What Backup Really Costs on OCI — And the Mistakes That Double the Bill

Backup is the line item nobody defends in a cloud business case and nobody notices growing. It has no owner, no roadmap, and no obvious ceiling — it just accretes, month after month, until somebody runs a cost report and finds that protecting the estate costs a meaningful fraction of running it. This is a numbers-first look at what backup actually costs on OCI, across the whole estate rather than just the databases, and which decisions move the total.

All prices below are Oracle list prices, pay-as-you-go, in euros, as of August 2026. Oracle meters storage per GB; we quote per terabyte throughout, because that’s the unit you actually size backups in. One TB here means 1,000 GB.

Four different things called “backup”

On OCI, “backup” covers four separate services with four separate price tags, and teams routinely budget for one and get billed for all four:

• Databases — RMAN backups going either to an Object Storage bucket or to Autonomous Recovery Service.
• Block and boot volumes — point-in-time snapshots of the volumes behind your VMs, stored in Object Storage.
• File systems — File Storage snapshots, which live inside the file system’s own metered capacity.
• Object data — buckets holding everything else: application exports, logs, on-premises archives.

Two structural facts catch people out immediately. First, OCI has no “back up this VM” button. A compute instance is protected by backing up its boot volume, or by baking a custom image — and custom images cost nothing extra to store, which makes them the cheapest way to keep a known-good build of a server. Second, block and boot volume backups are billed at Object Storage rates but appear under Block Storage in Cost Analysis. That single reporting quirk is why so many teams can’t find their backup spend when they go looking for it.

Databases: the number most budgets get wrong

Ask an architect what database backup costs on OCI and you’ll usually hear the Object Storage Standard rate — €23.72/TB/month. That number is real, and it’s the wrong one.

When you set a database’s automatic backup destination to Object Storage, OCI meters two SKUs, not one:

• Object Storage – Standard (B91628) — the storage itself, €23.72/TB/month.
• Database Backup Cloud Service (B90230) — the backup software component, €4.74/TB/month.

The real rate is €28.46/TB/month, about 20% above the number most estimates are built on.

And here is the part that surprises people: €28.46 is also the price of Autonomous Recovery Service. Not approximately — exactly. Oracle prices its database-integrated backup service at parity with the do-it-yourself bucket route.

Which makes the common instinct — “Recovery Service sounds like the expensive option, let’s just use a bucket” — a decision to pay full price for less:

• Incremental-forever. Recovery Service eliminates the weekly full backup entirely. Object Storage backups run a level 0 full every week plus daily level 1 incrementals; Recovery Service takes incrementals only and assembles a virtual full for any day you ask for. Less CPU, less I/O, a shorter backup window on the production database.
• Validated, immutable backups. Every backup is checked for anomalies and cannot be altered or deleted by anyone inside your tenancy — precisely the property you want when the threat model is ransomware rather than disk failure.
• Protection you can actually see. A per-database dashboard answers “when was the last good backup”, “how far back can I recover”, and “which database is consuming the backup spend” — questions that, with plain buckets, require somebody to go and look.

Zero Data Loss Autonomous Recovery Service (B95241) adds real-time redo protection — recovery to within a second of the incident instead of back to the last scheduled backup — at €37.20/TB/month, a 30% premium. It’s a per-database checkbox, so the premium goes only on the databases that can’t tolerate data loss. That pairing with MAA tiering is the design decision, not the price.

Switching an existing database over is a change of “Backup Destination” in the console: no outage, no gap in protection, and the old Object Storage backups stay usable and expire on their own schedule.

Block and boot volumes: the policy names mislead

Volume backups are incremental, but functionally equivalent to fulls — you can restore a complete volume from any one of them, and you never need to keep “the first full” around. They cost €23.72/TB/month.

OCI’s predefined policies read like a cost ladder. They aren’t one:

All three keep monthly backups for 12 months and a yearly backup for 5 years. Only the short end differs. Gold on a single 10 TB volume with a 5% daily change rate runs roughly €510/month — and Bronze on that same volume is nowhere near the bargain the name implies, because the expensive long tail is identical. If you need a genuinely cheaper profile you have to write a user-defined policy, which is also the only way to get scheduled cross-region copies or to protect a volume group.

