At CES 2026, NVIDIA announced the Rubin chip architecture platform that will replace the current Blackwell architecture. CEO Jensen Huang said the chips with this architecture will offer increased speed for training artificial intelligence (AI) models and inference while also offering power savings.
The Rubin platform, named after astronomer Vera Rubin, combines six chips to create a complete AI ecosystem. The chips are the Vera CPU, Rubin GPU, NVLink 6 Switch, ConnectX-9 SuperNIC, BlueField-4 DPU and Spectrum-6 Ethernet Switch.
Also announced are the NVIDIA Vera Rubin NVL72 system and the NVIDIA HGX Rubin NVL8 platform. The NVIDIA Vera Rubin NVL72 consists of 72 NVIDIA Rubin GPUs, 36 NVIDIA Vera CPUs, NVIDIA NVLink 6, NVIDIA ConnectX-9 SuperNICs and NVIDIA BlueField-4 DPUs. The NVIDIA HGX Rubin NVL8 consists of 8 NVIDIA Rubin GPUs supporting x86 generative AI systems.
The performance of training AI models is up to 3.5x faster than Blackwell, up to 5x faster in inference tasks with processing power reaching 50 PFLOPS. For every Watt of power used, it performs 8x more inferences than Blackwell.
Although the chip has only just begun production, Anthropic, Microsoft, OpenAI, xAI, Meta, Google Cloud and Amazon Web Services are among the first companies to say they are using it. HPE Blue Lion and Doudna supercomputers will also use it later.
