Oracle Database licensing on AWS. Why the same server needs twice the licenses.

On AWS, Oracle counts vCPUs and drops the 0.5 core factor, so identical hardware needs about twice the licenses it needs on premises. Instance family choice, RDS license model and option control decide what you actually pay.

How does Oracle count processors for Oracle Database on AWS?

Oracle counts virtual CPUs on AWS, and the core factor table that halves the count on Intel and AMD servers in your own data center does not apply. AWS is on Oracle's list of Authorized Cloud Environments, alongside Microsoft Azure and Google Cloud, and the list comes with its own counting rules.

• Multithreading on. Two vCPUs equal one Oracle processor license. This is the default on almost every EC2 instance type.
• Multithreading off. One vCPU equals one processor license.
• No core factor. The 0.5 multiplier for Intel and AMD is simply gone on AWS compute.
• Rounding. Take the vCPU count the instance presents, halve it where threads are on, and round up.

The arithmetic is mechanical. An r6i.8xlarge at 32 vCPUs is 16 Oracle processors. An x2iedn.2xlarge at 8 vCPUs and 256 GiB of memory is 4. The rules cover Oracle Database, Middleware and a defined set of Oracle Technology programs running on EC2 and on Amazon RDS, and nothing else.

Why does a lift and shift double the license count?

On premises, a 48 core Intel server needs 24 processor licenses at the 0.5 core factor. Run the same workload on AWS with the same physical capacity and it needs 48, because the factor stops at the Authorized Cloud Environment boundary and you gain nothing in exchange.

Say you move that server unchanged and license it at Enterprise Edition list price.

Discounts scale both columns by the same ratio, so the gap survives any price you negotiate. The full treatment of the multiplier is in our core factor analysis, and the on premises rules for virtualization sit in the partitioning policy.

Does disabling hyperthreading on EC2 cut the count?

No. An r6i.4xlarge is 8 Oracle processors at 16 vCPUs with threads on. Turn multithreading off and it presents 8 vCPUs, which the policy now counts one to one, so it is still 8 processors. You halve the compute and save no license.

Some workloads do run better single threaded, and that is a fair reason to change the setting. Just do not book it as a licensing saving.

How does the Named User Plus minimum work on AWS?

Named User Plus keeps its minimum of 25 users per processor, and the processor number is the AWS number. An 8 processor instance therefore carries a floor of 200 named users before anyone logs in.

Buyers who calculate the floor from the old on premises processor count, usually half the AWS figure, end up short. Our Named User Plus minimum examples work through the edge cases.

Which Oracle deployments on AWS does the cloud policy cover?

The cloud policy governs counting on EC2 and RDS only. Engineered Systems, most Oracle Applications, Java, VMware Cloud on AWS and Oracle Database@AWS all sit outside the vCPU mapping and count under their own terms, which is where most of the arguments start.

Where do the boundary cases land?

Oracle Database@AWS, which Oracle now markets as Oracle AI Database@AWS, is an Oracle cloud service running in AWS data centers. You buy it license included or bring your own licenses, it is sold through AWS Marketplace, and its usage counts toward your AWS commitments. The vCPU rules on this page do not apply to it.

VMware Cloud on AWS is the contested one. Our practical position is that a dedicated bare metal software defined data center counts as physical hosts under the partitioning rules, so the whole cluster is in scope wherever Oracle can run. The VMware and Oracle guide covers the host counting.

Which AWS purchasing, scaling and standby choices change the count?

Most AWS cost tools change the AWS bill and leave the Oracle count exactly where it was. These are the cases that catch most teams.

• Reserved Instances and Savings Plans. These are AWS commercial commitments. They reduce the Oracle footprint by zero.
• Spot instances. A Spot instance carries a full processor obligation for as long as it runs.
• Data Guard standbys. A standby applying redo is installed and running, so it is licensed whether it serves traffic or sits idle.
• Auto Scaling groups. Oracle counts the group at its peak concurrent instance count, not the average.
• Dedicated Hosts and bare metal. These are still counted on instance vCPU. The core factor does not come back on metal.

The same rules applied across Azure and Google Cloud are set out in Oracle licensing in cloud environments.

Why does instance memory drive the Oracle license bill on AWS?

Because the license is bought per vCPU and AWS instance families deliver very different amounts of memory per vCPU. Most Oracle databases are limited by memory before they run out of threads. Teams usually reach a memory target by buying a bigger size in the family they already run, and on AWS that buys Oracle licenses the database never uses.

