NVIDIA said on 12 October, at the Open Compute Project (OCP) Global Summit in San Jose, that it gave OCP its Vera Rubin rack and compute tray designs. They are part of its open MGX design. The updated MGX 1.2 supports racks up to 250kW and a liquid-cooled 5,000A busbar, a thick bar that carries power. NVIDIA also said it joined AMD, Google and Meta to publish Open Data Center Specification v0.7. Google, Meta, Microsoft and NVIDIA published a separate grid resilience spec. Meta is adopting NVIDIA's Spectrum-6 network switches. NVIDIA and Cisco announced USFP, a plug-in network module aimed at 400Gb/s per lane.
NVIDIA's post says USFP is planned, and Falcon support, a Google-developed network protocol, will come on future ConnectX-10 chips. It gives no dates, prices or shipping times for those, or for Visionbay's planned K-1 design. The 1.3x power efficiency gain for USFP is NVIDIA's own claim about a product that does not yet exist. The post does not say how many vendors will build to these specs. All facts here come from NVIDIA's blog.
NVIDIA says open specs let equipment makers build without waiting for each customer's private requirements. For teams that run or buy AI hosting, that could mean more hardware built to shared designs. The post gives no cost or timing figures.