Markdown edition

# NVIDIA's OCP AI factory push spans MGX racks, DSX code and a facility spec

> NVIDIA presented MGX 1.2 rack designs and four DSX software repositories at OCP. The cited facility specification predates the summit; partner designs and hardware support remain plans.

By BIG CHANGE Editorial

Published: 2026-10-13T03:25:18.604Z
Updated: 2026-10-13T03:25:18.604Z
Canonical: https://bigchange.ai/blog/nvidia-ocp-ai-factory-mgx-dsx-facility-spec

![Illustrative charcoal view down a broad aisle between two rows of generic data-center cabinets, with overhead cable trays and pipes.](https://bigchange.ai/api/media/file/nvidia-ocp-facility-design-hero-v1.png)
AI-generated conceptual illustration by BIG CHANGE. Generic facility design scene; no actual NVIDIA/OCP installation or tested hardware is depicted.

NVIDIA used the Open Compute Project (OCP) Global Summit on October 12 to present a set of AI data center designs and software contributions. For operators, the useful distinction is between a facility specification they can read, code repositories they can inspect, and rack designs NVIDIA says it has contributed. The announcement also contains partner plans and future hardware that are at a different stage.

The timing of one prominent document needs care. NVIDIA described the [Open Data Center specification](https://www.opencompute.org/wp-content/uploads/2026/10/Open-Data-center-Spec-Revision-0.7.0-4.pdf) alongside the summit news, but Revision 0.7.0 is marked effective **August 27, 2026**. Its version history records collaborative work by AMD, Google, Meta and NVIDIA. October 12 was the date NVIDIA drew attention to that work, not the effective date printed on the specification.

## The big change

NVIDIA's [OCP announcement](https://blogs.nvidia.com/blog/ocp-global-summit-vera-rubin-scale-ai-factories/) brings together three different layers of an AI facility: an updated MGX rack architecture, public DSX software repositories, and an OCP facility design specification. The combination gives builders documents and code to assess against their own requirements. It does not establish that any particular Vera Rubin installation is running, or that the announced partner designs and future networking products are available to buy.

## A facility document with measurable design targets

OCP's [Open Data Center: Next-Generation ML Infrastructure Design Principles, Revision 0.7.0](https://www.opencompute.org/wp-content/uploads/2026/10/Open-Data-center-Spec-Revision-0.7.0-4.pdf) is a facility-level design specification aimed at 2027–2028 AI and machine learning systems. It covers structural limits, row clearances, power density and cooling. For example, its Table 1 lists a 100-inch rack height target and an 82-inch minimum cold aisle; Table 2 sets a 12 MW minimum cluster-size baseline, with 24 MW preferred. These are design targets in a specification, not measurements of a built site.

Its version history is also instructive. The 0.7.0 revision broadens earlier Google-specific guidance into a document authored by four companies. A site designer can use the published figures as a common starting point for discussions with facility, power and cooling suppliers. The document itself says implementation risk rests with the user, and its appendix drawings are non-normative examples. Local engineering, safety review and product compatibility still need to be established for a project.

## MGX concerns the rack; DSX concerns operations

At the rack level, NVIDIA says it contributed Vera Rubin rack and compute-tray designs to OCP through its [MGX architecture](https://www.nvidia.com/en-us/data-center/mgx/). Its October announcement describes MGX 1.2 as adding blind-mate liquid cooling, rack-scale interconnect infrastructure, a liquid-cooled 5,000A busbar and native 19-inch equipment support. Those are NVIDIA's descriptions of the design. The OCP facility specification above addresses the data hall around racks; it is not an independent test of MGX hardware or a purchase listing for an MGX 1.2 system.

