Oracle licensing on VMware and other hypervisors. Choose the platform before an audit chooses it.

Oracle licenses a virtualized database by every host it could run on. On VMware, Hyper-V and Nutanix AHV that means the whole cluster, so cluster design sets the license bill and the audit exposure far more than the hypervisor brand does.

Oracle licenses a virtualized server farm by the hardware its database could run on, whatever hardware it happens to use. That rule decides which hypervisor is worth running Oracle on, how you design the cluster around it, and most of what an Oracle audit claim will be worth.

This guide covers the platform decision: which boundaries Oracle accepts, what each cluster design costs in licenses, and how to prove the boundary later. For audit defense on one specific platform, our pages on Oracle on VMware, Oracle on Hyper-V and Oracle on Nutanix carry the detail.

Which virtualization platform should Oracle Database run on?

Run Oracle on a boundary Oracle names by product, or on a dedicated cluster you can prove is sealed off from everything else. Four questions settle the choice, and only the first one is technical. Answer them in this order.

1. Does Oracle name the boundary by product? If the technology is not in Oracle's partitioning policy, no amount of engineering makes it reduce the count.
2. Can you prove it with dated records from before any audit notice? A boundary you cannot evidence will be argued away in the first meeting.
3. How many cores could an auditor reach on your worst day? Count them from vCenter, Failover Cluster Manager or Prism, and ignore the architecture diagram.
4. What will it cost to leave later? Every platform has an exit price, and once Oracle workloads and evidence are built around it, that price is higher than the entry price.

How does Oracle treat each platform?

The table sets out what you license on each option, before any argument about evidence starts.

Read the first four rows together. VMware, Hyper-V and Nutanix AHV all sit in the soft partitioning group, and that single fact settles most platform strategies. It is also the fact most often missed when a hypervisor migration is being planned.

The cloud rows follow a different rulebook. The authorized cloud counting rules and the OCI BYOL comparison cover them in detail.

What is the difference between soft and hard partitioning in Oracle licensing?

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

Oracle names the accepted technologies by product in its partitioning policy, the current version dated February 14, 2022. The document states that it may not be incorporated into any contract, and in most agreements it is not referenced in the ordering document either. That is the most useful fact you hold when a virtualization claim arrives.

Which technologies does Oracle accept as hard partitioning?

• Physical domains on the vendor platforms Oracle names, including Fujitsu PPAR and HP nPar.
• Capped logical partitions and containers, including capped IBM LPAR and Micro-Partitions and capped Solaris Zones.
• Oracle Linux KVM and Oracle VM Server, configured to Oracle's own core pinning rules.
• Older HP-UX partitioning, meaning vPar, Integrity Virtual Machine and Secure Resource Partitions, all capped.

VMware appears by name in the policy's list of soft partitioning examples, next to affinity management and resource managers. Hyper-V, Nutanix AHV and community KVM builds are absent from the accepted list, which puts them in the same group.

What each accepted technology requires in practice, and the evidence that holds it, is on the partitioning policy page and in the hard partitioning implementation guide. Platform specifics are in our pages on IBM LPAR, Solaris Zones and Oracle Linux KVM. Once the boundary is settled, the core factor page explains how the cores inside it are counted.

Why does Oracle count every host in a VMware 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 its data.

Why does a virtual machine that never moved still count?

Oracle points to the capability. A virtual machine that has never migrated still counts against every host it could reach, unless that reach was removed by design and you can show when it was removed. Our Oracle and VMware licensing white paper sets out the full case.

Newer platform versions widened the theoretical reach, and Oracle uses each widening to argue broader scope. Since vSphere 6.0, vMotion can move a running machine between vCenter Server instances, which puts a whole management domain into an opening claim. The version by version argument is on the Oracle on VMware page.

What does the cluster design cost in Oracle licenses?

Pricing it once usually ends the architecture debate. Take a mid size deployment and count it both ways at the Enterprise Edition list price of $47,500 per Processor, before any discount and before any option.

