NVIDIA Opens Vera Rubin Architecture to Help Scale AI Factories With Industry Partners

NVIDIA Tackles AI Factory Scaling Bottlenecks With Open Architecture

NVIDIA announced new partnerships and contributions at the Open Compute Project Global Summit in San Jose, aimed at creating a common foundation for next-generation AI infrastructure1

1

2

2

. The chip giant contributed key Vera Rubin rack and compute tray designs to the Open Compute Project through the NVIDIA MGX architecture, giving system builders common blueprints for developing AI factories2

2

. "The real bottleneck for the industry right now is how fast the world can manufacture and deploy AI factories," said AI Infrastructure Solutions Lead Joe DeLaere2

2

.

Open Standards for AI Infrastructure Drive Industry Collaboration

NVIDIA joined industry leaders in contributing to the third-generation MGX AI data centre designs through OCP's Open DC Initiative1

1

. The company is working with Google, Meta, AMD, and Cisco to establish open standards for AI infrastructure, helping turn proven AI factory designs into industry standards that others can adopt, manufacture, and scale3

3

. NVIDIA, AMD, Google, and Meta published version 0.7 of the Open Data Center Specification, which lays out common requirements for AI facility design, including power architecture, cooling, and physical layout3

3

.

Vera Rubin Architecture Advances Rack-Scale Liquid Cooling

The updated MGX 1.2 specification extends the NVIDIA MGX architecture for AI factories with rack-scale blind-mate liquid cooling, interoperable rack-scale interconnect infrastructure, and a liquid-cooled 5,000A busbar designed for higher-power AI systems1

1

. The architecture supports rack deployments up to 250kW while simplifying serviceability, deployment, and infrastructure reuse1

1

3

3

. The compute tray has no cables, no hoses, and no fans—everything is liquid-cooled and modular, turning hours into minutes for tray assembly and speeding up serviceability and overall rack build times4

4

.

Grid Resilience and Power Efficiency Shape Next-Generation AI Factories

Google, Meta, Microsoft, and NVIDIA published the OCP Grid Resilience Specification, designed to help AI facilities respond to utility grid disruptions and return to near-full load in less than two seconds3

3

. Visionbay unveiled plans for K-1, a next-generation AI factory platform designed around an 800V DC architecture and built with future NVIDIA Vera Rubin deployments in mind4

4

. The platform combines NVIDIA's 800V DC reference architecture with power and thermal technologies from Vertiv to create a scalable foundation for high-density AI infrastructure, addressing the challenge of rising rack power densities4

4

.

Vera CPU Rack Design Targets Agentic AI Workloads

The Vera CPU rack design specification extends the NVIDIA MGX architecture beyond accelerated computing, providing a purpose-built platform for the growing agentic AI era1

1

. While Vera Rubin NVL72 serves as the foundation for large-scale training and inference, the NVIDIA Vera CPU rack is designed to support tool calling, orchestration, and sandboxed execution, helping AI agents interact with applications, data, and services in real time1

1

. "Between model calls, agents need to do orchestration. They need to execute code. They need to process data, run evaluations," DeLaera explained2

2

. NVIDIA claims that its Vera CPUs deliver around twice the performance of its previous Grace GPU generation on several workload benchmarks, and a 1.8-times improvement on Terminal-Bench, which measures practical agent tasks involving code containers2

2

.

Meta Adopts Spectrum-6 for Enhanced Optical Networking

Meta is adopting NVIDIA's Spectrum-6 Ethernet switches, with switching capacity of 102.4 terabits per second, for its next-generation AI infrastructure3

3

. NVIDIA said Spectrum-6 offers twice the switching capacity and per-lane bandwidth of its previous Spectrum-4 generation3

3

. Meta and NVIDIA advanced open AI networking with NVIDIA Spectrum-X Ethernet SuperNIC and Spectrum-6 innovations1

1

. With multi-plane networks, when a component fails, the SuperNIC detects that in hardware and reroutes the traffic in less than 3 milliseconds—30 times faster than software-based alternatives4

4

.

NVIDIA and Cisco Define USFP Standard for 400Gb/s Networking

NVIDIA and Cisco announced the Ultra-Dense Small Form Factor Pluggable (USFP) standard, the industry's first module for 400Gb/s-based networking fabrics1

1

. According to NVIDIA, the USFP optical module design targets 400 Gbps per lane and four times the bandwidth density of traditional pluggable optics operating at 200 Gbps per lane3

3

. The USFP design will offer four times the bandwidth density of the existing 200G OSFP form factor, with NVIDIA claiming a 1.3-times improvement in power efficiency2

2

. This allows data center scalability to expand networks without larger switching systems, as more networking capacity can fit into the same physical footprint at lower power2

2

.

Google's Falcon Transport Integrates With NVIDIA Networking Ecosystem

NVIDIA and Google announced Falcon Transport support on future NVIDIA ConnectX-10 SuperNICs and BlueField DPUs1

1

. NVIDIA is adding support for Google's Falcon Transport to its Spectrum-X Ethernet networking ecosystem, beginning with the company's future ConnectX-10 SuperNICs2

2

. Falcon is a transport technology designed to manage data delivery efficiently in large-scale networks, capable of moving traffic over multiple paths, responding to congestion, recovering lost packets, and maintaining secure connections2

2

.