CONFIRMED: NVIDIA set out these announcements in a company blog post on 12 October 2026, at the Open Compute Project (OCP) Global Summit in San Jose. Descriptions of the specifications and their benefits are NVIDIA’s and its partners’.
NVIDIA said it “contributed key NVIDIA Vera Rubin rack and compute tray designs to OCP through the open NVIDIA MGX architecture” [1]. OCP is an industry group that shares open hardware designs for data centres; MGX is NVIDIA’s modular design for servers and racks. Vera Rubin is NVIDIA’s next AI computing platform.
MGX 1.2: racks up to 250kW
NVIDIA said the updated MGX 1.2 specification adds “rack-scale blind-mate liquid cooling, interoperable rack-scale interconnect infrastructure and a liquid-cooled 5,000A busbar designed for higher-power AI systems” [1]. A busbar is a thick metal bar that carries power through the rack; blind-mate means parts connect automatically when slid into place. NVIDIA said MGX 1.2 also adds “native 19-inch equipment support” and “fanless liquid-cooled compute trays”, and “supports rack deployments up to 250kW” [1].
A rack design for Vera CPUs
NVIDIA also published a Vera CPU rack design specification. It said the rack “is designed to support tool calling, orchestration and sandboxed execution, helping AI agents interact with applications, data and services in real time” [1]. NVIDIA described this as a design specification; the blog does not say Vera CPU racks are shipping.
Open standards for AI data centres and the grid
- Open Data Center Specification v0.7: NVIDIA said it “joined AMD, Google and Meta to publish” the spec, which “defines a unified blueprint for next-generation AI factory facility design — from power architecture to cooling infrastructure to physical layout” [1].
- OCP Grid Resilience Specification: published by Google, Meta, Microsoft and NVIDIA, it sets out how large AI sites “should behave during utility grid disturbances: staying connected, responding automatically and restoring to near-full load within less than two seconds” [1].
NVIDIA also released open-source software for its DSX AI factory platform: Config Manager, Cluster Readiness Engine, Rack Management Service and a Cluster API Provider for NVIDIA Infrastructure Controller [1].
Meta adopts Spectrum-6
“Meta is adopting NVIDIA Spectrum-6 102.4Tb/s Ethernet switches to build larger, higher-performance NVIDIA Vera Rubin AI factories on open networking,” NVIDIA said [1]. Meta’s Spectrum-6-based switch, N-IceCube, fits into IcePack, which NVIDIA describes as “Meta’s 100T, liquid-cooled rack”, and runs Meta’s open-source network software FBOSS [1].
NVIDIA said its Multiplane technology spreads GPU traffic “across up to eight independent network planes, enabling AI factories to scale beyond 100,000 GPUs with fewer switch tiers” [1]. If one part fails, traffic moves to the remaining planes.
“Extending FBOSS to Spectrum ASICs means we can run that same software stack across our fabrics,” said Omar Baldonado, senior director of data center and AI networking at Meta [1].
Google’s Falcon on ConnectX-10
“NVIDIA and Google also announced Falcon Transport support on future NVIDIA ConnectX-10 SuperNICs and BlueField DPUs,” the blog said [1]. Falcon is “a transport protocol designed to deliver data reliably over Ethernet networks, which can otherwise drop packets under congestion” [1]. NVIDIA said NVIDIA and Google network cards will be able to talk over a common transport. No dates were given.
NVIDIA and Cisco’s new USFP pluggable optics plan is covered separately: NVIDIA and Cisco Plan USFP.
Sources
- NVIDIA, “OCP Global Summit: Amid Ecosystem Adoption, NVIDIA Opens Vera Rubin Architecture to Help Scale AI Factories Worldwide”, NVIDIA blog, 12 October 2026. https://blogs.nvidia.com/blog/ocp-global-summit-vera-rubin-scale-ai-factories/
Related stories
- AM Intelligence’s extra 20,000 Nvidia Rubin GPUs: company plans for India and Malaysia
- NVIDIA GTC 2026: The $1 Trillion Infrastructure Bet Reshaping AI
- Lumentum, Astera Labs and Corning to Show PCIe Running Over Light Inside an AI Rack
- Arista Announces Ethernet Racks to Link Up to 144 AI Chips. Every Figure Is Arista’s Claim.
, a Vera CPU rack spec, and said Meta is adopting Spectrum-6 switches.){kind=link}