Rehosting Oracle to an authorized cloud voids the 0.5 Intel core factor, so the same 16 cores that needed 8 Processor licenses on-premises now need 16
Lift-and-shift is the only path that preserves your existing entitlements, and it is also the path most likely to double the license count on identical compute, because Oracle's cloud policy counts 2 vCPUs as 1 Processor with no core factor applied. Replatforming to a managed service forks the license model entirely (RDS License Included is Standard Edition 2 only, at $17,500 per socket versus $47,500 per Processor for Enterprise Edition), and refactoring off Oracle resets the cost model while leaving 22% support running on shelf licenses you no longer deploy. Decide the path against the license math before the SOW is signed, not after the first true-up.
Prepared by Redress Compliance · August 18, 2026 · Oracle advisory. Migration and cloud licensing engagements 2024 to 2026.
Executive summary
Rehost preserves entitlement type but can double entitlement quantity: 16 Intel cores on-premises need 8 Processor licenses at a 0.5 core factor, and the same 16 vCPUs in an Authorized Cloud Environment need 16 vCPUs counted at 2:1, so 8 licenses, only if you hold vCPU count flat.
Teams that size cloud instances by memory rather than vCPU routinely land on 8 vCPU shapes where 4 would serve, and pay $95,000 in list Enterprise Edition licenses they did not need.
Replatform to Amazon RDS License Included is a Standard Edition 2 decision, not an Enterprise Edition one, which caps you at 8 vCPUs and strips every option in a stack that lists at $122,000 per fully optioned Enterprise processor.
If your workload depends on RAC ($23,000 per processor) or Partitioning ($11,500 per processor), License Included is not an option and BYOL on RDS or EC2 is the only replatform route.
Refactoring off Oracle changes the deployment but not the invoice: support is 22% of net license fees on the ordering document and does not fall when deployment falls, so a fully migrated estate keeps billing until licenses are formally terminated or repriced.
Every dollar of net license discount you win at the same time is worth roughly $2.10 over a five-year hold, because support is calculated on net rather than list.
Instance shape selection is a licensing decision worth 50% of the license line: a db.r7i.2xlarge at 8 vCPUs and 64 GB needs 4 Processor licenses, and the same 64 GB at 4 vCPUs via Optimize CPU needs 2.
That is $95,000 of Enterprise Edition list price and $20,900 of annual support removed by a configuration flag, before any negotiation happens.
How each migration path is counted: the licensing mechanics of rehost, replatform, and refactor
The three migration strategies are not three flavors of the same license problem.
They are three different counting rules, and only one of them leaves your ordering document untouched. Rehost keeps the same SKU and the same metric, then recounts the deployment under Oracle's cloud policy: 2 vCPUs equal 1 Processor license with multithreading enabled.
1 vCPU equals 1 Processor with it disabled, and the Processor Core Factor Table is explicitly not applicable. Replatform to Amazon RDS License Included changes the product you are running, not just the host: License Included is Standard Edition 2 only, no options, and SE2 tops out at 8 vCPUs.
Bring your own license on RDS or EC2 preserves your metric and your options but binds you to the same 2:1 vCPU arithmetic. Refactor off Oracle removes the deployment entirely and changes nothing at all about what you owe: the ordering document survives.
And 22% support keeps billing on the net license fees you signed for, whether or not a single instance is running.
| Counting element | Rehost (EC2/Azure/GCP, BYOL) | Replatform (RDS License Included) | Replatform (RDS BYOL) | Refactor off Oracle |
|---|---|---|---|---|
| Metric | Processor or NUP, unchanged | Vendor-included, per instance-hour | Processor or NUP, unchanged | Metric persists on paper only |
| Core Factor Table | Not applicable | N/A | Not applicable | N/A |
| Counting ratio | 2 vCPUs = 1 Processor (MT on) | None to count | 2 vCPUs = 1 Processor | None deployed |
| Edition available | EE, SE2, full option stack | SE2 only, no options | EE or SE2, options as owned | None |
| Hard ceiling | None (cost is the ceiling) | 8 vCPUs (SE2 limit) | 8 vCPUs if SE2, none if EE | N/A |
| NUP floor | 25 per Processor (EE), 10 per server (SE2) | N/A | 25 per Processor (EE) | Floor still owned |
| Support at 22% | Continues | Continues on shelf licenses | Continues | Continues indefinitely |
The table's real message sits in the bottom row. Every path keeps paying 22% support on the original net license fees. Rehost and BYOL replatform at least deploy against that spend.
