Contents
Key takeawaysTwo vendors, one set of hostsBroadcom's push to consolidateHow Oracle counts on vSphereThe 10 to 30 host arithmeticDesigning the Oracle clusterNamed User PlusWhat we see in auditsChecking your exposureAnswering Oracle's auditorsContract terms to ask forMigration and hard partitioningWhat to do nextFAQBroadcom's per core VCF pricing rewards fewer, larger clusters, and Oracle licenses every host a database VM could reach. Grow a cluster from 10 to 30 hosts and the Oracle count triples unless you build a dedicated boundary first.
- The cluster is the boundary. Oracle treats VMware as soft partitioning and claims every host a database VM could migrate to through vMotion or DRS.
- Consolidation triples the count. Ten 32 core hosts need 160 Enterprise Edition processor licenses; thirty need 480, about $15 million more at list with no new databases.
- Dense hosts cost the most. One two socket AMD EPYC 9754 server carries 256 cores and 128 processor licenses, $6,080,000 at list before support.
- Rules and pinning do not help. Oracle's policy excludes affinity management and VMware controls, so DRS rules and CPU pinning leave the count unchanged.
- The policy is not your contract. Oracle's partitioning document says it may not be incorporated into any contract, and your agreement's processor definition governs.
- Small hosts suit the Oracle cluster. Broadcom bills every CPU at 16 cores or more, so 8 core CPUs cost the same on VMware as 16 core CPUs and half as much on Oracle.
- Named User Plus follows the cores. The 25 per processor minimum scales with the cluster, so switching metrics does not stop the count from tripling.
Why do Broadcom and Oracle pull your vSphere 8 cluster design in opposite directions?
Broadcom charges for VMware per core, and Oracle charges for its database on every physical core that could run it. Broadcom's subscription pricing rewards fewer, larger, denser clusters. Oracle's soft partitioning position treats the cluster as the license boundary, so every host you fold into a cluster that runs Oracle adds processor licenses.
The two vendors run separate audit cycles against the same servers. If you cut the VMware invoice without redrawing the Oracle boundary, you can multiply the Oracle bill without adding a single database.
- Broadcom. VMware is now sold only as subscription, built around VMware Cloud Foundation (VCF) and priced per core with a minimum of 16 cores per CPU. List prices rose through 2024 and 2025, and opening renewal quotes have run several times the old perpetual cost.
- Oracle. Oracle's rules have not moved. It still licenses Database on the physical cores that could run it, whatever vCPU count the virtual machine is given, and it left its partitioning policy exactly where it was.
Our 2026 map of Oracle on VMware exposure covers the full picture. This page isolates one mechanism: what cluster consolidation on vSphere 8 does to the Oracle count, and the design choices that contain it.
The VMware VCF Renewal: How to Prepare Before Broadcom Names the Price
How does Broadcom's per core pricing push you toward fewer, larger clusters?
Per core subscription pricing makes idle capacity expensive, so the quickest way to shrink a VCF quote is to run the same workloads on fewer licensed cores. In practice that means folding small clusters into a few large ones and replacing older hosts with high core count servers.
The 16 core per CPU minimum adds a second push. A CPU with 8 cores is billed as 16, so infrastructure teams buy denser chips to stop paying for cores they do not have.
What happened to the 72 core minimum?
In late March 2025, VMware distributors told partners that the minimum purchase would rise from 16 to 72 cores per order from April 10, 2025. It covered products from vSphere Foundation to VCF, and customers would pay whichever total was higher. The advice that came with it was to deploy servers with 32 to 64 cores per socket.
Broadcom withdrew the 72 core minimum around the date it was due to start, saying it had never announced a price change, and the 16 core per CPU rule remains. The episode still shaped 2025 hardware refresh plans toward dense hosts, the worst possible shape for a host that runs Oracle under soft partitioning.
Why disabling cores in the BIOS saves nothing
Broadcom's knowledge base states that all cores on a server where the software is installed must be licensed, including any cores deactivated in the BIOS. The tactic some teams used to trim VMware counts no longer works. It never helped with Oracle either, because Oracle counts the physical cores present in the hardware.
How the late renewal penalty shortens your planning time
Broadcom also introduced a penalty of 20 percent of the first year subscription price, applied retroactively, when a subscription is not renewed by its anniversary date. That deadline pushes teams into fast consolidation decisions, and the Oracle impact review is usually the step that gets skipped.
Our guides to negotiating vSphere Foundation and the wider Broadcom VMware licensing reset cover where Broadcom gives ground.
Cut Your VMware vSphere Foundation Cost in 2026
Per core subscription pricing, when VVF beats VCF, and what to check before you sign.
Get the white paper →How does Oracle count licenses for a database on a vSphere 8 cluster?
Oracle counts every physical core on every host the database virtual machine could run on, multiplied by the core factor. Its partitioning policy lists VMware as soft partitioning, and Oracle's position is that it may claim every host a virtual machine could migrate to. With vMotion and DRS enabled, a conservative reading licenses every host in the cluster.
