In short, at the scale any block service is intended to serve, HDDs/SSDs are going to fail, networks are going to be down, data corruptions are going to happen, unpredictable latencies are going to appear, etc. etc. All these failures are destined to happen at a not so infrequent rate. The service offers the abstraction/illusion of "HDD in the cloud that never fails (or may 9s SLA)" despite all that is pretty hard to pull off.