License Included and refactor do not, which is why both paths quietly create shelf licenses: you have bought a second cost model without retiring the first. Oracle will not volunteer a reduction, and support does not fall when deployment falls.
Two structural details decide most models. First, the four Authorized Cloud Environments are Amazon EC2, Amazon RDS, Microsoft Azure, and Google Cloud Platform. Google Cloud is on the list, contrary to a good deal of older advice still circulating in migration SOWs.
Anything outside those four is not covered by the policy and reverts to standard physical counting rules, which on a shared virtualization layer means the full host.
Second, the NUP/Processor crossover is exactly 50 users per Processor by construction, because Enterprise Edition prices NUP at $950 against $47,500 per Processor, one fiftieth.
Below 50 real users per Processor, NUP is cheaper; above it, Processor wins, and the 25 NUP per Processor floor on EE (10 NUP per server on SE2) means you license the higher of the floor and actual headcount, never the lower.
One last point on standing: Oracle's own policy document states it may not be incorporated into any contract, yet License Management Services enforces it as the practical basis for every cloud audit we defend. That gap is leverage, not comfort.
Treat the policy as the audit position you will face and negotiate contractual language if you want anything better.
The rehost trap: identical compute, higher license count
Run the arithmetic that catches most programs. On-premises, 16 Intel cores times the 0.5 core factor equals 8 Processor licenses. Lift that same workload to an Authorized Cloud Environment as 16 vCPUs with multithreading on, and the 2:1 ratio also gives you 8. Identical.
That is the answer teams present to the steering committee, and it is technically correct. The trap is that 16 physical cores with hyperthreading present 32 threads on-premises, and cloud vCPUs are threads, not cores.
A one-to-one replacement of visible compute means 32 vCPUs, which is 16 Processor licenses, double the on-premises count at $47,500 list each.
The trap lands even harder when teams size for memory rather than compute, because memory-optimized instance families scale RAM by stepping up the vCPU count you never asked for.
The counter-move is instance selection, which on Oracle workloads is a licensing decision, not an infrastructure one. Optimize CPU on the AWS M7i and R7i families lets you keep the memory and cut the vCPUs. A db.r7i.2xlarge at 8 vCPUs and 64 GB RAM requires 4 Processor licenses.
Reduce to 4 vCPUs at the same 64 GB and you require 2, a 50% cut against unchanged memory that the database actually uses. That is roughly $95,000 of list license and about $20,900 of annual support removed per instance, before discount.
Two related misconceptions cost real money in our engagements: Graviton and other ARM instances get no special treatment, because the 2:1 rule applies uniformly regardless of processor architecture on shared tenancy.
And non-ACE or dedicated-host instance types can revive the full-host position that VMware creates on-premises, where Oracle counts every core in the cluster rather than every core you assigned.
If cost per licensed workload is the deciding factor, model it across providers before you commit, as we set out in the multicloud comparison.
The headline saving in a rehost business case is almost always infrastructure, and the headline overrun is almost always licensing. Size the instance to the license count first, then check whether the memory and IOPS still work.
Doing it the other way round is how a cost-neutral lift-and-shift ends up at 2x the Processor count on the same application.
Avoid the Oracle Exadata core license trap
Oracle Exadata can lock you into full core licensing across X9M, X10M, and Cloud at Customer. The buyer side strategy to size the platform and cut the bill.
Get the white paper →The replatform fork: License Included caps your product, BYOL caps your shape
Replatforming to a managed service is sold to your infrastructure team as an operational simplification, and it is nothing of the kind. It is a product substitution decision that someone in a cloud architecture review is making on Oracle's behalf, usually without reading the option stack it deletes.
Amazon RDS License Included ships Standard Edition 2 only.
That single fact removes Real Application Clusters ($23,000 per Processor at list), Partitioning ($11,500), Advanced Security, Advanced Compression, Diagnostics Pack, and Tuning Pack from the menu, and it imposes SE2's hard ceiling: four vCPUs per socket.
