How does cross-region model replication ensure high availability in ChatGPT-scale systems?

Cross-region replication keeps synchronized model copies across multiple data centers to ensure availability during outages.

ChatGPT operates across multiple geographic regions to provide low latency and high availability. Cross-region replication ensures that model weights, configurations, and routing metadata are synchronized across data centers.

Replication strategies include active-active (multiple regions serve traffic simultaneously) and active-passive (fallback regions). Consistency is maintained via periodic checkpoints and versioned deployments.

This architecture ensures resilience against regional outages while balancing latency, cost, and consistency tradeoffs.