Nvidia just unveiled Rubin - and it may transform AI computing as we know it

Nvidia announced the Rubin AI supercomputing platform at CES 2026, a six-chip architecture (including the Vera CPU, Rubin GPU with up to 50 PFLOPS NVFP4, NVLink 6, ConnectX-9 SuperNIC, BlueField-4 DPU, and Spectrum-6 Ethernet) designed to cut LLM inference token costs by up to 10x and reduce GPU requirements for MoE training by 4x; configurations like the Vera Rubin NVL72 target large-scale, lower-cost AI deployment, with initial availability to partners such as AWS, Google Cloud, and Microsoft in the second half of 2026.