NVIDIA also linked four public software repositories under its DSX AI Factory effort: [Config Manager](https://github.com/dsx-ai-factory/nv-config-manager) for network configuration, [Cluster Readiness Engine](https://github.com/NVIDIA/cluster-readiness-engine) for cluster validation, [Rack Management Service](https://github.com/dsx-ai-factory/nv-rms) for rack operations, and a [Cluster API provider for NVIDIA Infrastructure Controller](https://github.com/dsx-ai-factory/cluster-api-provider-nico) for bare-metal Kubernetes provisioning. The repositories expose code and documentation to inspect. Config Manager's own README calls it a *Developer Preview* and says it is not recommended for production use. The Cluster API provider's README lists its support level as *Experimental* in the Compatibility section. Public code, by itself, does not verify the deployment times or reliability results claimed in the launch post.

The practical sequence for a data center team is to review the facility requirements in OCP's document, compare rack interfaces against the MGX materials, and evaluate the individual DSX repositories for its own stack. These artifacts operate at different layers; the sources do not establish a tested, interchangeable package spanning all three.

## Plans remain plans

The same [NVIDIA announcement](https://blogs.nvidia.com/blog/ocp-global-summit-vera-rubin-scale-ai-factories/) says Visionbay unveiled *plans* for an 800V DC K-1 facility design. It says NVIDIA and Cisco announced *plans* for a USFP optical module standard, and it places Falcon Transport support on *future* ConnectX-10 SuperNICs and BlueField DPUs. Those statements describe proposed designs and future support, not verified deployments or current product availability.

For procurement, the immediate value is a clearer set of documents to request and examine: OCP's versioned facility specification, NVIDIA's stated MGX rack interfaces, and the four software repositories. A decision about a real installation still depends on the applicable hardware specifications, supplier commitments, interoperability work and site engineering. BIG CHANGE reviewed the published documents and repositories; it did not test hardware or deploy the software.

## Sources & further reading

- [NVIDIA's OCP Global Summit announcement, October 12, 2026](https://blogs.nvidia.com/blog/ocp-global-summit-vera-rubin-scale-ai-factories/) describes the MGX 1.2 contribution, DSX repositories and partner announcements. Product capabilities and deployment benefits in this vendor post are attributed claims, not independent measurements.
- [OCP Open Data Center specification, Revision 0.7.0](https://www.opencompute.org/wp-content/uploads/2026/10/Open-Data-center-Spec-Revision-0.7.0-4.pdf) is the original facility document. Its cover gives an August 27, 2026 effective date; Sections 4–6 contain physical, power and cooling targets. Its appendix is explicitly non-normative.
- [NVIDIA Config Manager repository](https://github.com/dsx-ai-factory/nv-config-manager), [Cluster Readiness Engine repository](https://github.com/NVIDIA/cluster-readiness-engine), [Rack Management Service repository](https://github.com/dsx-ai-factory/nv-rms), and [Cluster API provider repository](https://github.com/dsx-ai-factory/cluster-api-provider-nico) expose the software NVIDIA linked. Their READMEs explain the components and prerequisites. Config Manager identifies its Developer Preview status; the Cluster API provider marks its support level Experimental.

## Sources

- [NVIDIA at OCP Global Summit, October 12, 2026](https://blogs.nvidia.com/blog/ocp-global-summit-vera-rubin-scale-ai-factories/) — Vendor account of MGX 1.2, DSX repositories and partner plans. Performance and deployment language is attributed to NVIDIA.
- [OCP Open Data Center specification, Revision 0.7.0](https://www.opencompute.org/wp-content/uploads/2026/10/Open-Data-center-Spec-Revision-0.7.0-4.pdf) — Primary facility specification, effective August 27, 2026; Sections 4–6 contain design targets and its appendix is non-normative.
- [NVIDIA Config Manager repository](https://github.com/dsx-ai-factory/nv-config-manager) — Public source and README; identifies Developer Preview status and advises against production use.
- [NVIDIA Cluster Readiness Engine repository](https://github.com/NVIDIA/cluster-readiness-engine) — Public source and README describing cluster validation.
- [NVIDIA Rack Management Service repository](https://github.com/dsx-ai-factory/nv-rms) — Public source and README describing rack-level operations.
- [Cluster API provider for NVIDIA Infrastructure Controller](https://github.com/dsx-ai-factory/cluster-api-provider-nico) — Public source and README for bare-metal Kubernetes provisioning.