NVIDIA opens Vera Rubin designs for AI factory buildouts

NVIDIA used the Open Compute Project Global Summit to expand its open infrastructure push for AI factories, contributing Vera Rubin rack and compute tray designs through the NVIDIA MGX architecture. The company said the work gives system builders a common foundation for next-generation AI factory infrastructure, spanning compute, networking, power, cooling and facility design.

The updated MGX 1.2 specification adds rack-scale blind-mate liquid cooling, interoperable rack-scale interconnect infrastructure, a liquid-cooled 5,000A busbar, native 19-inch equipment support and higher-density rack designs. NVIDIA said the architecture supports rack deployments up to 250kW while improving assembly, testing, serviceability and infrastructure reuse.

NVIDIA also expanded DSX OS with open source components for network configuration, cluster validation, rack management and bare-metal provisioning. Separately, NVIDIA joined AMD, Google and Meta on Open Data Center Specification v0.7, while Google, Meta, Microsoft and NVIDIA published an OCP Grid Resilience Specification for large-scale AI infrastructure.

The summit announcements extended across the partner ecosystem. Visionbay introduced K-1, an AI factory platform built around 800 VDC architecture for future Vera Rubin deployments. Meta said it is adopting NVIDIA Spectrum-6 102.4Tb/s Ethernet switches, NVIDIA and Cisco announced the USFP pluggable optics form factor, and Google Falcon Transport support is coming to future NVIDIA ConnectX-10 SuperNICs and BlueField DPUs.