Oracle licenses a boundary, not a virtual machine. The four step ladder Oracle climbs on a VMware estate, what each step costs at published list, and the configuration evidence that stops it.
Oracle does not license a virtual machine. It licenses a boundary, and on VMware it argues the boundary is every host a virtual machine could reach. This guide is about drawing that boundary, proving it, and pricing what it costs when you draw it wrong.
Because Oracle's partitioning policy treats VMware as soft partitioning, and soft partitioning does not reduce licensable scope under that policy. On that reading the licensable unit stops being the virtual machine and becomes the set of physical hosts the virtual machine could be moved to.
The counter position is that the document doing the arguing is not the document you signed. Oracle's partitioning policy carries an explicit statement that it exists for educational purposes and may not be incorporated into any contract.
It says it is guidance. That single paragraph is the reason every defended VMware settlement in our file is argued on contract language and deployment evidence rather than on whether VMware is soft partitioning.
Because most estates cannot answer the follow up question. Oracle asks where the database could have run in the audited period, and the customer discovers that vCenter no longer holds the answer.
The policy is weak paper. It beats no paper at all, every time.
In four escalating steps, each triggered by something specific in your configuration. Knowing which step you are on tells you what to fix and what to produce.
The four step boundary ladder, and what stops each step
| Step | Oracle's claim | What triggers it | Evidence that stops it |
|---|---|---|---|
| 1 | Every host in the cluster | Any Oracle program installed on any host in that cluster | Cluster membership export and host inventory with core counts |
| 2 | Every host in the vCenter | Migration configured or possible between clusters | Enforced host affinity, separated compute, dated change records |
| 3 | Every host that sees the storage | Oracle datastores presented to hosts outside the cluster | Storage presentation, masking and zoning records for the period |
| 4 | Every host in linked vCenters | Linked mode or migration across vCenter instances | vCenter topology diagram and the permission model that blocks it |
The contractual test is where the programs are installed and running. Steps 3 and 4 abandon that test entirely and substitute reachability, which is why they are the softest part of any Oracle claim and the first place to push back.
Oracle's licensing documentation permits an unlicensed failover node in a clustered configuration with shared storage for up to a total of 10 separate days in a calendar year. It is narrow, it is per calendar year, and it does not cover a virtual machine that drifts across hosts under load balancing.
Enough that the boundary is usually the single largest number in an Oracle audit. Here is the arithmetic on a mid sized estate, using only published list prices so you can reproduce it.
At the published Enterprise Edition list of 47,500 dollars per Processor in Oracle's technology price list, the contested position is 18.24 million dollars and the contained position is 2.28 million. Neither number is what you would pay after negotiation, but the ratio between them survives any discount.
Same workload, two boundaries, published list prices
| Line | Contested boundary | Contained boundary |
|---|---|---|
| Physical cores in scope | 768 | 96 |
| Processor licenses after 0.5 core factor | 384 | 48 |
| Enterprise Edition at list | 18.24m dollars | 2.28m dollars |
| Annual support at 22 percent of license fee | 4.01m dollars | 0.50m dollars |
| With Partitioning, Diagnostics and Tuning added | above 27m dollars | above 3.4m dollars |
Because the packs travel with the boundary. Every Processor license Oracle claims for the database is also claimed for Partitioning, Diagnostics Pack, Tuning Pack and anything else the estate uses, and those add roughly half again to the per Processor number at list.
This is also the reason a management pack switched on by a well meaning database administrator becomes a seven figure item at cluster scale rather than a five figure one.
Neither column. Settlements land between them, weighted by what each side can evidence, and are usually converted into forward cloud or license spend that Oracle can book rather than a back dated compliance invoice.
By making the boundary a property of the configuration rather than a property of your intentions. Four controls do the work, and each one closes a specific rung on the ladder above.
VMware distinguishes preferential rules from required ones. A "should run on hosts in group" rule is advisory and the scheduler is free to violate it under resource pressure or during maintenance, which means it does not constrain where the database could run.
Auditors who know vSphere ask for the rule type by name. If your dedicated cluster is enforced by a preferential rule, you have a naming convention rather than a boundary.
Sometimes the honest answer is to leave the hypervisor. Bare metal hosts, an Oracle approved hard partitioning technology, or an engineered system remove the argument rather than winning it.
They made the cheapest Oracle cluster geometry more expensive to build and pushed estates toward the geometry Oracle prefers. That is the part most VMware renewal analyses miss entirely.
The classic containment build used low core count processors in the Oracle cluster, because fewer physical cores means fewer Processor licenses. Under Broadcom's subscription terms each processor is billed at a minimum core count, commonly 16, so an eight core processor is charged as sixteen.
