On May 6, 2026, OpenAI, in collaboration with AMD, Broadcom, Intel, Microsoft, and NVIDIA, officially released the "Multi-Path Reliable Connection" (MRC), a new open network protocol. This protocol aims to enhance the speed and reliability of large-scale AI training clusters by reducing GPU idle time and optimizing computational efficiency. MRC employs a multi-plane network design, splitting a single 800Gb/s interface into multiple smaller links, allowing the system to connect approximately 131,000 GPUs with just two layers of switches, significantly reducing network power consumption and component count. Additionally, MRC introduces adaptive packet spraying technology, which disperses data packets from a single transmission task across hundreds of parallel paths, effectively avoiding core network congestion. Currently, MRC has been fully deployed in all of OpenAI's large supercomputers used for training cutting-edge models.