AI

NVIDIA Opens Vera Rubin Architecture for Scalable AI Factories

Monday, October 12, 2026Read Original

Details

  • NVIDIA announced at the October 2026 Open Compute Project Global Summit that it contributed Vera Rubin rack and compute-tray designs to OCP through its open MGX architecture.
  • The initiative involves NVIDIA, AMD, Google, Meta, Microsoft, Cisco, Visionbay and other ecosystem partners across computing, networking, power and cooling.
  • MGX 1.2 adds blind-mate liquid cooling, interoperable rack-scale interconnects, a liquid-cooled 5,000A busbar, configurable I/O and support for racks up to 250kW. NVIDIA also released open-source DSX components for network configuration, cluster validation, rack management and bare-metal provisioning.
  • Partners published the Open Data Center Specification v0.7 and OCP Grid Resilience Specification, while Meta adopted Spectrum-6 networking for Vera Rubin systems. NVIDIA and Cisco introduced the planned USFP optical form factor, and Google’s Falcon Transport will support future ConnectX-10 SuperNICs and BlueField DPUs.
  • Visionbay’s K-1 design applies NVIDIA’s 800V DC reference architecture, while Spectrum-6 Multiplane targets AI factories exceeding 100,000 GPUs and USFP targets 400Gb/s-per-lane networking fabrics.[1][3][8]

Impact

The announcements push AI infrastructure toward shared mechanical, electrical, networking and software interfaces, potentially reducing integration time and broadening participation among manufacturers, cloud providers and colocation operators. Over the next 12–24 months, open rack standards and DSX automation could shift investment toward high-voltage power delivery, liquid cooling, resilient Ethernet and interoperable deployment tooling as operators prepare for larger AI factories.[1][3][5]

Rift Dispatch
NVIDIA Opens Vera Rubin Architecture for Scalable AI Factories | RiftLab