in the second half of 2026.
