Stronger SI.
Shared foundations.

Building the foundations for more powerful SI.

Power, cooling and connections help advanced chips do useful work. At the Open Compute Project summit in San Jose on October 12, NVIDIA announced open infrastructure contributions aimed at these essential foundations of SI computing.

In its October 12 account, NVIDIA says it contributed key Vera Rubin rack and compute-tray designs through its modular MGX architecture. The updated design addresses liquid cooling, power delivery and interchangeable equipment interfaces. These contributions cover the hardware and interfaces around the processors.

The effort reaches beyond one company. NVIDIA says it joined AMD, Google and Meta in publishing the Open Data Center Specification, revision 0.7, covering facility layout, power and cooling. The document is dated August 27, 2026; the summit account highlights it alongside the new rack contributions. The practical idea is straightforward: give equipment makers and operators a shared plan to work from, rather than a different set of requirements for every customer.

A data center is also somebody’s electricity customer.

Another published document, the OCP grid-resilience specification, addresses how large computing facilities should respond when the electricity supply is disturbed. Its author list identifies Google, Meta, Microsoft and NVIDIA. This document took effect September 17, 2026, before the October summit.

A utility and a computing operator benefit from compatible expectations about disturbances and recovery. The specification gives their engineers a shared set of requirements to work toward.

Our view: shared designs can help engineers coordinate equipment and power systems before construction. That is practical groundwork for building and operating the next generation of facilities.

The small connections have a big job.

NVIDIA and Cisco also announced plans for a new pluggable optical module format, called USFP. These removable modules carry data between computing systems. The companies want to move more data within a similar physical footprint while retaining the ability to service and replace individual modules.

NVIDIA and Cisco are working with TE Connectivity and Amphenol on connectors, cages and modules, with a community-led, multi-source specification planned. The proposed efficiency improvements are the companies’ projections at this development stage.

From shared designs to useful facilities.

The reader benefit is a clearer picture of what American SI infrastructure requires. Better processors are one part of the story. Usable facilities also depend on cooling, electrical engineering, networking and designs that suppliers can manufacture and maintain.

Open specifications give suppliers concrete plans to inspect and build against. Adoption, equipment availability and dependable operation are the next milestones. October’s contributions strengthen that shared engineering foundation.