The industry generally regards three availability zones as the baseline for production-grade deployment: when an application is redundantly deployed across three availability zones, it can both fail over automatically on a single point of failure and maintain acceptable replication latency across the zones.