NVIDIA opens AI data centre designs as it pushes for industry-wide standards

NVIDIA is opening up more of its AI infrastructure designs and working with companies including Google, Meta, AMD and Cisco to establish common standards for next-generation AI data centres, the company announced on Monday.

At the Open Compute Project (OCP) Global Summit, NVIDIA said it had contributed its Vera Rubin rack and compute tray designs through its open MGX architecture, giving system manufacturers a common blueprint for building AI infrastructure.

According to NVIDIA, its updated MGX 1.2 specification supports rack deployments of up to 250 kW, with liquid cooling, standardised power and cooling interfaces, and higher-density rack designs. The company also introduced a Vera CPU rack design for AI agent workloads, including tool calling and orchestration.


NVIDIA, AMD, Google and Meta publish open standards

NVIDIA said it had joined AMD, Google and Meta in publishing version 0.7 of the Open Data Center Specification, which lays out common requirements for AI facility design, including power architecture, cooling and physical layout.

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

Also Read: AI models may soon self-replicate without access to data centers, Anthropic warns

The specifications are intended to give data centre operators and equipment manufacturers common design standards as they build infrastructure for increasingly power-intensive AI workloads, the company said.

Meta adopts NVIDIA networking technology

Meta is adopting NVIDIA’s Spectrum-6 Ethernet switches, with switching capacity of 102.4 terabits per second, for its next generation AI infrastructure. NVIDIA said Spectrum-6 offers twice the switching capacity and per-lane bandwidth of its previous Spectrum-4 generation.

The companies are also advancing Spectrum-X Multiplane technology, which distributes GPU traffic across up to eight network planes. NVIDIA said the technology can help AI factories scale beyond 100,000 GPUs while improving network utilisation and resilience.

In a separate announcement, NVIDIA and Cisco unveiled the Ultra-Dense Small Form Factor Pluggable (USFP) standard for optical networking. According to NVIDIA, the design targets 400 Gbps per lane and four times the bandwidth density of traditional pluggable optics operating at 200 Gbps per lane.