Three shapes, three use cases
Block storage for virtual infrastructure, attached over Fibre Channel SAN or iSCSI. Continuous availability here comes from synchronous replication with automatic failover.
HCI or dHCI delivers storage, compute and networking from the same box. The main advantage is integrated management, and the price is that compute and capacity grow together.
Scale-out on software-defined storage for file services over CIFS and NFS, or for object storage. Data can sit across several sites, which is how very high availability figures are reached.
What actually decides it
Rarely capacity. Usually the question of how quickly something has to be back when it is gone, and how many sites are involved in making that true. Both belong at the start rather than in the tender.
All-flash with NVMe is the answer to latency, not to availability. Those two get confused regularly.