Oracle virtualization licensing. Choose the platform before the audit chooses it.

Oracle licenses a virtual estate by the hardware its software could run on, not the hardware it actually uses. That one sentence decides your platform, your cluster design, and most of what an Oracle audit will ever be worth.

This page is the platform decision. Which hypervisor, which cluster shape, and what each choice costs in licenses before anyone argues about it. For the audit defense on a specific platform, the pages for VMware, Hyper V and Nutanix carry the detail.

Which platform should Oracle actually run on?

Four questions decide it, and only the first one is about technology. Answer them in this order and the platform picks itself.

1. Is the boundary one Oracle names by product? If it is not on Oracle's list, the boundary does not reduce the count no matter how it is engineered.
2. Can you evidence it, dated, from before any audit notice? A boundary you cannot prove is a boundary you do not have.
3. What is the blast radius if Oracle disagrees? Count the cores an auditor could reach on your worst day, not on your architecture diagram.
4. What does it cost to move later? Every platform has an exit price, and it is always higher than the entry price.

The platform options, with Oracle's posture on each

Read the top four rows together. Three of the mainstream hypervisors sit in the same bucket, which is the fact that decides most platform strategies and the one most often missed.

What is the difference between soft and hard partitioning?

Hard partitioning is a boundary Oracle accepts as a way to license a subset of a server. Soft partitioning is any method Oracle does not accept for that purpose, and the consequence is that you license every processor the software could run on.

The list of accepted technologies is short, named by product, and published in Oracle's partitioning policy. It is a policy document rather than a contract term in most agreements, which is the single most useful fact a buyer holds in this argument.

What Oracle accepts as hard partitioning

• Physical domains on the vendor platforms Oracle names.
• Capped logical partitions and containers, including capped IBM LPAR and capped Solaris Zones.
• Oracle Linux KVM and Oracle VM Server configured to Oracle's own core pinning rules.

The full list, what each one requires in practice and the evidence that holds it are on the partitioning policy page and the hard partitioning implementation guide. The rules for counting cores once the boundary is settled are on the core factor page.

Why does Oracle count the whole cluster?

Because on a soft partitioned platform Oracle's position is that the database can run anywhere the hypervisor could place it. Shared storage and live migration are what turn one virtual machine into a claim on every host that can see it.

The capability trap

It is the capability, not the act, that Oracle points to. A virtual machine that has never migrated still counts against every host it could reach unless the reach was removed by design.

Newer platform versions widened that theoretical reach, and Oracle uses each widening to argue broader scope. Which vSphere version you run therefore changes the size of the claim, and that specific argument is covered on the Oracle on VMware page.

What does the platform choice actually cost?

Price it once and the architecture argument ends. Take a realistic mid size estate and count it both ways at Enterprise Edition list, before any discount and before any option.

Intel core factor of 0.5 throughout, rounded up. The annual support on that difference is 2,173,600 USD at 22 percent, every year, for as long as the licenses exist.

Now add the options. A stack of Partitioning, Diagnostics Pack and Tuning Pack adds 24,000 USD per Processor, so the same 208 license gap widens by a further 4,992,000 USD. The option lines are counted at the same quantity as the database underneath them.

The cost nobody puts on the other side of the page

Dedicating a cluster to Oracle is the right licensing answer and it is no longer a free one. Broadcom moved VMware to subscription bundles priced per core, with minimum core counts per processor and per order that have been revised upward.

Check the current minimum before you size a small dedicated cluster. A three host Oracle cluster can now land below the order minimum, which means you buy VMware capacity you will not use in order to save Oracle licenses you would not have owed.

Run both numbers together. The comparison across platforms is on the Broadcom era VMware reference and the Nutanix and VMware cost comparison.

How do you contain Oracle on a shared hypervisor?

Containment is an architecture decision, not a negotiating position. You shrink the set of hardware the database could run on, and then you prove it.

1. Dedicate a cluster to Oracle and nothing else, with its own management boundary.
2. Pin the version and configuration so live migration cannot reach beyond that cluster.
3. Separate storage so the Oracle estate is not visible from outside it.
4. Export the evidence on a schedule and store it somewhere dated and unalterable.

Does separating storage matter?

Yes, and it is the control teams skip. Storage visibility is part of how Oracle argues reach, so separating the array can matter more than separating compute.

An Oracle estate on isolated storage is materially harder to claim across management domains, because the argument that a workload could have been started elsewhere has to survive the fact that the data was not there.

The five artifacts that make a boundary real

• Cluster membership exports, dated, showing which hosts existed in the Oracle cluster at each point in time.
• Storage presentation records showing which arrays and LUNs were visible to which hosts.
• Configuration exports for the settings that constrain placement.
• Change records for every alteration to the above, because gaps get read as concealment.
• A written internal position stating how you read the contract, produced before any notice arrived.

The last one carries more weight than people expect. A position written under audit pressure reads as a defense. The same position written two years earlier reads as a policy.

How do failover, standby and disaster recovery nodes count?

Separately from the cluster argument, and this is where estates quietly overpay and quietly overrun at the same time. Oracle publishes distinct treatments for failover, standby and testing environments.

The three cases and what each one costs

• Failover on shared storage. Oracle permits an unlicensed node in a failover configuration to run for a limited number of separate days in a calendar year. The allowance is ten separate days in the policy as published, and it is counted per calendar year, not per incident.
• Standby and disaster recovery. A standby that is mounted, open or applying redo is a running deployment and it is licensed. Data Guard configurations do not create a free second copy.
• Test and development. There is no free test license on Oracle Database. A development environment on a shared cluster is inside the same cluster argument as production.

