NVIDIA's new Rubin chips go big: 5x faster, way cheaper AI
NVIDIA just announced at CES 2026 that its Vera Rubin chips are officially unveiled.
These new chips promise up to five times greater inference performance and up to 10x lower cost per token for MoE inference than the previous generation.
CEO Jensen Huang showed off what they can do live on stage in Las Vegas.
What's inside?
The Vera Rubin lineup packs serious hardware: an 88-core CPU, loads of fast memory, a powerful GPU, and high-speed connections—all designed to work together smoothly.
You can even scale them up for massive AI projects, thanks to clever memory sharing across hundreds of GPUs.
Why does it matter?
With production ramping up later this year, these chips are set to power the next wave of AI—think smarter apps and quicker responses for everything from enterprise tools to creative projects.
NVIDIA's making sure there's enough muscle under the hood as demand for AI keeps growing.