Nvidia said on Monday that Google and Meta — two of its largest customers and its most prominent rivals in custom AI silicon — will adopt its networking hardware, in a series of announcements made at the Open Compute Project (OCP) Global Summit in San Jose. The OCP, founded by Facebook — now Meta — in 2011, is the industry group that publishes open standards for data center hardware so that equipment from different vendors can work together.

Google and Meta embrace Nvidia networking

The announcement comes despite both companies spending the past year working to loosen Nvidia's grip on their AI infrastructure. Meta signed a multi-billion-dollar deal to rent Google's Tensor Processing Units (TPUs) and has shipped its own MTIA chip designs, while Google now sells TPUs to outside customers. Networking sits at a different layer of the stack: SuperNICs, switches, and data processing units move data between the servers and racks inside AI data centers, so the interconnect serves a cluster's processors no matter which company designed them.

According to Nvidia's blog post, Meta is working with Nvidia on open AI networking built around the Spectrum-X Ethernet SuperNIC and Spectrum-6 switch chips. Google, separately, will support Falcon Transport on Nvidia's upcoming ConnectX-10 SuperNICs and BlueField data processing units.

Microsoft, meanwhile, showed off its own Maia 300 inference chip at the summit as it aims to reduce its reliance on Nvidia and AMD. Even so, Cryptopolitan reports that hyperscalers such as Microsoft account for more than half of Nvidia's quarterly revenue — a reminder that its biggest customers and its chip rivals are largely the same companies.

The networking announcements build on Nvidia's recent hardware momentum. The company began shipping Vera Rubin, its newest data center platform, in August after announcing it in March. Supermicro said on September 23 that it was shipping Vera Rubin racks, and CoreWeave said a week later that a customer was already running production workloads on the system. Nvidia also shared its rack and compute-tray designs with the Open Compute Project, allowing other companies to build compatible hardware.

Specifications and optics

Among the other announcements, Nvidia released the Open Data Center Specification v0.7, a shared blueprint for how AI factories should handle power, cooling, and layout. AMD, Google, and Meta joined Nvidia in publishing the specification. Published openly through the OCP, the blueprint gives other equipment makers a common reference for building AI data center hardware.

A second document, the OCP Grid Resilience Specification, was developed by Nvidia together with Google, Meta, and Microsoft. It sets out how large AI sites should ride out power grid disturbances and return to near-full load in under two seconds.

Nvidia also announced the Ultra-Dense Small Form Factor Pluggable (USFP) module alongside Cisco, as detailed in the companies' announcement. Pluggable optics are the fiber-optic transceiver modules that connect switches and servers inside a data center, and the USFP design runs at 400 gigabits per second per lane and reaches 12.8 terabits per second in total. According to the companies, it delivers four times the bandwidth density and roughly 1.3 times the power efficiency of 200-gigabit optics, with TE Connectivity and Amphenol working on the components.

Revenue backdrop

The announcements come as Nvidia's business continues to expand rapidly. The company's revenue more than doubled in its fiscal second quarter (May to July 2026), rising 106% from a year earlier to $96.2 billion, with its data center business accounting for $89 billion of the total — roughly 92% of revenue. Management predicts the coming fiscal year will be supply-constrained.