Summary
OpenAI published Multipath Reliable Connection, or MRC, as an open networking protocol through the Open Compute Project. The company said MRC is already deployed across its largest NVIDIA GB200 supercomputers and is designed to reduce congestion and failure impact in very large training clusters.
What changed
OpenAI released the MRC networking protocol and described its use in production across large frontier-training clusters.
Why it matters
Infrastructure bottlenecks are becoming a differentiator in frontier model development. By opening a networking protocol tied to large-scale training reliability, OpenAI is signaling that network design is now a first-order part of model capability and compute efficiency.
Evidence excerpt
OpenAI said MRC is already deployed across its largest GB200 supercomputers and released the specification through the Open Compute Project.