Nvidia has contributed key Vera Rubin rack and compute-tray designs to the Open Compute Project, giving equipment makers a shared specification for building next-generation AI data centers. The updated MGX 1.2 architecture supports deployments of up to 250 kilowatts per rack.
The design includes blind-mate liquid cooling, a liquid-cooled 5,000-amp busbar, standardized power and cooling interfaces, configurable I/O bays and support for conventional 19-inch equipment. Nvidia says the modular layout should simplify manufacturing, testing and field service while allowing vendors to build interoperable systems.
The company also released open-source components for its DSX AI Factory platform. They cover switch configuration, cluster health testing, rack management and bare-metal provisioning. Separately, Nvidia joined AMD, Google and Meta on an open data-center design specification, while Google, Meta, Microsoft and Nvidia published rules for keeping large facilities stable during utility-grid disturbances.
These contributions open parts of the surrounding infrastructure, not Nvidia’s accelerators themselves. Operators still need to validate power, cooling, networking and software choices for their own facilities. The practical change is that suppliers now have common targets for rack-scale systems that would otherwise require separate proprietary designs for each large customer.