OpenAI Introduces MRC (Multipath Reliable Connection): A New Open Networking Protocol for Large-Scale AI Supercomputer Training Clusters
OpenAI just solved the networking bottleneck that was supposed to limit supercomputers to 50,000 GPUs. MRC scales to 100,000+ on two-tier Ethernet.
Why it matters
OpenAI's new MRC protocol removes a critical infrastructure constraint for AI training at scale—network resilience and topology efficiency. This directly impacts how fast companies can build and train on massive clusters, affecting capex decisions and training timelines across the industry.
The key facts
6 to knowMRC enables supercomputers with 100,000+ GPUs using only 2-tier Ethernet switches
Multipath packet spreading across hundreds of paths simultaneously
Microsecond-level network failure recovery
Partnership: OpenAI, AMD, Broadcom, Intel, Microsoft, NVIDIA
Open protocol (industry-wide adoption potential)
Published May 7, 2026
Go to the source
MarkTechPostmarktechpost.com
Publisher excerpt: MRC (Multipath Reliable Connection) is a new open networking protocol developed by OpenAI in partnership with AMD, Broadcom, Intel, Microsoft, and NVIDIA that improves GPU networking performance and resilience in large-scale AI training clusters by spreading packets across hundreds of paths…