At CES NVIDIA unveils Rubin, says up-to 10x inference-token-cost cuts
NVIDIA just pulled the wraps off its new AI supercomputer, Rubin, at CES 2026.
The company says Rubin can deliver up to a 10x reduction in inference token costs and needs way fewer graphics cards to train big models compared to the older Blackwell platform.
This could make cutting-edge AI a lot more accessible (and cheaper) for everyone.
Rubin packs 6 tightly integrated chips
Rubin packs six tightly integrated chips, including the powerful Vera CPU with 88 Olympus cores and a GPU that delivers serious performance.
It also features upgraded networking technology for faster connections and smoother workload management.
NVIDIA plans to launch Rubin in the second half of 2026 and is teaming up with giants like AWS, Google Cloud, and Microsoft to help bring advanced AI tools to more people.