NVIDIA contributes Vera Rubin rack designs to OCP

NVIDIA contributed Vera Rubin rack and compute tray designs to OCP through its open MGX architecture.[1] NVIDIA says the contribution gives system builders a common foundation for developing next-generation AI factory infrastructure.[1] The architecture supports rack deployments of up to 250 kilowatts and includes a liquid-cooled 5,000-amp busbar, rack-scale blind-mate liquid cooling and interoperable rack-scale interconnect infrastructure.[1] The OCP specification, effective October 6, 2026, allows third parties to build racks and trays with common blind-mate power and liquid-cooling interfaces.[2] NVIDIA says the same MGX rack footprint supports GB300 NVL72 and is intended to support Vera Rubin NVL72 and related configurations.[4]

At OCP, NVIDIA expanded DSX OS with open-source components developed from its experience running DGX Cloud, the company says.[1] One component automates switch configurations at scale, while another stress-tests GPU clusters with real workload simulations before training.[1] A service provides rack hardware abstraction and rack-scale firmware management; another component turns raw AI hardware into production-ready clusters in hours instead of days.[1] Separately, Open Data Center Specification v0.7 defines a unified blueprint for next-generation AI factory facility design.[1] These tools and specifications address different layers of deployment, from facility planning to cluster setup.[1]

Google, Meta, Microsoft and NVIDIA published an OCP grid-resilience specification covering AI infrastructure behavior during utility-grid disturbances.[1] It calls for infrastructure to restore to near-full load in less than two seconds after a disturbance.[1] Meta and NVIDIA also announced advances in open AI networking involving Spectrum-X Ethernet SuperNIC and Spectrum-6.[1] NVIDIA and Cisco announced a standard for 400-gigabit-per-second-based networking fabrics, while NVIDIA and Google announced Falcon Transport support on future ConnectX-10 SuperNICs and BlueField DPUs.[1] NVIDIA said more than 20 industry partners are showcasing components and support for 800-volt DC data centers.[4]