Capped at 8 vCPUs total, with a 10 Named User Plus per server floor.
If the application you are replatforming depends on partition pruning for query performance, or on Diagnostics Pack because your DBAs live in AWR reports, License Included is not a cheaper version of your database.
It is a different database, and the remediation work lands on the project, not on Oracle.
BYOL goes the other way: you keep the full option stack you already own, but you inherit vCPU counting under the Authorized Cloud Environment policy, so 2 vCPUs equals 1 Processor with no core factor relief, and every option you deploy is licensed at that same inflated Processor count.
License Included caps what you can run; BYOL caps how large you can run it before the bill compounds across seven line items instead of one.
| Dimension | RDS License Included | BYOL on RDS or EC2 |
|---|---|---|
| Edition available | Standard Edition 2 only | EE, SE2, whatever you own |
| Options (RAC, Partitioning, Diagnostics) | Not available at any price | Available, licensed separately at list |
| Size ceiling | 8 vCPUs (SE2 policy limit) | No policy ceiling, cost ceiling only |
| Counting basis | Bundled in hourly rate | 2 vCPUs = 1 Processor, core factor void |
| Reference list price | SE2 $17,500 per occupied socket | EE $47,500 per Processor, plus options |
| Fully optioned EE exposure | Not applicable | $122,000 per Processor list, $26,840 annual support |
| Existing entitlement reuse | None, you re-buy as a rate | Preserved, and still on 22% support |
The one genuinely new piece of leverage here is bare metal.
Oracle's core factor is void in Authorized Cloud Environments because the policy counts vCPUs, but the September 2025 RDS bare metal launch for Enterprise Edition BYOL, extended to SE2 in January 2026 at roughly 25% below the virtualized rate.
Presents a shape where physical cores are the visible unit.
That is worth a written question to Oracle before you commit, because the policy document itself states it "may not be incorporated into any contract," meaning your position on any deployment shape is only as strong as the correspondence you hold.
Ask for a specific counting statement on the specific instance family, in writing, from someone with authority, and price both outcomes.
- Pull the option inventory before the architecture review, because License Included silently deletes RAC, Partitioning, and the management packs your DBAs already depend on.
- Model SE2's 8 vCPU ceiling against real peak CPU, not average, since exceeding it is a compliance event rather than a performance event.
- Get the bare metal counting position in writing, naming the instance family and the edition, before the SOW commits you to that shape.
- Price BYOL at the fully optioned number, $122,000 per Processor list with $26,840 annual support, so the vCPU multiplier is visible across every line and not just the base database.
Why refactoring is the only path that changes the cost model, and the only path Oracle bills you for anyway
Rehost and replatform both keep your workload inside Oracle's pricing gravity. Rehost keeps it there in the worst possible orientation: identical compute, more Processor licenses, because the core factor evaporates the moment the workload lands in an Authorized Cloud Environment.
Replatform keeps it there in a different currency, either as a bundled rate on a capped edition or as BYOL on vCPU counting. Refactor is the only path that leaves the gravity well entirely, and this is precisely why it is the only path that generates real resistance from your Oracle account team.
Watch the behavior: a rehost proposal produces a helpful architecture conversation and possibly a cloud credit. A refactor plan produces a license review request, a sudden interest in your Java estate, and a call to someone two levels above your CIO.
The commercial trap in refactor is that the invoice survives the migration by years. Oracle support is 22% of the net license fees on the ordering document, plus accumulated uplift. It attaches to what you own, not to what you run.
You can decommission the last Oracle instance on a Friday and receive the identical support invoice the following quarter, because nothing in your entitlement changed. Technical completion and commercial completion are separate events, and only one of them is on your program plan.
That gap is where refactor programs quietly fail their business case. The board approved a number that assumed the Oracle line went to zero when the application was retired.
What actually happens is that the application retires, the CSI stays intact because nobody wants to trigger repricing on the remaining lines, and the shelf licenses sit there accruing 22% for three more renewal cycles.
Redress sees this pattern in most multi-year exit programs we review: the technical savings are real and the licensing savings are entirely unrealized.
The footprint review has to happen before renewal, not after decommissioning, because the renewal is the only moment Oracle is contractually engaged with the question.
