Responses are served from the node nearest your user, so cold starts stop being a tail-latency problem.