Broadcom documents the counting rules for its subscription editions in its own core counting guidance, and the minimum order size on the Cloud Foundation line was raised again during 2025. The perpetual editions are gone, replaced by the subscription bundles described on the VMware Cloud Foundation page, so read your own quote rather than a summary.
Two dedicated Oracle cluster builds under a 16 core per processor floor
| Build | Physical cores | Oracle Processor licenses | VMware cores billed |
|---|---|---|---|
| 2 hosts, 2 sockets, 8 core processors | 32 | 16 | 64 |
| 2 hosts, 2 sockets, 16 core processors | 64 | 32 | 64 |
| 2 hosts, 2 sockets, 24 core processors | 96 | 48 | 96 |
Read the first row carefully. You pay Broadcom for 32 cores you cannot use, and you save 16 Processor licenses of Oracle Enterprise Edition, which is 760,000 dollars at list. The small host build still wins by a wide margin.
The dangerous Broadcom response is the obvious one. Facing per core and per order minimums, infrastructure teams merge small clusters into fewer large ones to hit thresholds efficiently.
That is exactly the wrong shape. Every cluster merge that pulls Oracle hosts into a general purpose cluster hands Oracle step one of the ladder for free, and it happens inside a project nobody labelled as a licensing project.
With dated configuration exports covering the audited period, produced in a controlled sequence. Assertion is worthless here and screenshots taken this week prove nothing about 2023.
vCenter does not keep task and event history indefinitely. Retention is a configurable database setting and is commonly left at a small window such as 30 days, so by the time an audit asks about a period two years back the record is gone.
Fixing this costs almost nothing. Export tasks and events to a retained location on a monthly schedule and store them with the change tickets.
Our Oracle audit response playbook covers the wider sequence, and the partitioning policy in detail sits alongside this page.
The standard advice is that a dedicated Oracle cluster makes you safe. We disagree, and the engagement record is unambiguous about why. In roughly 20 to 30 VMware matters Fredrik Filipsson worked in 2024 and 2025, several estates had a cluster named for Oracle and still lost the boundary argument, because the affinity rules were preferential, the datastores were visible to other clusters, and vCenter had discarded the migration history months earlier. A dedicated cluster is a design intention. What Oracle prices is what you can evidence, so the deliverable is not the cluster, it is the dated file that proves the cluster behaved as designed for every quarter in the audit reach.
Three cuts of our advisory engagement file frame the size of the boundary question.
Source: Redress Compliance advisory engagement file, 2024 to 2025.
A cluster named Oracle is a naming convention. A cluster with required affinity rules, isolated storage and four years of dated exports is a boundary.
Take these in order. The first three cost nothing but calendar time and close most of the gap before any negotiation opens.
Oracle's audit position says yes for every reachable host, on the basis that VMware is soft partitioning. That position rests on a policy document rather than on your contract, and estates that produce dated cluster evidence have settled the virtualization line at 10 to 25 percent of the opening claim in our file.
No, and the document says so itself. It states that it is provided for educational purposes and may not be incorporated into a contract, which is why the defensible argument runs on your ordering document and your deployment evidence rather than on the policy text.
Only if it is enforced and evidenced. The cluster must use required host affinity rules rather than preferential ones, present Oracle storage to Oracle hosts only, and be backed by dated exports covering the whole audit period.
A "should run on hosts in group" rule is preferential and the scheduler may override it during contention or maintenance. A "must run on hosts in group" rule is required and is enforced, which is what makes it usable as a licensing boundary.
On two socket hosts with 24 core processors it is 768 physical cores, which after the 0.5 x86 core factor converts to 384 Processor licenses. At the published Enterprise Edition list price of 47,500 dollars that is 18.24 million dollars before options and before the 22 percent support line.
Indirectly but significantly. Per processor core minimums make small host Oracle clusters more expensive to license from Broadcom, and the pressure to consolidate into fewer large clusters widens the boundary Oracle can argue, so the VMware redesign needs an Oracle owner in the room.
For the full audit reach in your agreement, which is usually longer than the default vCenter retention. Set task and event retention deliberately, export monthly to storage outside vCenter, and keep the change tickets with the exports.
It can. Bare metal, an Oracle approved hard partitioning technology, or an authorized cloud environment each replace the reachability argument with a countable boundary, and the right answer depends on how much unused capacity you are willing to buy to get simplicity.
Hard versus soft partitioning, the cluster wide claim, and how to bound Oracle licensing in a virtual estate.
Used across more than five hundred enterprise engagements. Independent. Buyer side. Built for procurement leaders running the next renewal cycle.
Oracle prices the question by what your VMs could do. You settle it by proving what they did.
500+ enterprise clients. 11 vendor practices. Industry recognized. One conversation can change what you pay for the next three years.
One buyer side briefing a week. Pricing moves, audit signals, and the levers that work. No vendor spin.
