Google and Meta are adopting Nvidia's artificial intelligence (AI) networking hardware. The two companies had sought to reduce reliance on Nvidia by developing their own AI chips, but are cooperating with Nvidia in networking.

On Oct. 12, local time, blockchain outlet Cryptopolitan reported that Nvidia disclosed the plan at the Open Compute Project (OCP) Global Summit in San Jose, California.

Meta will develop open AI networking technology based on Nvidia's Spectrum-X Ethernet SuperNIC and Spectrum-6 switch chips. Google plans to support Falcon Transport technology on Nvidia's next-generation ConnectX-10 SuperNIC and BlueField data processing unit (DPU).

The two companies have shown moves over the past year to reduce reliance on Nvidia. Meta signed a multibillion-dollar contract to rent Google's tensor processing units (TPUs), and developed and launched its own AI chip, MTIA. Google is also selling TPUs to external customers.

Microsoft (MS) unveiled its own inference AI chip, Maia 300, at the event, showing a move to reduce reliance on Nvidia and AMD. Still, Cryptopolitan said large cloud companies including MS account for more than half of Nvidia's quarterly revenue.

Nvidia also shared rack and compute tray designs with OCP. It said this was to allow other companies to build hardware compatible with the designs.

Nvidia also unveiled "Open Data Center Specifications v0.7" covering power supply, cooling and facility layout standards. AMD, Google and Meta participated in the joint announcement. It also announced the "OCP Grid Resilience Specification" developed with Google, Meta and MS. The specification sets standards for large AI data centres to respond to grid disruptions and return to near-maximum load within 2 seconds.

With Cisco, it announced ultra-high-density ultra small form-factor pluggable (USFP) optical modules. The product supports a transmission speed of 400 gigabits per second (Gbps) per lane, with total bandwidth reaching 12.8 terabits per second (Tbps). It said bandwidth density is 4 times higher and power efficiency about 1.3 times higher than existing 200 Gbps optical technology. TE Connectivity and Amphenol are also participating in component development.

Nvidia began shipping its next-generation AI platform Vera Rubin, unveiled in March, from August. Supermicro announced shipments of Vera Rubin racks on Sept. 23, and CoreWeave said a week later that customers were running real operational workloads on the system.

Nvidia's fiscal 2026 second-quarter (May to July) revenue rose 106 percent from a year earlier to $96.2 billion. Revenue from its data centre business was $89.0 billion, accounting for most of the total. Nvidia executives forecast supply constraints would continue into the next fiscal year.