Every row holds the same memory and runs the same database, yet the Oracle list cost varies sixteen fold. The higher instance hour rate on the dense family is nowhere near enough to close that gap.

In the reviews we ran, moving a memory bound database from a general purpose family to a memory dense one removed 40 to 70 percent of the processor count, with no commercial negotiation at all.

Can you reduce active cores instead of changing instance family?

Often, yes. You can set the number of active CPU cores and the threads per core when you launch an EC2 instance. RDS for Oracle offers the same control through processor features, on bring your own license instances only. Memory stays the same.

An r6i.8xlarge set to 4 cores with 2 threads presents 8 vCPUs, which is 4 Oracle processors with all 256 GiB intact. Three conditions come with it.

• You pay for the whole instance. AWS bills the full instance hour, so compare this against the memory dense family on total cost.
• RDS needs care. Changing the setting causes a brief outage, and every later change of instance class must restate the core and thread values.
• Keep the evidence. Oracle's cloud policy document does not mention the setting, so save the launch configuration and the API output showing the active cores. Oracle will ask how the instance was built.

Do clock speed, Graviton or burstable instances help?

Clock speed is the second sizing factor after memory. A workload limited by throughput can often run on half the vCPUs of a general purpose family if you move it to a high frequency family.

• Graviton. Not an option on RDS, which offers no ARM instance classes for Oracle. Oracle does publish ARM builds of Database 19c for Linux, so treat Graviton on EC2 as a support question to settle before anyone builds a license plan around it.
• Burstable T family. CPU credits do not reduce the count. Oracle counts the vCPUs the instance presents, whether it is bursting or idle.

Run every sizing comparison on total cost, and price the same physical machine both ways before you sign the migration business case.

How should you license Amazon RDS for Oracle?

RDS for Oracle comes in two license models, and the choice decides who answers to Oracle in an audit. License included is Standard Edition 2 only, with the Oracle license inside the hourly rate. Bring your own license runs Standard Edition 2 or Enterprise Edition from your own contract, counted on the vCPU of the instance class.

You can switch license model later by modifying the instance. The edition is harder to undo. RDS allows a move from Standard Edition 2 to Enterprise Edition, by restoring a snapshot, if you hold an unused Enterprise Edition license, and it allows no move from Enterprise Edition back down.

What happens when a database outgrows Standard Edition 2?

The policy caps your own Standard Edition 2 licenses at 8 Amazon vCPUs. That ceiling, far more often than performance, is what pushes a database onto Enterprise Edition. The step is a jump to a different price list with a separate options catalog, and our note on SE2 vCPU limits covers the counting in detail.

• Standard Edition 2 at list. $17,500 per socket, with every 4 vCPUs on AWS counted as one socket, rounded up. An 8 vCPU instance is two sockets, or $35,000.
• Enterprise Edition at list. The same instance is 4 processors at $47,500 each, more than five times as much before a single option is added.

Which RDS features add licenses you did not plan for?

The standby and the replica. Under bring your own license, a Multi-AZ deployment needs entitlement for both the primary and the standby instance.

A read only replica runs Active Data Guard, a separately licensed Enterprise Edition option at $11,500 per processor at list, billed on the replica's vCPU as well as the primary's. A mounted replica avoids Active Data Guard but still needs Enterprise Edition on both sides.

Is license included RDS cheaper than bringing your own license?

Only for databases that run a few hours a day. License included looks safer because the compliance question disappears. On any database running more than a few hours a day, the Oracle premium inside the hourly rate ran 10 to 18 percent above a bring your own license position in the environments we reviewed.

Meter real running hours for two billing cycles before you choose. Then split the fleet on purpose: license included for spiky and short lived databases, your own licenses for steady state. The entitlement rules behind that choice are in the bring your own license guide.

What does an Oracle audit look for on AWS?

Oracle targets AWS deployments because the vCPU question is widely misread and easy to verify remotely. Five findings account for most of what we see in those audits.

1. Hyperthreading assumed. The two vCPU rule gets applied to an instance running single threaded, or the reverse, because no one checked the running instance.
2. Oversized instances. A memory problem solved by doubling the license footprint for capacity the database never uses.
3. Forgotten copies. A read replica or a cross region standby running with no matching license.
4. Options spread from a golden image. Partitioning, Advanced Compression or the Diagnostics Pack enabled in an AMI and inherited by every instance built from it.
5. The wrong user floor. The Named User Plus minimum counted against on premises processors instead of the AWS number.

How can you check your own position before Oracle does?

