NVIDIA CEO Jensen Huang introduced the new Vera Rubin AI Platform during his keynote at CES 2026 in Las Vegas on January 5. This signals a move to releasing new AI chips each year after the launch of the Blackwell platform.  

Highlights from the Vera Rubin launch 

• The platform is named after astronomer Vera Rubin and is built to meet the growing computing needs of next-generation agentic AI.  

• The Rubin platform, featuring the Rubin GPU, is designed for high efficiency. It offers 5x better AI inference performance (50 petaflops of NVFP4) and trains models 3.5 times faster than the earlier Blackwell architecture.  

• The platform includes 6 main components:  

• Rubin GPU  

• Vera CPU with 88 ARM cores  

• NVLink 6 switch  

• ConnectX-9 SuperNIC  

• Bluefield 4 DPU  

• Spectrum 6 Ethernet switch  

• Huang said Rubin is already in full production, and shipments to major partners are expected in the second half of 2026.  

• The Rubin platform has been designed to lower the cost of generating AI tokens by up to 10x compared to previous architectures.  

Background: After Blackwell Ultra 

NVIDIA introduced the Vera Rubin platform as it continues to release new products every year. For 2026, the main high-performance product is the Blackwell Ultra GB300.  

• Blackwell Ultra, which has 128GB of GDDR7 memory and 3x faster attention capabilities, will be widely deployed in early 2026 to support advanced long-context AI.  

• Blackwell Ultra will be the main product for 2026, but the Rubin platform was announced at CES to prepare for an even more powerful AI platform coming in late 2026.  

Major partners for the Rubin platform include Microsoft, AWS, Google Cloud, Meta, and XAI. The systems will be integrated into Dell and HPE servers.  

Six-Chip Architecture Design 

The Rubin architecture is a six-chip system that combines advanced components for improved performance and efficiency. NVIDIA says the Rubin platform uses extreme codesign across the six chips to slash training time and inference token costs.  

Performance Breakthrough Over Blackwell 

NVIDIA’s own tests show that Vera Rubin performs much better than the current Blackwell generation, establishing new standards for AI processing.  

The improved performance meets the evolving needs of AI systems, especially for networks that process large data sets in multiple steps. During his talk, Huang said, “Vera Rubin is intended to address the basic challenge that we have: the amount of computation necessary for AI is skyrocketing.”  

Product Status and Market Development 

The Rubin Architecture is now in full production, having finished its testing phase. Huang told the CES audience, “Today I can tell you that Vera Rubin is in full production, and said more expansion is planned for later this year.”  

Cutting-Edge Storage and Infrastructure Solutions 

New architecture brings major upgrades to storage and connectivity thanks to upgraded Bluefield and NVLink systems. Dion Harris, NVIDIA’s Senior Director of AI Infrastructure Solutions, explained the importance of these changes: “As you start to enable new types of workflows like agent-based AI or long-term tasks that put a lot of stress and requirements on your KV cache,”  

The Vera CPU is designed for agent-based thinking tasks, and the two Rubin GPUs offer powerful parallel processing for demanding AI workloads.  

Strategic Market Impact 

NVIDIA’s rapid development has made it the world’s most valuable company, and the Rubin architecture is poised to further extend its lead in the AI market. Instead of upgrading just one part, the new 6-chip design offers a complete solution for next-generation AI across robotics, healthcare, and heavy industry.  

Leading cloud providers and research groups are already planning to adopt Rubin technology, demonstrating strong confidence in its capabilities and NVIDIA’s ongoing leadership as AI technology evolves.

Source: Nvidia Launches Vera Rubin Architecture at CES 2026 with Major Performance Gains