High availability implementation of AWS NAT instances

50% savings to replace a managed service with a self-managed instance (trust me, even if things are fully automated there will still be gray failures, weird hiccups, congestion events, etc) doesn’t sound like a great idea for most companies.