Every one of those findings can be checked with tools you already have. Build the evidence from the running environment, dated, before an auditor asks for script output.

• Threads and cores. Read the CpuOptions block (CoreCount and ThreadsPerCore) from the EC2 DescribeInstances call, and the ProcessorFeatures field from DescribeDBInstances for RDS. On EC2, confirm with lscpu inside the running instance.
• Options in use. Query DBA_FEATURE_USAGE_STATISTICS in every database and reconcile each flagged option to an entitlement or a documented reason it is switched off. Our audit script analysis shows what Oracle's own scripts collect.
• Inventory. List every binary, region, replica and Auto Scaling peak with a date. AWS License Manager can track Oracle license counts by vCPU across accounts.
• Entitlement. Match an active Customer Support Identifier to every deployed license.

Then price the gap yourself. A settlement opened on your numbers lands very differently from one opened on the auditor's spreadsheet. The option prices that drive most findings are listed in our Enterprise Edition options guide.

What discounts does Oracle give when you run Oracle Database on AWS?

Oracle treats AWS as competitive ground and prices accordingly. In the deals we see, Enterprise Edition base licenses close at 60 to 75 percent off list and options at 40 to 70 percent. With discipline, the annual support uplift can be held at 0 against the 8 percent Oracle proposes by default.

Treat the support line as the real negotiation. A license discount is a single event, while the uplift compounds every year you keep the licenses. The wider commercial approach is in Oracle cloud negotiations.

What will the Oracle account team say, and how should you answer?

• "The cloud policy is published, so the count is fixed." Agree, and count to it. Then move the discussion to the price per processor, the support uplift and the terms for the licenses you move.
• "Your migration puts you out of compliance, so buy now." Ask which instances they mean and show your own dated count. A deficit you have already sized and priced leaves little room for pressure.
• "A ULA covers the cloud move." Ask how AWS deployments count at certification. Standard ULA terms often exclude public cloud deployments from the certified count unless that right was negotiated in, so read your own agreement; see our ULA on AWS guide.
• "License included RDS takes compliance off your hands." True for Standard Edition 2 only, and paid for inside every instance hour. Size the running hours first.

What contract terms should you ask for?

• A support uplift cap of 0 for the term. It compounds, so it is worth more than an extra point of license discount.
• A dated reference to the cloud policy version. The policy is not part of your contract, so tie your count to the version in force when you sign.
• Price holds on the options you expect to need. Active Data Guard and Partitioning are the usual ones once replicas and larger tables arrive.
• Audit scope limited to the ordering document. It keeps a review of your AWS databases from spreading to programs outside that order. The audit clause redline has sample wording.

What have we seen in Oracle on AWS reviews in 2024 and 2025?

Across roughly 30 to 40 Oracle on AWS environments I reviewed in 2024 and 2025, the licensed core count ran 20 to 35 percent above what the workload required. The median reduction was 28 percent of licensed cores. It came from right sizing: memory dense instance families, correct vCPU counts, and options stripped back to what was used.

Why we do not count the migration itself as a saving

The usual advice is that moving Oracle to AWS cuts cost because you retire hardware and stop paying for a data center. On the Oracle line we disagree. Relocating an unchanged server loses the core factor, as the table at the top of this page shows, and none of the savings in our reviews came from the relocation.

The migrations that saved money did it through sizing, plus moving steady state databases to bring your own license. Buyers who treated the migration as an infrastructure project and left Oracle sizing to the platform team paid twice: once in licenses and again in the support stream that follows them every year.

AWS does not change what Oracle charges per core. It changes how many cores Oracle is allowed to count.

Give the license owner a seat in every instance selection decision. A platform engineer picking an r6i because the team already uses it can commit the company to four times the Oracle cost of an x2iedn with the same memory.

What to do next

1. Recount from the vCPU. Rebuild the processor count from each instance's vCPUs and thread setting. A count carried forward from on premises is almost always wrong, usually by half.
2. Test memory bound databases on a denser family. Price each one on a memory dense family and with reduced active cores. This is the largest single saving available, and it needs no negotiation.
3. Run the options scan and strip what is unused. Every flagged option needs an entitlement or a documented reason it is off, especially anything inherited from a golden image.
4. Split RDS on metered hours. Decide license included or your own licenses per database, based on two billing cycles of real running hours.
5. Open the Oracle conversation on your own data. Bring a priced, documented and independently reviewed position before anyone at Oracle asks for script output. The Oracle practice runs these reviews for buyers.

Frequently asked questions