Big Tech Five Join Forces to Develop Next-Generation Networking Technology to Handle AI Data Surge

▤ Summary

OpenAI, working with AMD, Broadcom, Intel, Microsoft and NVIDIA, released a new networking protocol called MRC (Multipath Reliable Connection) to address congestion in massive GPU training clusters. MRC spreads traffic across hundreds of network paths to avoid bottlenecks, reduce jitter and keep large-scale training jobs running without interruption. The specification and accompanying research were published through the Open Compute Project, and NVIDIA says parts of the technology are already running in major AI training clusters using its Spectrum-X Ethernet. The design also promises simpler network topologies that can connect 100,000+ GPUs with fewer switch tiers while giving operators a single dashboard to monitor and control traffic.

▦ Linked Companies

Broadcom Inc.

$50M to $100M

NVIDIA Corporation

$1B+

AMD

$25M to $50M

Intel Corporation

$1B+

OpenAI

$25M to $50M