OCI vs AWS pricing for a typical production compute tier
A like-for-like comparison of on-demand compute across OCI and AWS for a four-node application tier, including what changes once commitments and ARM instances enter the picture.
Comparing compute prices across clouds is harder than it should be, because nobody sells the same unit. AWS sells fixed instance shapes; OCI sells flexible shapes where you pick OCPUs and memory independently. To compare honestly you have to normalise to vCPU-hours and GB-hours and then check that the underlying silicon is comparable.
Normalising the unit
One OCI OCPU is a full physical core with two hardware threads — roughly equivalent to two AWS vCPUs, which are threads. Comparing 'per OCPU' against 'per vCPU' without that conversion overstates OCI's price by about 2x, and it's the single most common mistake in vendor-produced comparisons.
| Provider / shape | Unit price | Normalised per vCPU-hour |
|---|---|---|
| AWS m7i (Intel) | per-instance, fixed shape | ~$0.0504 |
| AWS m7g (Graviton ARM) | per-instance, fixed shape | ~$0.0408 |
| OCI E5 Flex (AMD) | ~$0.024 / OCPU-hr + ~$0.0015 / GB-hr | ~$0.012 + memory |
| OCI A2 Flex (Ampere ARM) | ~$0.01 / OCPU-hr + ~$0.0015 / GB-hr | ~$0.005 + memory |
Figures are public on-demand list prices in US regions, gathered August 2026. Discounts (committed use, reserved capacity, EDP/enterprise agreements) change the picture materially — always confirm against the provider's live calculator before deciding.
Worked example: four-node app tier, 8 vCPU / 32 GB each
| Configuration | Approx. monthly (730 hrs) |
|---|---|
| AWS 4x m7i.2xlarge (8 vCPU / 32 GB) | ~$1,180 |
| AWS 4x m7g.2xlarge (Graviton) | ~$955 |
| OCI 4x E5 Flex (4 OCPU / 32 GB) | ~$420 |
| OCI 4x A2 Flex (4 OCPU / 32 GB) | ~$260 |
On list price alone, that's a 60–75% delta. But list price is not what large estates pay, and that's where honest comparison starts rather than stops.
What closes the gap
- Savings Plans and Reserved Instances: a 3-year commitment on AWS can cut compute 40–60%, which materially narrows the difference.
- Enterprise agreements: negotiated discounts on either side rarely appear in any public comparison.
- ARM migration effort: OCI's biggest compute savings come from Ampere shapes, which means rebuilding containers for aarch64 and validating every dependency. That's real engineering time.
- Ecosystem gravity: managed services, IAM patterns, and CI/CD tooling are the actual switching cost, not the VMs.
What doesn't close the gap
Egress and block storage. Compute discounts are negotiable and heavily competed; data transfer and storage list prices move far less. That's why, in most real migrations we look at, the durable saving comes from the data layer rather than from the instances — even though instances are where the conversation always starts.
The right answer is workload-specific. A steady, ARM-friendly, egress-heavy platform is a strong candidate to move. A workload deeply wired into managed AWS services with modest data transfer usually isn't, and we'll say so.