NVIDIA Opens Vera Rubin Architecture to OCP to Scale AI Factories

At the Open Compute Project (OCP) Global Summit, NVIDIA announced new partnerships and contributions to create a common foundation for next-generation AI infrastructure. The company contributed key NVIDIA Vera Rubin rack and compute tray designs to OCP through the open NVIDIA MGX architecture, providing system builders with a standardized base for developing AI factory infrastructure.

MGX 1.2 and Vera CPU Rack

NVIDIA joined industry leaders to contribute to the third-generation MGX AI data center reference design. The updated MGX 1.2 specification extends the open architecture with rack-scale blind-mate liquid cooling, interoperable rack-scale interconnect infrastructure, and a liquid-cooled 5,000A busbar for higher-power systems. It supports rack deployments up to 250kW and adds native 19-inch equipment support.

The NVIDIA Vera CPU rack design extends the MGX architecture beyond accelerated computing to support the agentic AI era. While Vera Rubin NVL72 handles large-scale training and inference, the Vera CPU rack is designed for tool calling, orchestration, and sandboxed execution, allowing AI agents to interact with applications and data in real time.

Open Source Software for DSX AI Factory

NVIDIA launched new open source software components for the NVIDIA DSX AI Factory platform to unify computing, networking, power, and cooling. These components, developed from experience running NVIDIA DGX Cloud, include:

• NVIDIA Config Manager: Automates switch configurations to deploy and maintain consistent AI factory network fabrics.
• NVIDIA Cluster Readiness Engine: Stress-tests GPU clusters with real workload simulations to prevent job crashes.
• NVIDIA Rack Management Service: Treats racks as first-class units for improved firmware management.
• Cluster API Provider for NVIDIA Infrastructure Controller: Accelerates bare-metal deployment, turning raw hardware into production-ready clusters in hours.

Industry Standards and Networking Advances

NVIDIA, along with AMD, Google, and Meta, published the Open Data Center Specification v0.7, defining a unified blueprint for AI factory facility design. Additionally, the OCP Grid Resilience Specification was published to ensure AI infrastructure stays connected and restores to near-full load within two seconds during grid disturbances.

In networking, Meta and NVIDIA announced that Meta is adopting NVIDIA Spectrum-6 102.4Tb/s Ethernet switches. Meta’s new N-IceCube switch, running the Facebook Open Switching System (FBOSS), slots into the IcePack liquid-cooled rack. This collaboration supports agentic AI factories of more than 100,000 GPUs using Spectrum-X Multiplane technology, which distributes traffic across up to eight independent network planes for improved resilience and bandwidth utilization.

Power and Optical Innovations

Visionbay unveiled plans for K-1, a next-generation AI factory platform designed around an 800 VDC architecture for the Vera Rubin era. The design uses cluster-level power centers to centralize AC-to-DC conversion, simplifying power delivery for high-density infrastructure.

NVIDIA and Cisco announced the Ultra-Dense Small Form Factor Pluggable (USFP) standard, the industry’s first module for 400Gb/s-based networking fabrics. USFP provides four times the bandwidth density of traditional 200Gb/s-per-lane optics and roughly 1.3x greater power efficiency. Furthermore, NVIDIA and Google announced Falcon Transport support on future NVIDIA ConnectX-10 SuperNICs and BlueField DPUs, improving network efficiency for large-scale AI infrastructure.