So a database that sits on two hosts of a 30 host cluster is counted on thirty. We cover the reachability argument in why Oracle says the whole cluster could run the database, and the factor itself in our Oracle core factor guide.
Why DRS rules and CPU pinning do not reduce the count
Oracle's policy says soft partitioning may not be used to determine or limit the number of licenses required. It names Affinity Management and VMware among the soft partitioning technologies. In Oracle's reading, DRS host rules, CPU affinity settings and vSphere CPU pinning all fall into that category, so none of them shrink the licensable core count.
If your current position depends on host affinity rules, read what partial cluster licensing can and cannot do before an audit letter arrives. Affinity rules are useful evidence of intent, yet Oracle will not accept them as a license boundary.
Is Oracle's partitioning policy part of your contract?
Not automatically, and that distinction is where audit defense starts. The policy document says it is for educational purposes only, reflects Oracle's policies in effect as of February 14, 2022, and may not be incorporated into any contract. Your license agreement defines the licensable processor.
Check your ordering documents and amendments for any reference to the policy or its web address. Where it is absent, your negotiating and audit defense position is materially stronger, although Oracle will still argue the cluster boundary. Our Oracle partitioning policy guide walks through the wording.
How much does growing a cluster from 10 to 30 hosts add to the Oracle bill?
Tripling the hosts triples the Oracle count. Take a common x86 host with 2 sockets and 16 cores per socket, 32 physical cores in all. Intel x86 carries a core factor of 0.5, so each host needs 32 x 0.5 = 16 processor licenses, and a 10 host DRS cluster of the same specification needs 160.
Consolidate into a 30 host VCF cluster and the obligation on that cluster rises from 160 to 480 processor licenses, without a single new Oracle database instance. At the Enterprise Edition list price of $47,500 per processor, that is roughly $15 million more before support.
| Cluster configuration | Physical cores | Core factor | Processor licenses | EE list ($47,500 each) | Annual support (22 percent) |
|---|---|---|---|---|---|
| Single 2 socket, 32 core host | 32 | 0.5 | 16 | $760,000 | $167,200 |
| 10 host cluster (32 cores each) | 320 | 0.5 | 160 | $7,600,000 | $1,672,000 |
| 30 host cluster after VCF consolidation | 960 | 0.5 | 480 | $22,800,000 | $5,016,000 |
| One 2 socket AMD EPYC 9754 host (128 cores per socket) | 256 | 0.5 | 128 | $6,080,000 | $1,337,600 |
Why the dense host row is the one to watch
A two socket AMD EPYC 9754 host carries 256 physical cores. At a 0.5 factor that is 128 processor licenses for one server, four fifths of the entire 10 host cluster, and it is exactly the server shape a per core VMware quote rewards.
Oracle has not lowered factors to offset rising core density. Its core factor table update of January 28, 2026 added Intel Xeon 69xx, 67xx, 65xx and 63xx variants and several older series, all at 0.5. The dense chips that trim a VMware quote raise the Oracle count at the same rate the core count climbs.
What should a dedicated Oracle cluster look like on vSphere 8?
It should be small, isolated and built from hosts with modest core counts. Keep every Oracle database on a dedicated cluster that is never folded into the general purpose VCF cluster, with no vMotion path and no shared datastore to other hosts. Our walkthrough on designing a dedicated Oracle VMware cluster gives host counts and vMotion boundary rules.
The old perpetual VMware clusters often formed a clear Oracle boundary by accident. Subscription consolidation removes that fence, so you have to build it on purpose, and the VMware redesign is the natural moment to cap the Oracle count.
A worked example: four ways to host the same Oracle workload
Say your Oracle databases need about 48 physical cores of capacity, with room for one host to fail. The table compares four hypothetical designs at Enterprise Edition list with 22 percent support. The last column shows the cores Broadcom bills under its 16 core per CPU minimum.
| Design | Hosts and cores | Oracle processor licenses | EE list | Annual support | VMware cores billed |
|---|---|---|---|---|---|
| A. Oracle VMs in the 30 host general cluster | 30 hosts x 32 cores = 960 | 480 | $22,800,000 | $5,016,000 | Shared with all workloads |
| B. Dedicated cluster of AMD EPYC 9754 hosts | 2 hosts x 256 cores = 512 | 256 | $12,160,000 | $2,675,200 | 512 |
| C. Dedicated cluster, two 16 core CPUs per host | 4 hosts x 32 cores = 128 | 64 | $3,040,000 | $668,800 | 128 |
| D. Dedicated cluster, two 8 core CPUs per host | 4 hosts x 16 cores = 64 | 32 | $1,520,000 | $334,400 | 128 |
Design D keeps 48 cores available after one host failure and costs Broadcom the same 128 cores as design C, because 8 core CPUs are billed at the 16 core minimum. On the Oracle side it halves the count. The gap between designs A and D is $21,280,000 at list, for identical databases.
Why we advise against dense hosts for the Oracle cluster
The usual advice during a VCF refresh is to standardize every cluster on the newest high core count servers. Fewer sockets shrink the VMware quote, and one hardware standard is simpler to run. For the cluster that runs Oracle, we recommend the opposite.
