NVIDIA said on 12 October 2026, at the Open Compute Project (OCP) Global Summit in San Jose, that it contributed key Vera Rubin rack and compute tray designs to OCP through its open MGX architecture, and that Meta is adopting its Spectrum-6 102.4Tb/s Ethernet switches to build larger Vera Rubin AI factories.
NVIDIA also announced USFP, a pluggable optics form factor with Cisco, and Falcon Transport support with Google on future ConnectX-10 SuperNICs and BlueField DPUs.
Business Pill · SWITCHING COSTS
A one-minute explainer of switching costs: what it costs a buyer to move from one supplier to another. It teaches the general idea only and says nothing about any organisation in this story.
The key insight: As we read it, NVIDIA used OCP to open the rack, tray and facility designs around Vera Rubin, named Meta as an adopter of its Spectrum-6 switches, and announced optics and transport work with Cisco and Google.
NVIDIA says it joined industry leaders in contributing to the third-generation MGX AI data center reference design through OCP’s Open DC Initiative. For Vera Rubin AI factories, it says 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.
According to NVIDIA, MGX 1.2 also adds native 19-inch equipment support, higher-density rack designs, configurable I/O bays and fanless liquid-cooled compute trays. It says the architecture supports rack deployments up to 250kW.
NVIDIA says a separate Vera CPU rack design specification is meant to support tool calling, orchestration and sandboxed execution for AI agents, while Vera Rubin NVL72 remains the foundation for large-scale training and inference.
The companies NVIDIA names in its 12 October 2026 OCP Global Summit blog and its USFP release with Cisco: Meta, Google, Cisco, Microsoft, AMD, Visionbay, Vertiv, TE Connectivity and Amphenol. Meta is highlighted as the company NVIDIA names as adopting its switch hardware, Spectrum-6 Ethernet switches.
Meta Adopts Spectrum-6
NVIDIA says Spectrum-6 delivers twice the switching capacity and per-lane bandwidth of Spectrum-4, with a lossless Ethernet AI fabric at 200G per lane. Meta’s Spectrum-6-based switch, N-IceCube, slots into IcePack, Meta’s liquid-cooled rack, and runs FBOSS, Meta’s open source network operating system.
NVIDIA says its Spectrum-X Multiplane technology spreads GPU traffic in hardware across up to eight independent network planes, so AI factories can scale beyond 100,000 GPUs with fewer switch tiers.
Omar Baldonado, senior director of data center and AI networking at Meta, said: “Extending FBOSS to Spectrum ASICs means we can run that same software stack across our fabrics.”
What NVIDIA Said On Stage
NVIDIA’s Ian Buck at the OCP Global Summit on assembly time, with the slide comparing a 90-minute GB200 NVL72 tray build with one minute for Vera Rubin: โIt takes 1 minute to assemble and manufacture that compute tray … literally producing thousands of racks a week at this point.โ These are NVIDIA’s figures.
Buck on Vera Rubin NVL72, with the slide citing SemiAnalysis AgentX on DeepSeek-V4-Pro: โYou get 2x, 10x, at the extreme 30x more token performance experienced by the user … reduce the cost of those tokens by 45x … Vera Rubin is in full production.โ These are NVIDIA’s claims.
Buck previewing Vera Rubin Ultra NVL576: โOne giant 576 GPUs, all connected with NVLink … designed to take us to 10 trillion parameter models to enjoy when we get there next year.โ That timing is NVIDIA’s roadmap, not a shipped product.
USFP: Pluggable Optics With Cisco
NVIDIA and Cisco announced plans for the Ultra-Dense Small Form Factor Pluggable (USFP) module, which they say is designed to be the first to support 400Gb/s-per-lane electrical signals, with capacity up to 12.8Tb/s.
The release says the first planned USFP transceivers will pack 16 full-duplex lanes in an OSFP-sized footprint, with room to expand to 32 lanes. Against 200Gb/s-per-lane pluggable modules, it claims 4x the bandwidth density and approximately 1.3x the power efficiency.
