The NVIDIA Rubin platform has arrived, and it marks a major shift in how the world will build and scale artificial intelligence. NVIDIA unveiled this next‑generation system at CES, presenting six new chips that work together as one powerful AI supercomputer. The launch signals NVIDIA’s push toward faster, more efficient and more secure AI development for organizations of every size. The NVIDIA Rubin platform relies on a strategy of tight hardware and software codesign. This approach cuts inference token costs by up to ten times. This efficiency reflects broader Data Center GPU trends, where high-performance GPUs are driving faster AI training and inference across enterprise and cloud environments. These gains open the door for broader adoption of advanced AI workloads.
NVIDIA based the platform on six components: the Vera CPU, Rubin GPU, NVLink 6 Switch, ConnectX‑9 SuperNIC, BlueField‑4 DPU and Spectrum‑6 Ethernet Switch. Together, they reduce training time and improve performance across complex tasks. Jensen Huang, founder and CEO of NVIDIA, said the demand for AI computing has surged, and Rubin arrives at the right time to meet that demand. Rubin takes its name from Vera Rubin, the astronomer whose work changed the understanding of the universe. The system includes the Vera Rubin NVL72 rack‑scale solution and the HGX Rubin NVL8 system. These units highlight NVIDIA’s push toward larger, more capable and more secure AI environments.
NVIDIA Rubin platform introduces several important innovations. NVIDIA enhanced its NVLink interconnect, Transformer Engine, Confidential Computing and RAS Engine. The new Vera CPU also plays a key role. These advances speed up reasoning tasks and long‑sequence processing. They also cut inference costs to a fraction of earlier levels. As a result, AI teams can train more complex models with fewer resources. Major cloud providers and AI labs will adopt Rubin in 2026. Support will come from AWS, Google Cloud, Microsoft, Meta, Dell, Lenovo, Oracle, HPE and many others. Leaders in AI research, such as OpenAI, Anthropic, Mistral AI and xAI, also plan to use the platform. They expect Rubin to help them train larger models and serve long‑context systems at faster speeds.
Many industry leaders praised the launch. Sam Altman of OpenAI said more compute leads to more capable models, and Rubin helps scale that progress. Mark Zuckerberg of Meta said the NVIDIA Rubin platform delivers the jump in efficiency needed to serve advanced models to billions of people. Satya Nadella of Microsoft said Rubin will support the world’s most powerful AI superfactories. These responses show strong confidence in NVIDIA’s new architecture. NVIDIA designed Rubin to support the growing demand for multistep reasoning and high‑resolution video generation. These workloads require strong processing, long‑context memory and fast communication between chips. Rubin meets these needs with 3.6TB/s of GPU bandwidth and 260TB/s in its NVL72 rack. This bandwidth exceeds the scale of the entire internet, enabling massive training jobs to run more smoothly.
The company also introduced a new AI‑native storage platform powered by BlueField‑4. It manages large context memory and improves responsiveness for agentic AI. Its secure architecture allows teams to isolate workloads without losing performance. Rubin-based systems will appear in the second half of 2026. AWS, Microsoft, Google and Oracle will be among the first to deploy them. CoreWeave will also add Rubin to its platform, offering support for advanced reasoning and MoE workloads. Server makers such as Dell, Lenovo, HPE, Cisco and Supermicro will follow with their own Rubin-ready products. The NVIDIA Rubin platform stands as a major leap in AI infrastructure. It improves efficiency, reduces costs and prepares data centers for the next wave of AI innovation. As organizations move toward agentic AI and larger models, Rubin sets the foundation for the future.
