There's a formula that I like to use when architecting data storage (such as Veeam repositories): "C+P+F+A=$", where...
C: Capacity - "How much data storage will you need? What are your data compression and data deduplication expectations, if any?"
P: Performance - "What are your storage performance metrics (how fast would you like it to be)?" What type of workload are you architecting for (for example: Small block random reads/random writes, large block sequential writes). Are you more concernd about IOPS or are you more concerned about throughput(MB/sec)?
Note: Backup and archive workloads are most often sequential, but can, at times, be random.
F: Features - "What do you plan to do with your data (snapshots, replication, etc...)? Do you want your data to be immutable? What about tiering your data to less expensive disk over time? What about data archival?"
A: Availability - "How important is the availability and continuity of the data to you?"
$: Investment - "How much will you need to invest in data storage?"
C: Let's say that based on various Veeam sizing exercises and leveraging built in Veeam deduplication and compression features you determine that you will need to have at least 100TB of usable capacity in your on-site Performance Tier.
Consider the impact of trying to pull 100TB of backups from a public clould provider back to on-premise.
Note that backups should focus on maximizing throughput. SureBackup and Instant Recovery should really focus on IOPS. Forward Incremental backups perform only sequential writes to the target repository, however there is a requirement to schedule periodic full backups which will take additional time to create.
[75 IOPS/disk] X [Block Size] x [n disks]/[6x penalty] = [Backend Throughput]
F: Do you plan to immediately copy your Veeam backups to a public cloud provider? Do you plan to archive your Veeam backups to a Capacity Tier in a public cloud provider after XX days and then move to cold storage after YY days? How long do you want the Veeam backups in your Performance Tier to be immutable as part of your Ransomware prevention strategy? Do you plan to test your backups via SureBackup (you should)?"
A: "Do you want to be able to survive a double disk failure without data loss by utilizing RAID 60?"
Example 1: 100-drive Dell XE7100 with an XE7440 compute sled, 100x 18TB drives. 1.8PB RAW, 1.25PB after RAID6, hot spares, file system format. Contact me for an UNOFFICIAL pricing estimate.
Example 2: 90-drive SuperMicro X11 90-bay single node storage server, 90x 18TB drives, 1.6PB RAW, estimate that will be about 1.1PB usable after RAID6, hot spares, and filesystem formats. Contact me for an UNOFFICIAL pricing estimate.
HPE Apollo example: The fastest solution from HPE and Veeam gets even faster - HPE StorageExperts
"Recent lab tests demonstrate how backup performance gets even faster on HPE Apollo as a hardened, immutable Linux repository for Veeam Backup and Replication. According to Veeam technical blogger Chris Donohoe, this means HPE and Veeam customers can have everything they want in a data protection solution – security, performance, and convenience."
Note that most cloud providers charge 2 cents per GB per month which equates to $1.20 per GB over 5 years.