NVIDIA and Cisco say they are working with TE Connectivity and Amphenol on USFP cages, connectors and modules, and plan an open industry specification through a community-led multi-source project.
Falcon, Standards and Software
NVIDIA says future ConnectX-10 SuperNICs and BlueField DPUs will support Google’s Falcon Transport, so NVIDIA and Google NICs can communicate over a common transport while applications keep standard RDMA programming interfaces.
NVIDIA also says it joined AMD, Google and Meta to publish the Open Data Center Specification v0.7 for AI factory facility design, and that Google, Meta, Microsoft and NVIDIA published an OCP Grid Resilience Specification covering how AI sites should behave during grid disturbances, including restoring to near-full load within less than two seconds.
It added open source DSX OS components: Config Manager, Cluster Readiness Engine, Rack Management Service and a Cluster API Provider that it says turns raw hardware into production-ready clusters in hours instead of days. NVIDIA says Visionbay unveiled plans for K-1, an 800 VDC AI factory platform combining NVIDIA’s 800 VDC reference architecture with Vertiv technology.
As we read it, NVIDIA is standardising the layer around its chips: it says MGX 1.2 covers liquid cooling, a 5,000A busbar, compute trays and rack interconnect, and that the Open Data Center Specification covers facility design from power architecture to cooling.
As we read it, the networking news points the same way: NVIDIA says Meta runs its own FBOSS operating system on Spectrum-6 switches, and that NVIDIA and Google NICs can talk over Falcon while applications keep standard RDMA interfaces.
As we read it, USFP keeps optics pluggable: NVIDIA and Cisco say it extends a deployment model proven across multiple generations of data center networks, with TE Connectivity and Amphenol building parts and an open specification planned.
Gilad Shainer, senior vice president, NVIDIA, 12 October 2026
“Building the worldโs largest AI factories starts with rethinking the smallest of connections.”
Three Implications
SYSTEM BUILDERS NVIDIA says the MGX contribution gives system builders a common foundation, and that MGX 1.2 supports rack deployments up to 250kW with fanless liquid-cooled compute trays.
HYPERSCALE NETWORKS NVIDIA says Meta is adopting Spectrum-6 102.4Tb/s switches, and that Multiplane lets AI factories scale beyond 100,000 GPUs with fewer switch tiers.
OPTICS SUPPLIERS NVIDIA and Cisco say TE Connectivity and Amphenol are building USFP cages, connectors and modules, with additional partners expected to join.
The framework puts it this way: “Close the SCARCE layer (keep proprietary) + Open the ABUNDANT layer (commoditize).”
As we read it, what NVIDIA opened at OCP is the rack, tray, facility and optics layer around its Vera Rubin systems, and NVIDIA says these open standards give manufacturers a common target to build for.
The multiples NVIDIA and Cisco state for USFP in their 12 October 2026 release, against 200Gb/s-per-lane pluggable modules: double the lane count and the bandwidth per lane, 4x the bandwidth density and approximately 1.3x the power efficiency. These are the companies’ claims for a planned form factor.
What Is Not Established
NVIDIA gives no ship dates or prices for USFP, and calls it a plan; the 4x density and 1.3x efficiency figures are the companies’ own claims. NVIDIA does not say how many Spectrum-6 switches Meta will deploy or when. The Open Data Center Specification is at version 0.7, and which system builders will adopt MGX 1.2 is not stated.
Business Engineer Framework
The Open vs Closed Meta-Framework
A Business Engineer framework on which layers of a technology stack to keep proprietary and which to open.
At OCP, NVIDIA put Vera Rubin rack and compute tray designs into the open MGX specification, named Meta as a Spectrum-6 adopter, and announced USFP optics with Cisco and Falcon support with Google. The figures are NVIDIA’s and its partners’ own.
A note on sourcing. On 12 October 2026 we read NVIDIA’s OCP Global Summit blog and its USFP release with Cisco in full. The specifications, multiples and partner statements are the companies’ own, USFP is a planned form factor, and the Open Data Center Specification is at version 0.7. Nothing here is investment advice.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
