OpenAI on Wednesday published a new networking protocol designed to make large-scale AI training clusters faster and more resilient, releasing it as an open specification through the Open Compute Project. The Multipath Reliable Connection protocol, developed in collaboration with Microsoft, AMD, Broadcom, Nvidia, and Intel, addresses two core challenges facing massive GPU clusters: network congestion and hardware failures.
The protocol is already deployed in OpenAI and Microsoft’s largest training facilities, including the Oracle site in Abilene, Texas, and Microsoft’s Fairwater supercomputers, where it has been used to train GPT-5.5 and other models.
How MRC Works
MRC relies on “packet spraying,” scattering data across hundreds of network paths simultaneously to prevent any single link from becoming congested, according to Mark Handley, OpenAI’s networking lead. This approach produces “flatter” networks that consume less power and compute. When paths fail, MRC detects and reroutes traffic in microseconds, allowing training jobs to continue without interruption.
The protocol also pairs with SRv6, or IPv6 Segment Routing, which prescribes exact data paths through the network rather than requiring switches to make routing decisions — reducing energy consumption at the switch level.
“We want to use as much compute as we can get, but also we want to make sure that we’re using it efficiently and effectively, and this is a critical component of that,” Greg Steinbrecher, OpenAI’s workload lead, told The Deep View.
An Open Standard to End Fragmentation
Steinbrecher emphasized that OpenAI is not seeking to differentiate on this technology but rather to move the entire industry past what it considers a legacy bottleneck. “Several players in the industry have their own in-house implementations of protocols … that type of market fragmentation is bad for the networking industry,” he said. “You want everyone’s energy going in one direction and pushing together, and then everyone moves faster as a result.”
The release comes as the AI industry faces mounting pressure on compute resources. By establishing MRC as an open standard, the coalition aims to accelerate adoption of Ethernet-based AI fabrics and reduce reliance on proprietary networking stacks — a shift that could reshape how data centers are built for the next generation of AI training.
