NVIDIA drops Vera Rubin AI servers: serious power for next-gen AI
NVIDIA just announced its new Vera Rubin AI servers at CES 2026, aiming to take AI performance to a whole new level.
The NVL72 rack packs 72 Rubin GPUs and 36 Vera CPUs, plus some high-speed networking tech—all built for massive-scale AI work.
Expect these beasts to hit the market later in 2026.
What's inside?
Each Rubin GPU delivers up to 50 petaflops of inference power and comes with super-fast HBM4 memory (288GB at 22TB/s).
The Vera CPU brings 88 Arm cores and supports 1.5TB of speedy LPDDR5X memory.
All together, one NVL72 rack theoretically can reach exaflop-level performance—yeah, that's a lot.
Why does it matter?
Rubin GPUs are way more efficient than NVIDIA's last-gen Blackwell chips—four times better at training huge models and 10 times cheaper for running them. Plus, you'll need fewer GPUs overall.
These servers aren't just about raw speed—they're well-suited for big workloads like Omniverse simulations, robotics, and self-driving cars.
The bottom line
If you're into cutting-edge AI or dreaming up the next big thing in tech, this is the kind of hardware that could make it happen—even if most of us won't have one under our desks anytime soon.