AI 데이터 폭증 대응…빅테크 5사, 차세대 네트워킹 기술 공동 개발

OpenAI warns that network congestion and link/equipment failures increasingly disrupt large-scale ML training—causing delays, jitter, and occasional pauses that can force restarts from checkpoints and waste GPU/time—and emphasizes preventing single-link failures as these workloads can act as "fault amplifiers"; the MRC project addressing this was led by OpenAI with technical contributions from AMD, Broadcom, Intel, Microsoft, and NVIDIA and is coordinated by the Open Compute Project.