NVIDIA's Rubin: The next-gen AI supercomputer that slashes costs
NVIDIA just dropped Rubin, its new six-chip AI supercomputer, at CES 2026.
Designed for training massive AI models, Rubin promises to cut inference token costs by up to 10x and uses four times fewer GPUs than the previous Blackwell system.
Rubin is built for the world's largest and most advanced AI systems—think seriously big AI projects.
What makes Rubin special?
Rubin packs a punch with its Vera CPU (88 Arm-compatible cores), the high-powered Rubin GPU (50 petaflops and up to 288GB memory per chip), and a super-fast NVLink 6 Switch that links up to 72 GPUs into one mega-brain.
All these parts work together for ultra-efficient, large-scale AI training.
Who's it for?
This isn't something you'll find in your gaming rig—Rubin is aimed at big tech companies and organizations running huge AI workloads.
If you're building or deploying giant language models or need serious computing muscle, keep an eye out when shipments start in late 2026.