NVIDIA said on October 12 that it contributed Vera Rubin rack and compute-tray designs to the Open Compute Project through its MGX architecture. The OCP Global Summit announcement concerns shared infrastructure designs and software, not a new consumer graphics card or an independently measured performance result.
The company describes MGX 1.2 as adding rack-scale blind-mate liquid cooling, shared interconnect infrastructure and a liquid-cooled 5,000-amp busbar. It also lists native 19-inch equipment support, configurable I/O bays and fanless liquid-cooled compute trays. These are design features described by NVIDIA, not proof that every data center has adopted them.
NVIDIA says the architecture supports rack deployments up to 250 kW. That is a supported design scale, not the measured electricity use of a particular operating rack. Higher supported power should not be translated into household energy savings or an efficiency benchmark without separate evidence.
The same announcement lists new open-source DSX OS components for switch configuration, cluster readiness checks, rack management and bare-metal provisioning. NVIDIA presents them as tools cloud providers and software partners can integrate. Their release does not by itself establish production suitability for a specific operator's environment.
NVIDIA also describes open data-center and grid-resilience specifications developed with other companies. The grid-resilience specification sets expected behavior during disturbances, including restoring near-full load in under two seconds. A specification target is not a verified result for every facility built against it.
Some other items in the announcement remain future plans. Falcon Transport support is described for future ConnectX-10 SuperNICs and BlueField DPUs, and the USFP optical module is framed as a planned path to higher-bandwidth networking. Keeping contributions, current software and future hardware support separate avoids turning an infrastructure announcement into an availability guarantee.