On October 12, 2026, NVIDIA announced that it is contributing key Vera Rubin rack and compute-tray designs to the Open Compute Project (OCP). The company says the contribution will give equipment manufacturers a shared foundation for building high-density AI infrastructure rather than requiring every supplier to develop its own mechanical and electrical design from scratch.

The announcement came at the OCP Global Summit in San Jose, California. NVIDIA described its updated MGX architecture as a modular framework spanning compute, networking, power distribution and liquid cooling. The company says the latest designs can support racks consuming up to 250 kilowatts. That is a rack-level design capability, not a claim that every installed Vera Rubin rack operates at that power.

The MGX 1.2 specifications include liquid-cooled, fanless compute trays, blind-mate liquid-cooling connections and interoperable rack-scale infrastructure. Blind-mate connectors are designed to engage as a component is inserted, potentially reducing the number of manual hose connections during maintenance. Whether a deployment actually achieves faster service depends on its configuration, procedures and parts availability.

Another component is a liquid-cooled busbar designed for currents as high as 5,000 amperes. A busbar distributes electrical current within the rack. That current rating is not the same as the rack’s power consumption: electrical power also depends on operating voltage, and engineers must account for thermal performance, protection and fault conditions.

NVIDIA is also extending the MGX approach to a Vera CPU rack specification. Unlike a GPU-centered rack, this design targets CPU-heavy workloads such as AI-agent orchestration, tool execution and sandboxed processes. NVIDIA argues that these supporting tasks increasingly matter as AI systems spend more time interacting with applications and external services.

The broader Vera Rubin NVL72 design integrates 72 Rubin GPUs with 36 Vera CPUs, NVLink switching, networking and data-processing hardware. NVIDIA’s hardware documentation describes 18 compute trays and nine NVLink switch trays in a standard NVL72 rack. These component counts help explain why manufacturers want repeatable assembly, test and service procedures.

The company also announced work with ecosystem partners on facility-level infrastructure. NVIDIA said it joined AMD, Google and Meta on the Open Data Center Specification, while Google, Meta, Microsoft and NVIDIA worked on a grid-resilience specification. These documents address issues beyond the server itself, including electrical architecture, cooling and how large computing sites respond to disturbances on the power grid.

For data-center operators, the challenge is not simply installing more GPUs. High rack densities increase demands on facility power distribution, coolant delivery, heat rejection, maintenance access and operating procedures. Standardized interfaces can reduce integration friction, but operators still need to validate compatibility and site-specific safety requirements.

There is also a distinction between opening infrastructure designs and opening NVIDIA’s proprietary chips or software. OCP contributions may help competing manufacturers build interoperable systems, but they do not make Rubin GPUs, NVLink implementations or NVIDIA software open-source. The commercial impact will depend on how widely manufacturers adopt the designs and how reliably the resulting systems perform.

The new effort fits a broader industry shift already covered by BitcoinVersus.Tech, including LITEON’s 1 MW power rack, Delta’s 800 VDC and cooling plans, and Supermicro’s Vera Rubin rack shipments. These developments underscore how power and cooling equipment are becoming as important to AI deployment as the processors themselves.

The underlying announcement is documented in NVIDIA’s October 12 report. Independent reporting from SDxCentral also describes the MGX 1.2 changes. The announced benefits remain vendor claims until independently demonstrated in deployed systems.

BitcoinVersus.Tech Editor’s Note:
We volunteer daily to ensure the credibility of the information on this platform is Verifiably True. If you would like to support to help further secure the integrity of our research initiatives, please donate using the QR code provided