NVIDIA debuts Rubin AI promising up to 10x cost cuts
NVIDIA just introduced its new Rubin AI platform at CES 2026, promising to make building and running advanced AI much more affordable.
Rubin is designed to slash the cost of using AI models by up to 10 times and cut down the number of graphics cards needed for training by four times, making powerful AI more accessible than ever.
Rubin's extreme codesign unites 6 chips
Rubin brings together six specialized chips using a unique "extreme codesign" approach.
At its core is the Vera CPU with 88 custom Olympus cores, paired with a high-powered GPU that delivers massive computing speed.
Features like NVLink 6 Switch and ConnectX-9 SuperNIC mean faster connections between chips and smoother networking overall.
Rubin to reach clouds H2 2026
Rubin will roll out in setups like the Vera Rubin NVL72, starting with big names such as AWS, Google Cloud, and Microsoft sometime in the second half of 2026.
NVIDIA hopes this move helps more people tap into advanced AI without breaking the bank or needing tons of hardware.