OCI cost is decided in the order document, not the console
The commitment you sign, the shape of the drawdown against it, and the meters your architecture happens to hit move more money than any instance you tune: OCI bills against Universal Credits, a prepaid pool drawn down as you consume, and the commitment punishes in both directions, forfeiture of unused credits at term end, and overage invoiced at the rate card in your order. Across our cost reviews, the commitment was almost always the wrong size or the wrong shape, and the architecture was rarely the problem.
Prepared by Redress Compliance · August 8, 2026 · Oracle advisory. Based on 30 to 40 OCI cost reviews run 2024 to 2025.
Executive summary
The commitment punishes both directions, and it was oversized in most estates.
Credits not consumed by term end are forfeited and nonrefundable, excess usage invoices monthly at the rate card established in your order, and both penalties are contractual.
Untouched by any console tuning after signature: annual commitments were oversized 20 to 40 percent to reach a headline discount tier, then partly forfeited at term end.
A deeper discount on a commitment you cannot draw is a worse deal, and the overage rate card must be your discounted one, in writing, or growth is priced at a number you never negotiated.
The shape matters more than the total, because migrations ramp and commitments do not.
Most programs consume little in the first two quarters and accelerate later, so a flat annual figure signed against a back loaded plan forfeits the first half of the year by construction, the worked curve ending at 90 percent consumed and 10 percent forfeited.
The fix is a ramped commitment, a lower first year or defined annual steps, agreed in the order: model the curve from twelve months of actual usage, include only workloads with a funded migration date, hold a slice of demand outside the commitment.
And negotiate the top up rate up front, because you will almost certainly need one.
The meters hide money the compute count never shows.
One OCPU is two vCPUs on x86, so comparing OCPU rates against hyperscaler vCPU rates makes OCI look roughly twice as expensive as it is, with Arm shapes at one to one.
Flexible shapes meter memory separately at one to 64 gigabytes per OCPU, so a generous template ratio bills invisibly across the fleet.
Block storage bills performance in volume performance units independent of capacity, the Balanced default carrying 10 units per gigabyte where Lower Cost carries zero, changeable on live volumes.
And reserved capacity bills at 85 percent of rate even when unused, until somebody deletes the reservation.
The idle tail and the unchecked accruals fund the whole review. Idle and orphaned resources, mostly block volumes, stopped compute, and forgotten capacity reservations, accounted for 8 to 15 percent of the bill, recoverable without touching a single commercial term.
And nobody had checked the Support Rewards accrual rate against their own order document, with several estates counting rewards against support invoices that were never eligible.
The rewards saturate too, at 25 cents per dollar covering the eligible technology support invoice only once OCI spend reaches about four times it, beyond which extra spend earns nothing usable.
The purchase models, priced against forecast error
| Model | How you pay | Forecasting too high costs | Forecasting too low costs |
|---|---|---|---|
| Pay as you go | Metered, no commitment | Nothing | Nothing, but the highest unit rate throughout |
| Annual Universal Credits | A prepaid pool, drawn monthly | Forfeiture of the unused balance | Overage at the order rate card |
| Multi year commitment | Prepaid at a deeper tier | Forfeiture, repeated, on an untrusted forecast | Overage, plus a lost repricing at renewal |
The drawdown curve is the whole sizing argument.
The worked quarter by quarter: 10 percent of the year consumed in Q1 against a flat 25 percent commitment slice, 18 in Q2, 30 in Q3 drawing down the surplus, 32 in Q4, ending at 90 percent consumed and 10 forfeited, paid for capacity nobody used while the migration was still on its curve.
A ramped commitment absorbs exactly this, and the buffer rules complete it: funded migration dates only, since a roadmap is not consumption; a demand slice held outside the commitment; the smaller discount fully used over the larger one partly forfeited.
And the top up rate negotiated before it is needed.
The four meters that do the quiet work
- The OCPU normalization: one OCPU is a physical core, two execution threads, so x86 comparisons against vCPU rates overstate OCI by roughly half, and Arm A1 shapes run one to one, changing the comparison again. Normalize before benchmarking anything.
- The flexible shape memory ratio: memory meters separately per gigabyte hour, from one to 64 gigabytes per OCPU, so a fleet built from a generous template carries a charge no OCPU count reveals. Check the ratio across the estate.
- The block storage performance tier: volume performance units bill independently of size, Balanced defaulting to 10 per gigabyte and Lower Cost to zero, adjustable on live volumes, the free lever on non production, archives, and log stores.
- The autonomous autoscaling ceiling: compute autoscaling on an Autonomous Database runs to three times base silently, a ceiling that belongs in the forecast rather than the surprise column.
The OCI FinOps brief
The commitment sizing method, the drawdown modeling, the meter map, and the renewal sequence worked on a representative estate.
