I'm a senior network architect among many things, I have CCIE and have built notable sections of the internet. I have zero doubt. sheer distance traveled between all 6 constituent parts are crucial to consider In no particular order those core parts are orchestrating code ( vapi, competitors to vapi, or custom code ) LLM STT TTS telephony integration location of human caller each of these requires data going to and from and each time it does so it must route through the 'orchestrating code' each time data traverses these paths, there is a time cost just for the data to travel to and from let alone the additional processing time ontop of that for each step. Technically speaking, latency is never 100% static its always variable by some number of miliseconds. You want ALL OF THESE to be as close as possible If I setup everything to be colo'd in ashburn virginia USA, but my client wants to have people call from india or australia, then that experience is going to be garbage. One bad link can sour the whole experience. this is the biggest aspect to keep in mind Hope that helps 🙂