OpenAI explained how it rebuilt WebRTC for low-latency voice AI

OpenAI described how it rewrote the WebRTC stack for ChatGPT Voice and the Realtime API so conversations with AI stay smooth, without pauses or stutters, even under global load. The company separated routing from session termination, used a relay + transceiver pairing, and embedded a hint for the first packet in the ICE ufrag. That let it scale voice AI without huge ranges of open UDP ports and with lower network latency.