Verify the current wording of the failover allowance in the Oracle Database Licensing Information manual for your release before you rely on it. It is one of the few genuinely useful allowances Oracle publishes and it is worded tightly.

Two failure modes we see repeatedly. Estates that license a standby they never open, and estates that run a failover node for a quarter and treat it as covered. The detail is on the disaster recovery licensing page.

Does moving off VMware fix the Oracle problem?

No, and this is the most expensive misunderstanding in the current market. Hyper V and Nutanix AHV sit in exactly the same Oracle bucket as VMware.

Buyers moving off VMware for Broadcom pricing reasons are making a sound decision about one vendor and no decision at all about the other. The Oracle count follows the workload onto the new platform unchanged.

What a platform move does and does not change

Only three rows in that table move the Oracle number, and two of them cost something real elsewhere. Decide the Oracle question and the hypervisor question separately, then price them together. The migration options are compared on the VMware migration alternatives page and the virtualization diversification playbook.

How do you defend a virtualization claim?

With evidence and with the contract, in that order. The opening claim is a position, not a settlement, and it is built on a policy document rather than on your order form.

• Map actual deployment host by host before responding to any script output.
• Document isolation with configuration exports, not assertions.
• Separate policy from contract, and hold the line on what was actually signed.

The Oracle Master Agreement governs what Oracle may verify. The partitioning stance sits outside it in most agreements, and the cloud licensing policy sits outside it too.

Value any genuine shortfall against the Oracle Technology Price List rather than accepting a number, so the remediation discussion starts from a figure you built. Then take a view on your own exposure with the virtualization risk assessment.

Where the common advice on Oracle virtualization is wrong

The common advice in 2026 is to get off VMware, and to treat that migration as the answer to the Oracle exposure as well as to the Broadcom bill. We disagree. A move to Hyper V or to Nutanix AHV changes the Oracle count by exactly zero, because Oracle treats all three as soft partitioning and takes the count on the physical host either way. What moved settlements 30 to 60 percent in the engagements behind this page was never the hypervisor brand. It was a dedicated cluster, a storage boundary, and dated evidence produced before any notice arrived. Fix the boundary first, then choose the hypervisor on price and operations.

It is the capability, not the act, that Oracle counts. Remove the capability and the argument disappears.

White Paper · Oracle

Oracle & VMware Licensing

Bound Oracle licensing on VMware. Read it free.

Read the white paper

What should a buyer do next?

Seven steps, in order, and the first three cost nothing but a week of somebody's time.

1. Map every Oracle instance to its physical host and cluster. Include development, test, disaster recovery and anything a supplier runs for you.
2. Count the worst case. Every core an auditor could argue is reachable today, priced at list. That is your actual exposure number.
3. Identify the clusters that mix Oracle with anything else. Those are where the entire gap lives.
4. Dedicate and isolate a cluster for the Oracle estate, sized against your hypervisor vendor's current core minimums.
5. Pin versions, configuration and storage so placement cannot reach beyond the boundary.
6. Export the isolation evidence and date it, before any audit notice exists. Evidence created afterwards is worth a fraction as much.
7. Have an independent adviser pressure test the boundary against Oracle's policy and against your signed agreement, separately.

If you only do one of these, do the second. Most estates have never put a number on the worst case, and the number is what starts the conversation internally.

Frequently asked questions

Does Oracle accept VMware as hard partitioning?

No. Oracle treats VMware as soft partitioning, so its position is that you license every processor the database could run on rather than only the hosts where it actually runs. The same treatment applies to Hyper V, Nutanix AHV and community KVM builds.

Will moving from VMware to Nutanix or Hyper V reduce my Oracle bill?

No. All three are soft partitioning in Oracle's policy, so the count is taken on the physical host on any of them. A platform migration is a sound decision about your hypervisor vendor and no decision at all about Oracle. Fix the cluster boundary first, then choose the hypervisor on price and operations.

Is Oracle's partitioning policy part of my contract?

Usually not. The partitioning document is a policy, not a contract term in most agreements, and it is not normally referenced in the ordering document either. That distinction is the foundation of defending a virtualization claim, and it changes the conversation from compliance to commercial.

Can live migration increase my Oracle license requirement?

In Oracle's view, yes. Because live migration gives a virtual machine the capability to run on other hosts, Oracle argues those hosts are in scope even if the machine never moved. It is the capability rather than the activity that the claim is built on.

How much does a dedicated Oracle cluster actually save?

On a realistic mid size estate, moving Oracle off an eight host shared cluster onto a three host dedicated cluster cuts the count from 256 Processor licenses to 48. That is 9.88 million USD at Enterprise Edition list and 2.17 million USD a year in support, before options are counted.

Do failover and standby nodes need licenses?

A standby that is mounted, open or applying redo is a licensed deployment. Oracle does publish a failover allowance permitting an unlicensed node in a failover configuration to run for a limited number of separate days per calendar year, ten in the policy as published. Verify the current wording in the Licensing Information manual before relying on it.

What counts as hard partitioning Oracle will accept?

Physical domains on the platforms Oracle names, capped IBM LPAR and capped Solaris Zones, and Oracle Linux KVM or Oracle VM Server configured to Oracle's core pinning rules. The accepted list is short and it is named by product, so anything not on it does not reduce the count regardless of how it is engineered.

How inflated are virtualization audit claims?

In our engagements the opening claim ran three to eight times actual deployment, because Oracle counted every reachable host. Documented isolation reduced the settlement by 30 to 60 percent, with a median around 45 percent. The reduction came from evidence rather than from argument.