AI/TLDR

NVIDIA · 2026-06-01 · major

NVIDIA Vera Rubin Hits Full Production — Next-Gen Rack-Scale AI Supercomputer Unifies 72 Rubin GPUs, 36 Vera CPUs, NVLink 6, BlueField-4, and Spectrum-X for 3.6 Exaflops NVFP4 Inference, 20.7 TB HBM4, and 10× Throughput Per Megawatt Over Grace Blackwell

NVIDIA's third-gen MGX platform enters volume production with 150+ Taiwanese suppliers and 350+ factories. NVL72 racks ship in fall 2026, training MoE models with one-quarter the GPUs of Blackwell and 10× cheaper inference per token.

NVIDIA Vera Rubin NVL72 rack-scale AI platform

NVIDIA's Vera Rubin AI factory platform is now in volume production, with NVL72 racks shipping fall 2026.

Key specs

Rubin gpus per rack72
Vera cpus per rack36
Nvfp4 inference3,600 PFLOPS
Hbm4 memory20.7 TB
Hbm4 bandwidth1.58 PB/s
Olympus cpu cores3,168
Cost per token vs gb200 nvl7210x lower
Throughput per mw vs grace blackwell10x higher

What is it?

Vera Rubin is NVIDIA's next-generation rack-scale AI compute platform, succeeding Grace Blackwell. The flagship NVL72 unit combines 72 Rubin GPUs, 36 Vera CPUs, NVLink 6 switches, BlueField-4 DPUs, and Spectrum-X Ethernet in a single liquid-cooled rack that NVIDIA bills as one accelerator.

How does it work?

Each NVL72 packs 20.7 TB of HBM4 memory at 1.58 PB/s bandwidth and 3,600 PFLOPS of NVFP4 inference. The Vera CPU contributes 3,168 ARM-compatible Olympus cores. NVIDIA's third-generation MGX reference design lets system builders compose five racks into a single agentic AI supercomputer with 260 TB/s of NVLink Switch bandwidth.

Why does it matter?

NVIDIA says Vera Rubin trains mixture-of-experts models with one-quarter the GPUs of Blackwell and delivers 10× lower cost per token at inference. Production ramping across 150+ Taiwan suppliers and 350+ factories signals NVIDIA's customers can finally start building the next round of frontier AI factories at scale.

Who is it for?

hyperscalers and frontier-AI labs planning 2026–2027 capacity

Try it

Available from Dell, HPE, Lenovo, Supermicro starting fall 2026

Sources · 4 outlets

Tags

  • nvidia
  • vera-rubin
  • rack-scale
  • agentic-ai
  • ai-infrastructure
  • data-center
  • nvl72
  • bluefield-4
  • spectrum-x
  • computex-2026

← All releases · Learn AI