**[OpenAI Leads AMD, NVIDIA, Intel, Microsoft, and Broadcom in Rare Collaboration to Tackle AI Network Layer]** Golden Finance reported on May 7, citing Fast Technology, that OpenAI officially released the MRC (Multipath Reliable Connection) protocol today through the Open Compute Project (OCP), addressing GPU network communication bottlenecks in large-scale AI training. The protocol was jointly developed by OpenAI, AMD, NVIDIA, Intel, Microsoft, and Broadcom over two years and is now in practical use within supercomputing clusters equipped with NVIDIA GB200. The core issue MRC aims to solve is: during the training of large-scale AI models, even a single instance of data transmission delay can interrupt the entire training process, causing GPUs to idle collectively while waiting. The larger the cluster, the more frequent the delay issues caused by network congestion, link failures, and device malfunctions. MRC's solution is to split a single 800Gb/s network interface into multiple smaller links—for example, connecting one interface to eight different switches to construct eight independent 100Gb/s parallel networks, rather than relying on a single 800Gb/s network.