The Intel core factor of 0.5 applies throughout, rounded up. Annual support on the difference is $2,173,600 at 22 percent of the license fee, charged every year for as long as the licenses stay on support.

How do database options widen the gap?

Options are counted at the same Processor quantity as the database underneath them. Partitioning at $11,500, Diagnostics Pack at $7,500 and Tuning Pack at $5,000 add $24,000 per Processor, so the same 208 license gap widens by a further $4,992,000 at list.

Support on those three options is $5,280 per Processor a year, or $1,098,240 on the 208 licenses. Before you assume a pack is unused, query DBA_FEATURE_USAGE_STATISTICS and check CONTROL_MANAGEMENT_PACK_ACCESS, which defaults to DIAGNOSTIC+TUNING on Enterprise Edition. Our Diagnostics and Tuning Pack guide shows how the usage gets recorded.

What does a dedicated cluster cost on the VMware side?

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 a minimum of 16 cores licensed for every physical CPU and order level minimums that have been revised upward.

The per CPU floor interacts with Oracle's count. Say you build the three host Oracle cluster with two 12 core CPUs per host instead of two 16 core CPUs. Oracle's count falls, while VMware still bills 16 cores for each of the six CPUs.

The smaller CPUs save 12 Oracle licenses, $570,000 at list, and change nothing on the VMware invoice. Before you size a small dedicated cluster, check the current order minimum with your reseller as well.

A three host Oracle cluster can land below that minimum, and then you buy VMware capacity you will not use in order to save Oracle licenses you would not have owed. Run both numbers together, using the Broadcom era VMware reference, the 2026 Oracle on VMware update and our Nutanix and VMware cost comparison.

How do you contain Oracle licensing on a shared hypervisor?

You shrink the set of hardware the database could run on, and then you prove it. Containment is an architecture decision made by your infrastructure team, and it is settled long before anyone negotiates.

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 datastores are not visible from outside it.
4. Export the evidence on a schedule and store it somewhere dated and unalterable.

The detailed design, host by host, is in our dedicated VMware cluster containment guide.

Does separating storage matter?

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

When the Oracle datastores are presented only to the Oracle hosts, a claim across management domains gets much harder to sustain. The argument that the database could have been started elsewhere has to survive the fact that its data was never there.

Which records make a boundary hold up in an audit?

• Cluster membership exports, dated, showing which hosts were 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 written position carries more weight than most teams expect.

A position written under audit pressure reads as a defense. The same position written two years earlier reads as a policy.

How do you check your own position?

Pull the same data an auditor would ask for, from the tools you already run. Keep each export with its date and the name of the person who ran it.

How do failover, standby and disaster recovery nodes count for Oracle?

They are treated apart from the cluster argument, and this is where companies overpay and overrun at the same time. Oracle publishes distinct rules for failover, standby and testing, mostly in its data recovery licensing policy.

How does each case count?

• Failover on shared storage. An unlicensed spare node in a failover cluster may run the licensed programs for up to ten separate 24 hour periods in a calendar year. The nodes must share one logical disk array in a single data center, and the ten periods are a yearly total across every 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, and remote mirroring does not qualify for the failover allowance.
• Backup restore tests. To test physical copies of backups, the policy allows you to run a program on an unlicensed server, including one in a cloud, up to 4 times a calendar year, with no test longer than 2 days. The right covers backup testing only, so a full DR rehearsal on a mirrored or standby site is licensed in the normal way.
• Test and development. There is no free test license for nonproduction copies of Oracle Database. A development environment on a shared cluster sits inside the same cluster argument as production.

Verify the current wording for your release in the Oracle Database 19c Licensing Information manual before you rely on it. The failover allowance is one of the few useful allowances Oracle publishes, and it is worded tightly.

Which failover mistakes do we see most?

Two errors come up repeatedly. Some companies buy full licenses for a spare failover node that only ever runs inside the allowance. Others keep production on the failover node for a whole quarter after an outage and treat it as covered, when the allowance was used up in the first fortnight.

