Why Oracle Cloud bills look different, and where the savings are real
OCI prices egress and compute differently enough that a like-for-like comparison with AWS misleads in both directions. Here is what genuinely moves and how we model it.
Oracle's pricing pitch is aggressive and it is not marketing noise: on several specific lines, OCI is meaningfully cheaper than the other hyperscalers. It is also not cheaper on everything, and the comparison people usually run — take the AWS invoice, price the same instance shapes on OCI — gets the answer wrong because the cost model differs structurally.
Here is what actually differs.
Egress is the headline, and it is real
Data transfer out is where OCI's pricing diverges most. A generous monthly free allowance and a per-GB rate well below the equivalent elsewhere change the economics of any workload that ships a lot of data to users or to another cloud.
For a media platform, a public API with large responses, a backup target or an analytics estate sending results out, this alone can dominate the comparison. For an internal application serving a few hundred employees, it is noise. Look at your actual egress line before you assume either.
Compute: flexible shapes, and the ARM allowance
Flexible shapes let you set the OCPU and memory independently, in one-unit increments. Elsewhere you pick a t-shirt size and pay for whatever the size includes. If your workload needs 6 OCPUs and 40GB, you buy 6 and 40 rather than rounding up to the next size. On a fleet, that rounding is real money.
Note the OCPU/vCPU distinction when comparing: one OCPU on x86 is a full physical core with two hardware threads, so it roughly corresponds to two vCPUs elsewhere. Comparing OCPU count to vCPU count directly makes OCI look twice as expensive as it is, and this specific confusion sinks more comparisons than any other.
Ampere A1 ARM compute is priced low and comes with a substantial always-free allowance. For workloads that run well on ARM — most interpreted languages, Go, modern JVM — this is the largest compute lever available, and the migration is the same build problem described in the Graviton piece.
Preemptible instances for batch and CI, at a steep discount, complete the picture.
Databases: where the comparison gets complicated
If you are running Oracle Database, licensing dominates everything else and the comparison is a licensing conversation, not an infrastructure one — bring your own licence versus licence-included, and the Support Rewards programme that discounts your on-premises support bill based on OCI spend. That programme is frequently the actual financial argument for OCI in an Oracle shop, and it is not visible on any pricing page.
Autonomous Database with auto-scaling and auto-stop for development environments is genuinely cheap to run when idle, which matters more than the headline rate for non-production.
For PostgreSQL and MySQL, OCI's managed offerings are competitive but the comparison is ordinary.
What does not get cheaper
Be honest about these in any business case:
- The ecosystem. Fewer managed services, fewer third-party integrations, a smaller pool of engineers who have run it. That is a real cost even when it is not a line item.
- Regions. Fewer than AWS or Azure, which matters for data residency and latency in specific geographies.
- Migration. Moving a workload costs what moving a workload costs, and the savings need to clear that within a sensible horizon.
How to model it honestly
- Take your real usage, not your instance list. Actual CPU and memory used, actual egress GB, actual storage and IOPS.
- Convert OCPU correctly, and size flexible shapes to the real requirement rather than the current shape.
- Price egress separately and prominently. It is where the difference concentrates.
- Include the Support Rewards offset if you hold Oracle support contracts.
- Add a migration cost and a ramp period, and require the payback to be under a year for a workload move to be worth proposing.
The pattern we see most often is not a full migration but a targeted one: move the egress-heavy or ARM-friendly workload, keep the rest. That is a legitimate outcome and usually a better one than a religious single-cloud position — our cloud work covers all four providers precisely because the right answer is often a split, and Skyline exists to show you the current estate across them before anyone argues about the destination.
What to do this week
Pull your egress GB for last month and your average CPU utilisation across the fleet. Those two numbers determine whether an OCI comparison is worth doing at all, and both take ten minutes to find.