Get the white paper →Support Rewards, and the commitment renewal
The rewards check takes an afternoon and several estates failed it: the accrual rate is an order level term, 25 cents per eligible OCI dollar or 33 with an active ULA, the offset applies to technology license support only.
And the saturation ceiling means OCI spend above roughly four times the eligible invoice earns credit that expires unused, arithmetic worked fully in the Support Rewards guide.
The renewal is where the review converts to money: the consumption history replaces the forecast, the drawdown shape justifies the ramp, the idle tail is cleaned before the baseline snapshots, and the commitment resizes to the curve you can prove rather than the tier the discount suggested.
The wider FinOps operating rhythm runs in the OCI FinOps framework, and the BYOL against license included decision on every database workload in the cloud licensing guide.
- Percentile standing for your exact deal size and industry, from real closed transactions
- Scenario simulation before the call: test alternative terms and see the financial impact of each
- A negotiation playbook, talking points, and a two page executive brief on day one
What we saw across OCI cost reviews, 2024 to 2025
Across roughly 30 to 40 OCI cost reviews Fredrik Filipsson and the Redress team ran between 2024 and 2025, the commitment was almost always the wrong size or the wrong shape, and the architecture was rarely the problem:
Signed to reach a headline discount tier, then partly forfeited at term end.
Orphaned volumes, stopped compute, and forgotten reservations billing at 85 percent.
The review sequence that worked runs commercial first, technical second: check the overage rate card is the discounted one, check the Support Rewards accrual and eligibility against the order, model the drawdown against the commitment, and only then walk the meters, the memory ratios.
The performance tiers, and the reservations, because the console work optimizes inside boundaries the order document already set.
A deeper discount on a commitment you cannot draw is a worse deal than a smaller one you will fully use, and that single sentence, applied at signature, would have saved more than every instance tuning session in the file combined.
Your first five moves
- Verify the overage rate card is your discounted one, in writing, or growth prices at a number you never negotiated.
- Size the commitment to a drawdown curve, ramped in the order, funded migrations only, with the top up rate pre agreed.
- Sweep the idle tail quarterly: orphaned volumes, stopped compute, and reservations billing at 85 percent unused.
- Check the Support Rewards rate and eligibility against your order, where several estates counted ineligible invoices.
- Normalize OCPUs before any benchmark, and check the memory ratios and storage tiers fleet wide. The Oracle practice runs the review with you.
Frequently asked questions
How does OCI billing work?
Against Oracle Universal Credits, a prepaid pool of dollars drawn down as services consume: credits unused at term end are forfeited and nonrefundable, and usage above the pool invoices monthly at the rate card established in your order.
The three upstream decisions, the purchase model, the commitment size and shape, and the meters your architecture hits, move more money than any console tuning.
How big should an OCI commitment be?
Sized to a drawdown curve, never an annual number: migrations consume little early and accelerate, so a flat commitment against a back loaded plan forfeits the first half year by construction.
Model twelve months of actual usage, include only workloads with funded migration dates, hold a demand slice outside the commitment, prefer the smaller discount you will fully use, and negotiate the top up rate up front.
Is OCI cheaper than AWS or Azure?
The comparison breaks before it starts unless you normalize: Oracle bills x86 compute per OCPU, a physical core equal to two vCPUs, so OCPU rates against hyperscaler vCPU rates make OCI look roughly twice as expensive as it is, while Arm A1 shapes run one to one.
Normalize the units, include the memory metering on flexible shapes, and then benchmark.
Where does OCI spend hide?
In the idle tail and the quiet meters: orphaned block volumes, stopped compute, and forgotten capacity reservations billing at 85 percent of rate accounted for 8 to 15 percent of bills in our reviews, while flexible shape memory ratios.
The Balanced storage tier's 10 volume performance units per gigabyte against Lower Cost's zero, and the autonomous autoscaling ceiling of three times base all bill outside the compute count.
How do OCI Support Rewards interact with cost optimization?
As an order level term to verify, not assume: rewards accrue at 25 cents per eligible OCI dollar, 33 with an active ULA, offset technology license support only, and saturate once OCI spend reaches about four times the eligible invoice, beyond which extra spend earns nothing usable.
Several estates in our reviews were counting rewards against invoices that were never eligible, an afternoon's check nobody had run.
What happens if you overshoot or undershoot an OCI commitment?
Both directions are contractual penalties: undershoot and the unused credits forfeit at term end, nonrefundable; overshoot and the excess invoices monthly at the rate card in your order, which must be your discounted card in writing or growth prices at list.
Commitments oversized 20 to 40 percent to reach discount tiers, then partly forfeited, were the single most common finding in our reviews.
