PANews reported on May 6th that OpenAI announced a collaboration with AMD , Broadcom , Intel , Microsoft , and NVIDIA to launch a new network protocol, Multipath Reliable Connection ( MRC ), for interconnecting GPUs in large-scale AI training clusters . MRC is based on RoCE and extends SRv6 source routing, splitting single data transmissions and "spraying" them across hundreds of paths, significantly reducing core congestion and bypassing link and switch failures in microseconds. OpenAI has already deployed MRC in the Stargate and Microsoft Fairwater supercomputers , built in collaboration with OCI , connecting over 100,000 GPUs using two-layer switches while reducing power consumption and the number of devices. The MRC specification has been made available to the industry through the Open Compute Project.