NVIDIA Contributes Vera Rubin Designs to OCP and Opens Parts of DSX OS

NVIDIA Vera Rubin OCP is taking shape through contributions of rack and compute tray designs to the Open DC initiative. NVIDIA announced this on October 12 at the OCP Global Summit in San Jose. The designs are provided through NVIDIA’s open, modular MGX architecture; this does not mean that the GPUs themselves or the company’s entire proprietary hardware stack are being opened.

The company also announced open-source releases of selected DSX OS components. These include tools for network configuration, cluster readiness testing, rack management, and bare-metal provisioning. These components address AI infrastructure operations, from hardware preparation to configuring larger clusters.

NVIDIA Vera Rubin OCP and the MGX Framework

The contribution is tied to an expansion of the MGX 1.2 architecture. NVIDIA says this version targets racks consuming up to 250 kW. The design is expected to include liquid cooling and standardized interfaces for power and cooling.

This level of power density increases the demands on physical data center infrastructure. For AI systems, rack design now includes not only servers and networking equipment, but also power delivery, the distribution and removal of cooling fluid, and the integration of operational tools. The goal of common specifications is to simplify compatibility between individual parts of the supply chain.

NVIDIA cites faster deployment, greater energy efficiency, and lower costs. However, the announcement does not provide independent operational measurements that would confirm these product and architectural goals.

DSX OS Releases Tools for Cluster Operations

In this announcement, DSX OS is intended to cover practical tasks involved in building and managing AI clusters. Network configuration prepares communication between systems, cluster readiness testing is meant to verify their status before deployment, and rack management together with bare-metal provisioning handles the basic preparation of physical servers.

The open-source release of these tools does not mean that NVIDIA is making its entire software or hardware ecosystem available. According to the available information, the announcement concerns specific DSX OS operational components and rack and compute tray designs within MGX.

OCP Already Hosts Related Specifications

The Open Compute Project also hosts the Open Data Center Specification revision 0.7. Its authors are AMD, Google, Meta, and NVIDIA, and the document addresses data center designs for AI and ML systems for the 2027–2028 timeframe. The specification was effective as of August 27, 2026, before the October summit.

Another document is the interface specification between hyperscalers and the utility grid for disturbance performance. Google, Meta, Microsoft, and NVIDIA contributed to its preparation. This document also predates the summit: it has been effective since September 17, 2026. It is therefore not accurate to attribute its publication directly to the OCP Global Summit.

For AI infrastructure suppliers, the key point is that parts of rack, power, cooling, and operational software design are moving toward common open interfaces. However, it remains unclear to what extent hardware manufacturers or data center operators will adopt the specific Vera Rubin MGX designs.

What to Watch Next

• specific OCP contributions and licenses for Vera Rubin MGX designs,
• OEM and ODM partner products based on MGX 1.2,
• independent deployments and measurements of high-power-density racks,
• final versions or implementations of standards for resilience to utility-grid disturbances.

Verified and updated: October 13, 2026, 06:20