Sequencing therefore inverts the intuitive order. You do not refactor and then clean up the licensing. You establish the terminatable set, the repricing exposure on the surviving set, and the target renewal date first, and then you sequence the technical work to land ahead of it.
If your refactor finishes in month 14 and your renewal is in month 8, you have committed to a further full term of support on licenses you already know you will not deploy.
The math on what to do with shelf licenses runs through the discount arithmetic. Because support is calculated on net license fees rather than list, every dollar of net discount is worth roughly $2.10 over a five-year hold: one dollar of license plus five years of the support it never generates.
That number cuts both ways.
It means the discount you fought for on the original purchase is doing quiet work you are still paying for, and it means trading shelf licenses into a new commitment is often worth more than terminating them outright, provided you get the credit applied to net rather than list on the incoming lines.
Termination is cleanest and gets the 22% off your P&L permanently, but Oracle will usually respond by repricing what remains.
Repurposing (moving the entitlement to a workload you are not exiting) preserves the discount structure and costs nothing extra, and it is the move most estates should default to.
The decision is not rehost versus replatform versus refactor. It is whether your migration program has a commercial completion date alongside its technical one. Rehost and replatform are cost events you can model in advance.
Refactor is a cost event you have to negotiate for, at a specific moment, against a vendor whose contract gives them no obligation to reduce anything.
So the practical reading is this: treat rehost and replatform as arithmetic and refactor as negotiation.
Build the refactor case on the renewal calendar, decide termination versus repurpose versus trade before you tell Oracle anything, and hold the $2.10 figure in view when the account team offers to "help with the transition" in exchange for a term extension.
What we see in migration engagements: the recurring patterns
The evidence base for everything above is a published Oracle document that Oracle itself disclaims. The cloud licensing policy PDF carries language stating it is for educational purposes and "may not be incorporated into any contract." That single sentence is the whole negotiation.
It means the 2 vCPU to 1 Processor ratio and the removal of the Core Factor Table are not contractual obligations for most customers, whose Oracle Master Agreement and ordering documents say nothing about vCPUs at all.
It also means, in our experience across migration engagements, that License Management Services will still open every AWS, Azure, or Google finding on exactly that basis, price it at list, and wait for you to argue.
Buyers who assume the disclaimer protects them discover that the practical audit basis and the contractual basis are two different documents, and only one of them arrives with a settlement number attached.
The asymmetry is the point: Oracle gets to enforce a rule it never had to negotiate, and you get to litigate it after the workload is already running.
The correct time to resolve that is before the target design is frozen, which is why the Authorized Cloud Environment core counting policy belongs in the contract review, not the infrastructure review.
Stripping the 0.5 Intel core factor turns 16 cores from 8 Processor licenses on-premises into 16 in an Authorized Cloud Environment.
A fully optioned Enterprise Edition processor (RAC, Partitioning, Advanced Security, Advanced Compression, Diagnostics, Tuning, Active Data Guard) lists at $122,000 with $26,840 in annual support, 2.6 times the base database.
The repeat failure modes cluster tightly. First, Standard Edition 2 instances found above eight vCPUs, which breaches the SE2 ceiling outright and converts a $17,500 per socket position into an Enterprise Edition claim at $47,500 per Processor, retroactive to first deployment.
Second, option stacks carried into cloud shapes by image copy: RAC licensed on a single-instance cloud database, Partitioning licensed where no partitioned table survived the move, Diagnostics and Tuning licensed because the AWR pack was never disabled.
The option is no longer used, but the support line still bills, and at $26,840 per Processor per year on a full stack, that is the most expensive shelfware in the estate.
Third, dual-running periods where source and target are both fully licensed for months, an exposure covered in detail in our guide to dual-running Oracle during a cloud cutover.
Fourth, and the most common of all, shelf support left running 12 to 36 months after cutover because nobody attached a termination date to the source licenses.
Support is 22% of net license fees and does not fall when deployment falls, so the on-premises estate keeps invoicing at full rate while the cloud estate invoices alongside it.
- 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
Your first five moves
- Freeze the vCPU count as a licensing artifact, not an infrastructure detail, and have the target architect sign off a per-instance vCPU inventory before the design gate, because at 2 vCPUs per Processor with no core factor every unreviewed instance-size decision is a $47,500 list decision.