Three more things worth knowing before you rely on them:

• Volume groups are how you get a consistent point-in-time snapshot across several volumes at once. An application whose data spans multiple volumes, backed up volume by volume, produces an inconsistent set that merely looks like a backup.
• Cross-region copies are charged as storage in both regions plus outbound transfer, and the destination region’s Vault keys aren’t copied for you — the policy fails without an explicit key OCID in the target region. A daily backup can take up to 24 hours to land.
• Deleting a full backup can raise your bill. Remove one from a chain and the next incremental is promoted to full; its billed size can jump substantially. Tidying up volume backups is not automatically a saving.

File systems, objects, and the estate around them

File Storage is the expensive corner of the estate at €279/TB/month — roughly twelve times Object Storage. Its snapshots are copy-on-write, so a snapshot of an unchanged file system consumes no additional space, and each snapshot’s exclusive bytes metric tells you exactly how much you’d reclaim by deleting it. That metric is the one to check before assuming a snapshot policy is cheap.

Object Storage is where everything else ends up, and its tiers are the main cost lever:

• Standard — €23.72/TB/month, no minimum retention, no retrieval fee.
• Infrequent Access — €9.30/TB/month, but a 31-day minimum and €9.30 per TB retrieved.
• Archive — €2.42/TB/month, a 90-day minimum, and objects must be restored before reading: up to an hour to the first byte, and the restored copy stays available for 24 hours by default (configurable up to 240).

Requests are effectively free for backup workloads — the first 50,000 per month cost nothing, and beyond that it’s fractions of a cent per 10,000. Lifecycle policies move objects down the tiers on a schedule; retention rules make them immutable for a fixed period, which is the right control for regulated data and an expensive one-way commitment everywhere else.

Above all of this sits Full Stack Disaster Recovery, at €0.0119 per OCPU-hour. It doesn’t store anything — it orchestrates the failover of a whole application stack. Worth knowing it exists, and worth not confusing with a backup product.

What each destination costs

Recovery Service bills in “vGB” — the storage that weekly fulls, daily incrementals and archived redo logs would have consumed over a calendar month. You size it exactly the way you’d size a bucket.

Universal Credits and Annual Flex rates are lower than these list prices and vary by commitment and region. The structure is the durable part; confirm the cents in the OCI Cost Estimator.

Retention is the lever, not the storage tier

Most cost-reduction effort goes into storage tiers. Retention is where the money is, and Oracle publishes its own rule of thumb for how database backup volume scales:

Base Database Service offers 7, 15, 30 (default), 45 or 60 days. Recovery Service ships policies at 14, 35 (default), 65 and 95 days. Autonomous Database defaults to 60 days — the longest of the three, and on the ECPU compute model backups are billed separately from database storage rather than bundled into it, as they were under OCPU. That single default, unexamined after an ECPU migration, is one of the most common unexplained jumps on an OCI invoice.

Going from 60 days to 30 saves €341 a year on one 1 TB database. On twenty databases it is roughly €6,800 a year for one configuration change — assuming, of course, that 30 days genuinely covers your recovery and regulatory obligations. That’s the question to answer, not “how much can we keep”.

What it costs in practice

A 10 TB production database, 30-day retention. Roughly 25 TB of backup at €28.46 = €711/month, or €8,538 a year. On the zero-data-loss tier the same database costs €930/month. The €219/month delta is the number to weigh against the cost of an hour of lost transactions — a far more useful comparison than arguing about the premium in the abstract.

Regulatory retention: a 10 TB backup kept for 12 months.

Long-term retention backups sit in Object Storage for 24 hours, then move automatically to Infrequent Access for the rest of the term — half the cost, no lifecycle rules to write. Restoring all 10 TB later costs €93 in retrieval fees, which against €1,722 of savings is not a close call. LTR runs from 90 days to 10 years.

On-premises databases backed up into OCI. Same RMAN mechanism, but here the archive path is legitimate: Database Backup Cloud Service on Archive Storage totals €2.89/TB/month. Ten terabytes costs €29/month instead of €285 — a 10× difference, for backups you accept won’t be readable within the hour.

Where the money quietly leaks

In cost reviews, the recurring findings are rarely about the databases at all.

