NVIDIA Opens Vera Rubin Architecture to OCP to Standardize AI Data Center Scaling

NVIDIA $NVDA has contributed the foundational blueprints for its next-generation Vera Rubin infrastructure to the Open Compute Project (OCP), effectively open-sourcing the facility-level designs needed to build gigawatt-scale AI data centers. [cite: 1.2.1] The updated MGX 1.2 architecture specification introduces standardized blind-mate liquid cooling at the rack scale, an interoperable interconnect design, and a liquid-cooled 5,000A busbar designed to handle the massive power density of Rubin-class systems. [cite: 1.2.1] A dedicated NVIDIA Vera CPU rack specification has been released specifically to support agentic AI workloads, handling real-time tool calling and sandboxed execution in parallel with the heavy lifting done by the Vera Rubin NVL72 training and inference stacks. NVIDIA also published the Open Data Center Specification v0.7 alongside AMD $AMD, Google $GOOG, and Meta $META, creating a unified facility blueprint spanning power architecture, cooling infrastructure, and physical layout. Meta announced it will integrate NVIDIA's Spectrum-6 102.4Tb/s Ethernet switches into its own 100T IcePack liquid-cooled racks, utilizing the open-source Facebook Open Switching System (FBOSS) to manage networks supporting over 100,000 GPUs. “We are excited about our close partnership with NVIDIA in making multiplane fabrics operationally scalable and performant, a critical requirement both for our own infrastructure builds and for our joint cloud partners.” — Omar Baldonado, Senior Director of Data Center and AI Networking at Meta. [cite: 1.2.1]