How OpenAI delivers low-latency voice AI at scale

OpenAI describes how it rearchitected its WebRTC stack to deliver low-latency voice interactions at scale for ChatGPT voice and the Realtime API. The team addressed three infrastructure constraints: one-port-per-session media termination, stateful ICE/DTLS session ownership, and global routing latency. They implemented a split relay plus transceiver architecture that preserves standard WebRTC client behavior while optimizing internal packet routing.