- Run the SE2 versus Enterprise Edition and License Included versus BYOL comparison before the SOW is signed, modelling $17,500 per socket against $47,500 per Processor plus options, and check every proposed Standard Edition 2 instance against the 8 vCPU ceiling now rather than after LMS finds it.
- Get Oracle's cloud policy position in writing or priced into the contract, because the policy PDF says it may not be incorporated into any contract while LMS enforces it anyway, so either secure a written cloud counting addendum or budget the difference between your contractual position and the policy position as a named audit reserve.
- Tie source license termination or repricing to the cutover date, not to project close, with a dated support reduction request filed against the specific CSI lines, so you are not paying 22% on decommissioned Processors for the 12 to 36 months it typically takes anyone to notice.
- Use the refactor decision as renewal leverage while support is still live, because a credible, costed plan to move a workload off Enterprise Edition is worth far more at the table than after the licenses are shelved, and every dollar off the net license price is worth roughly $2.10 across a five-year hold once avoided support is counted, as covered in our work on optimizing the footprint before renewal.
Frequently asked questions
Does rehosting Oracle to AWS or Azure require new licenses?
No, rehosting preserves your existing Processor or Named User Plus entitlements, since you bring your own license into an Authorized Cloud Environment.
What changes is the quantity required, because the Oracle Processor Core Factor Table does not apply in Authorized Cloud Environments, so the 0.5 Intel factor is gone and vCPUs are counted at 2 per Processor license with multithreading enabled.
If your cloud shape has more vCPUs than your on-premises core count times the core factor, you need more licenses for the same workload.
Which clouds does Oracle treat as Authorized Cloud Environments?
The current policy names Amazon EC2, Amazon RDS, Microsoft Azure, and Google Cloud Platform. Google Cloud is on the list despite a lot of older guidance saying otherwise, so check the effective policy date rather than a consultant deck from 2022.
Anything outside those four, including most hosting providers and private clouds, falls under standard on-premises rules where the core factor still applies but so do the full-host virtualization arguments.
Can I run Oracle Database Enterprise Edition on Amazon RDS License Included?
No. RDS License Included is Standard Edition 2 only, which means no Enterprise Edition, no RAC, no Partitioning, no Advanced Security, and no Diagnostics or Tuning Pack.
If your workload needs any of that, you must use BYOL on RDS or EC2 and keep paying for Enterprise Edition at $47,500 per Processor plus the option stack, which reaches $122,000 per processor fully optioned.
How many vCPUs can Standard Edition 2 use in the cloud?
Eight. Oracle's cloud policy allows four vCPUs per socket for Standard Edition, capped at 16 vCPUs for legacy SE and 8 vCPUs for SE1 and SE2.
Any SE2 instance running above 8 vCPUs is a live compliance exposure and should be resized or reclassified before an audit finds it, because the remediation price is the Enterprise Edition gap, not a small true-up.
If I refactor off Oracle entirely, does my support bill stop?
Not automatically. Support is 22% of the net license fees on your ordering document, it is not a usage fee, and it does not fall when deployment falls.
You have to formally terminate or reprice the licenses, and Oracle's repricing rules on partial terminations frequently make the remaining support base higher per unit than before, so model the termination outcome before you switch anything off.
Is Oracle's cloud licensing policy contractually binding?
Oracle's own policy PDF states it may not be incorporated into any contract, so for most customers it is not contractual. In practice, Oracle License Management Services enforces it as the basis for cloud audit findings.
That gap is negotiation leverage: if the counting rule materially drives your cost, ask for the counting method to be written into the ordering document or an amendment rather than relying on a document Oracle can revise unilaterally.
Does using AWS Graviton or ARM instances reduce Oracle license count?
No. Graviton and other ARM-based shared-tenancy instances follow the same 2 vCPUs equals 1 Processor rule, because Oracle's cloud licensing policy applies uniformly regardless of processor architecture.
The lever that actually works on shared tenancy is reducing vCPU count at constant memory, for example using Optimize CPU to run 64 GB on 4 vCPUs instead of 8, which halves the license requirement.
