There are two shapes of backup storage on the market and they are priced on different axes. One sells you a fixed quota for a fixed monthly figure. The other sells you per-gigabyte elasticity with the transfer metered separately. Both are legitimate. They suit different operations, and picking on headline price alone is how teams end up with the wrong one.
This is a comparison of backup storage cost between the two models, with the arithmetic written out and the assumptions named.
The two shapes
Flat quota. You buy a size. You pay for that size whether it is empty or full. There is no per-request accounting and no separate transfer charge. NexonHost's backup storage is this shape: €5 per TB per month at every tier from 1 TB to 600 TB, no setup fee, delivered as a chrooted SFTP account over SSH.
Metered object storage. You pay for bytes stored, and transfer and API calls are accounted separately — sometimes generously, sometimes not. Backblaze B2, as published in September 2026, is $6.95 per TB per month with free egress up to three times your average monthly stored volume and $0.01 per GB beyond that; its Class A, B and C API transactions are free. AWS S3 gives the first 100 GB per month of internet egress free, aggregated across services and regions, and meters beyond it.
The important correction to a claim you will see repeated: not all object storage punishes restores. B2's 3× allowance means an ordinary restore is usually free. The generalisation "object storage will bankrupt you on egress" is a decade out of date at the value end of the market, and it is worth saying so before comparing.
Worked example: 10 TB, restored twice a year
Assume 10 TB of backup data, growing slowly, with two full restores a year of 3 TB each.
| Model | Monthly storage | Restore cost | Notes |
|---|---|---|---|
| Flat quota at €5/TB | €50 | €0 | Quota is the whole invoice |
| B2-style, $6.95/TB, 3× free egress | $69.50 | $0 | 3 TB is well inside a 30 TB allowance |
| Metered, $0.01/GB, no free allowance | varies | $30 per restore | Small, but per event |
| Hyperscaler with tiered egress | varies | Meaningful above 100 GB/month | The tail you cannot forecast |
At this size and this restore pattern, the first two rows are close enough that the decision is not really about money. The invoice is €50 or roughly $70 a month, both are predictable, and both survive a restore without drama.
Where they separate is at the edges: an operation that restores constantly — for staging refreshes, for test data, for a CI pipeline that pulls a production dump nightly — can exceed a 3× multiple without noticing, and the flat quota simply does not have that failure mode. Conversely, an operation whose backups are 200 GB does not want to buy a terabyte.
Costs that appear in neither pricing page
Currency and term. One of these is in euros and one in dollars. Compare on the day you buy, not on the day you read a blog post.
The empty quota. Flat quotas charge for space you have not filled. Buying 10 TB for 3 TB of data because you might grow is a real overspend that a per-GB model would not have charged.
Minimum retention periods. Some object tiers bill a minimum duration per object — delete after three days and you still pay for thirty. This quietly destroys the economics of short-retention incremental chains. Check it before choosing an archive tier.
Retrieval latency tiers. The cheapest per-GB storage is usually cold, with a restore that begins in hours rather than seconds. That is not a cost, it is a recovery-time objective, and it belongs in the same conversation.
Engineering time. An SFTP target works with rsync, Restic, Borg, Duplicity, rclone, Veeam, Proxmox Backup Server and a NAS out of the box. An S3 target works with the same list plus anything cloud-native. Both are fine; the cost is zero for whichever one your existing jobs already speak, and a day of work for the other.
Which model fits which operation
| Your situation | The shape that fits |
|---|---|
| Predictable dataset, want a fixed invoice | Flat quota |
| Backups already run over SSH (Restic, Borg, rsync) | Flat quota |
| Frequent restores or staging refreshes | Flat quota |
| Small or wildly variable dataset | Metered object storage |
| Tooling is S3-native and you want versioning and object lock | Metered object storage |
| Serving files to users, not just backing up | Object storage plus a CDN |
| Finance wants one line item that never moves | Flat quota |
That last row is not a joke. A backup that generates a variable invoice generates a monthly conversation, and the version of this decision that survives contact with a finance team is often the boring one.
The comparison that actually matters more than price
Whichever shape you choose, three properties decide whether the backup works when you need it:
- Separate credentials. The key that reaches the backup target must not be the key sitting on the production server. Otherwise the copy dies with the machine in exactly the compromise scenario it exists for.
- Not a mounted writable share. A permanently mounted network drive is inside the blast radius of ransomware. Push to the target; do not mount it.
- A restore you have actually performed. Time it, on a scratch machine, quarterly. That measurement is your real recovery-time objective, and it is usually worse than the guess.
Price differences of 30% between providers are irrelevant next to any one of these being wrong.
Where the working copy belongs
A point that gets muddled: backup storage is not primary storage. If the live dataset is large — media, archives, log retention, surveillance footage — that belongs on a storage server or a storage VPS, with the backup quota holding a copy of it. Both shapes of backup storage are tuned for sequential bulk transfer, not for random access under load, and running a workload directly against either will disappoint.
The pattern that works is boring: fast local storage for the working set, a snapshot on the machine for the "undo the last hour" case, and an offsite copy on a schedule with a prune policy. See offsite backup storage sizing for the arithmetic on that third copy, and storage server vs storage VPS for the first one.
The short version
If your backups already speak SSH, your dataset is predictable and you want an invoice that does not move, a flat quota is the simpler instrument and — at €5 per TB with no retrieval line — usually the cheaper one too. If your tooling is S3-native, your data volume swings, or you need object lock and versioning, metered object storage is the right tool and the modern value-tier egress allowances mean restores will not ambush you.
What does not work is choosing on the storage number alone and discovering the transfer terms during an incident. Read both columns before you sign, and test a restore before you need one.
Related reading: offsite backup storage sizing, RAID levels explained, and VPS vs dedicated server.
Sources
- B2 Cloud Storage Pricing and B2 transaction pricing, Backblaze, retrieved 7 September 2026 — $6.95/TB/month, free egress up to 3× average monthly storage then $0.01/GB, Class A/B/C transactions free.
- Amazon S3 Pricing, AWS, retrieved 7 September 2026 — first 100 GB/month of data transfer out to the internet free, aggregated across services and regions.
- Bandwidth billing, DigitalOcean, retrieved 7 September 2026 — $0.01 per GiB additional outbound transfer.
- Plan specifications and pricing as published on nexonhost.com on 7 September 2026; check the product page for current figures.