Both are fixable with a log of when each node ran Oracle. The detail is on our disaster recovery licensing page, and cloud based recovery sites are covered in Oracle disaster recovery in the cloud.

Does moving off VMware fix the Oracle licensing problem?

No. Hyper-V and Nutanix AHV sit in exactly the same Oracle group as VMware, so the Oracle count follows the workload onto the new platform unchanged. This is the most expensive misunderstanding in the current market.

Companies leaving VMware for Broadcom pricing reasons are making a sound decision about one vendor and no decision at all about the other.

What does a platform change actually change?

Three rows reliably reduce the Oracle number, and two of those cost something real elsewhere. The cloud row changes the counting rule, which can cut or raise the count. Decide the Oracle and hypervisor questions separately, then price them together. The options are compared on the VMware migration alternatives page and in our virtualization diversification guide for CIOs.

Why we would not lead with a VMware exit

The common advice in 2026 is to get off VMware and count that migration as the answer to the Oracle exposure as well as the Broadcom bill. We disagree. A migration to Hyper-V or Nutanix AHV changes the Oracle count by exactly zero, because Oracle treats all three as soft partitioning and counts the physical host either way.

The settlement reductions in our engagements never came from the hypervisor brand. They came from 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.

What have we seen in Oracle virtualization audits in 2024 and 2025?

Across roughly 35 to 45 Oracle audit and advisory engagements I worked between 2024 and 2025, the shape of the virtual environment was the largest single driver of the opening claim. Three patterns came up again and again.

• Management domain claims. Opening claims counted every host reachable from a single management domain, which inflated the demand to three to eight times actual deployment.
• No isolation at all. In those environments, one Oracle virtual machine put 20 or more hosts in scope.
• Documented boundaries. Once isolation and version boundaries were documented, settlements fell 30 to 60 percent, with a median around 45 percent.

The reductions came from records the customer could produce, rather than from argument. Where the records had to be rebuilt after the notice, they carried visibly less weight.

How do you defend an Oracle virtualization audit claim?

Lead with evidence of where the database could actually run, then hold Oracle to the contract. The opening claim is Oracle's position, and it rests on a policy document that is not part of your order form.

• Map actual deployment host by host before you respond to any script output.
• Document isolation with configuration exports, since a written assurance that nothing moved carries no weight.
• Separate policy from contract, and hold Oracle to what was actually signed.

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

Value any real shortfall yourself against the Oracle Technology Price List, so the remediation discussion starts from a figure you built. Then take a view on your own exposure with the virtualization risk assessment. If a notice has already arrived, start with what to do when the audit letter lands.

What will Oracle's auditors say, and how should you answer?

• "Every host in vCenter is in scope." Ask which clause of your agreement says so. The Processor definition counts cores where the programs are installed or running, and the policy that extends it to reachable hosts is expressly not part of any contract.
• "Live migration was possible, so the hosts count." Show the dated cluster, storage and configuration exports proving the reach did not exist, and ask Oracle to identify a host outside the Oracle cluster that could see the datastores.
• "Please run our collection scripts across the whole vCenter." Offer the output for the Oracle cluster and the hosts running Oracle programs. Data collection should follow the audit clause, which covers your use of Oracle programs and does not extend to an inventory of your hypervisor.
• "A cloud or unlimited license purchase would close this out." Value the shortfall first at list and at your discount. Consider any purchase only after the number is agreed, and never as the way to agree it.

What to do next

Work through these seven steps in order. The first three take about a week of one person's time, and if you do only one, do the second. Most companies have never put a number on their worst case, and that number is what starts the internal conversation.

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. Price every core an auditor could argue is reachable today at list. That is your real exposure number.
3. Find the clusters that mix Oracle with anything else. Those are where the entire gap lives.
4. Dedicate and isolate a cluster for Oracle, 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 test the boundary against Oracle's policy and, separately, against your signed agreement.

Frequently asked questions