• Uncommitted multipart uploads. Object Storage bills for uploaded parts whether or not the object was ever committed. A failed upload leaves orphaned parts that don’t appear in the object listing and are never cleaned up automatically. This is the single most common silent charge we find. One lifecycle rule targeting multipart-uploads with the ABORT action fixes it permanently.
• Versioning on a backup bucket. Every overwrite becomes another retained copy. Backups are already immutable by design; versioning on top of them multiplies volume for no recovery benefit.
• Archive tier used for operational backups. 90-day minimum, up to an hour before the first byte is readable, and delete early and you’re billed for the full 90 days anyway.
• Infrequent Access without counting retrieval. 31-day minimum plus €9.30 per TB retrieved. If your restore testing pulls from IA regularly, the tier saves you nothing.
• Cross-region replication switched on estate-wide. Autonomous Database bills replicated backups at 2× the backup volume. Volume copies are charged in both regions plus transfer.
• Backup retention lock enabled “to be safe”. On Autonomous Database it cannot be switched off and the retention period can no longer be changed without going through your Oracle account team.
• Local storage as the backup destination. Backups in a DB system’s Fast Recovery Area disappear with the DB system. Cheap until the one day it matters.
• Terminated resources that left their backups behind. On database termination you choose between deleting backups within 72 hours and deleting them per policy. The second is correct for long-term retention — and is why tenancies quietly accumulate years of backups belonging to things that no longer exist.
• Orphaned volume backups. Manual (non-policy) block and boot volume backups never expire. Nobody remembers boot volumes.
• Standby databases in the DR plan. Automatic backups can’t be configured on a standby, only on-demand ones. A runbook that assumes otherwise has a hole in it.
• Moving from RMAN or dbcli scripts to console-managed automatic backups. This creates a new backup configuration, and the previous backups stop being managed — they’re neither rotated nor deleted.

The licensing trap hiding in your backup job

There’s one backup optimisation that costs nothing in cloud spend and can be expensive in an audit.

RMAN’s BASIC compression algorithm is included in every Oracle edition. The LOW, MEDIUM and HIGH algorithms are part of the Advanced Compression Option, a separately licensed product. On OCI it’s bundled into Enterprise Edition High Performance and Extreme Performance — but if you’re running Standard Edition, or BYOL against licences that don’t include ACO, then a DBA who set MEDIUM to shrink the backup window has cut your storage bill and created a compliance finding.

It’s a small setting, it’s invisible in any cost report, and it’s exactly the kind of thing that surfaces during an Oracle licence review rather than before one.

A checklist before your next invoice

• Database backup destination set to Autonomous Recovery Service, not Object Storage — same price.
• Retention set from your actual RPO/RTO and regulatory obligations, not left at the default.
• Multi-year regulatory retention handled through LTR, not by extending daily retention.
• Zero-data-loss tier enabled per database, not estate-wide.
• A lifecycle rule aborting uncommitted multipart uploads on every backup bucket.
• Versioning off on backup buckets, or on for a reason you can state.
• Volumes on user-defined policies rather than Gold by default, and multi-volume applications in volume groups.
• Custom images used where you only need a known-good server build — they cost nothing to keep.
• Cross-region replication only where the DR plan actually calls for it.
• File Storage snapshot policies checked against their exclusive-bytes figures.
• Tags plus a budget alert on the backup SKUs, so growth surfaces in weeks rather than quarters.
• An audit of orphaned backups: deleted VMs, terminated databases, manual backups that never expire.
• A restore test each quarter that measures RTO — including time spent restoring from Archive.

Bottom line

Backup on OCI is cheap per terabyte and expensive by accident. The two decisions that move the total are which destination you back up to — where Oracle has removed the cost argument entirely by pricing Recovery Service at parity — and how long you keep what you keep. Everything else is hygiene: lifecycle rules, tiering, volume policies that match the workload, and knowing which backups in your tenancy no longer belong to anything.

None of it requires a bigger budget. It requires somebody to own the line item. That’s the work we do, independently and without reselling infrastructure — if backup spend has been growing without an explanation, let’s talk.

Related reading: Which OCI database service is right for you · High availability & disaster recovery on OCI · OCI vs AWS for Oracle workloads.