Oracle bills every physical core in that cluster, and most database workloads hit limits in clock speed, memory and storage throughput long before they use 128 cores per socket. Buy high frequency processors with modest core counts for the Oracle hosts, size for failover and nothing more, and put the dense hardware in the general purpose clusters.
What else widens the boundary besides the compute cluster?
Storage and networking can extend reachability past the cluster you designed. The storage vMotion scope trap sets out the isolation requirements, and these are the paths to close:
- Datastores. Present Oracle datastores only to the hosts in the Oracle cluster.
- vMotion network. Put the Oracle cluster's vMotion adapters on their own VLAN with no route to other clusters.
- vCenter. Avoid cross vCenter vMotion for the Oracle cluster, since auditors have argued reachability across vCenter instances too.
- Standby and test copies. Give them their own dedicated hosts, and never park them in the general cluster during a migration.
Does Named User Plus licensing avoid the core inflation?
No, because the Named User Plus (NUP) minimum scales with the same physical core count. The Enterprise Edition minimum is 25 NUP per processor. A 16 core Intel server needs at least 16 x 0.5 x 25 = 200 NUP, roughly $190,000 at list, however few named users you actually have.
At $950 per NUP, that floor costs half the processor price per processor, so NUP can lower the bill where user counts are small. The link to hardware stays. Across a consolidated 30 host cluster the NUP floor triples along with the processor count, and real user numbers are rarely low relative to cores in a general purpose cluster.
What have we seen in Oracle audits of consolidated VMware clusters?
Across roughly 30 to 40 Oracle virtualization engagements in 2024 and 2025, soft partitioned environments faced license claims a median 3.5 times the cores actually running Oracle. More broadly, the average Oracle audit claim in VMware environments runs 3 to 5 times what the customer believed it owed.
In one representative case, an enterprise that believed it was licensed for 8 processors faced a $15 to $20 million back license claim for 120, plus 22 percent annual support from then on. The 112 processor gap is $5,320,000 at Enterprise Edition list, so the claim must also have covered items such as options and back support.
Consolidation does not create the audit multiplier. It enlarges the base the multiplier is applied to.
A 30 host cluster gives Oracle three times the surface of a 10 host cluster to build a claim on. The common failure is one of sequence, where the team consolidates first and discovers the Oracle scope expansion in an audit notice eighteen months later.
How do you check which hosts Oracle could count today?
Start from the Oracle virtual machines and work outward to every host they could reach. Oracle's auditors will ask for vCenter data, so build the same picture first and fix what you can.
- List the Oracle VMs. Match each Oracle Database installation to its VM, cluster and current host. RVTools exports the vInfo, vHost and vCluster views in one pass.
- Count physical cores per host. In PowerCLI, a host's ExtensionData.Hardware.CpuInfo properties NumCpuPackages and NumCpuCores give sockets and physical cores. Record the CPU model for the core factor.
- Map vMotion and storage reach. Note which clusters share vMotion networks, distributed switches, datastores and vCenter, and whether cross vCenter vMotion is configured.
- Keep the migration history. Export vCenter migration events on a schedule, because vCenter keeps event history for a limited time and an auditor may ask where a VM ran last year.
Store the exports with change records and boundary diagrams. Our audit evidence pack guide lists what holds up when Oracle asks.
What will Oracle's auditors say, and how should you answer?
Expect the same few lines in almost every VMware review, and answer each without conceding the count in the first meeting.
- "Every host the VM could run on needs a license." Ask which clause of your agreement says so, and point to the policy's own statement that it is not part of any contract. Then show that no path exists outside the dedicated cluster.
- "DRS rules are soft partitioning." Agree, and explain that you rely on physical isolation of compute, network and storage instead.
- "Your VM ran on another host last year." Ask for the date and the evidence. If a migration did happen, deal with that host for that period only.
- "A ULA or a move to OCI would settle this." Price the dedicated cluster first. A ULA negotiated during an audit usually starts from Oracle's claim figure, which pushes its price up.
Which contract terms should you ask for before you consolidate?
Ask when you are placing an order or renewing support, because that is when Oracle has a reason to agree. None of these terms is standard, so expect to negotiate each one.
- A named environment. An ordering document clause licensing Oracle Database on the hosts of a named cluster, listed by host name or serial number, which turns the boundary into a contract term.
- No reference to the partitioning policy. Keep the policy and its web address out of new ordering documents, so your agreement's processor definition stays the only rule.
- Host replacement rights. The right to replace hosts in the named cluster with the count recalculated on the new hardware only, so refreshes do not reopen scope.
- Agreed audit evidence. Acceptance that vCenter exports of the named cluster, its network and its datastores prove deployment.
Should you migrate off VMware or use hard partitioning instead?
Only after you have priced a dedicated cluster, which often caps the count at lower cost than a full migration. Moving Oracle off VMware does not remove exposure by itself, and OCI may or may not be cheaper depending on your metric and workload. Weigh both in our stay on VMware versus OCI cost comparison.
For teams that need Oracle to license only the cores actually running the database, Oracle approved hard partitioning is the structural fix. It is a different architecture from vSphere and needs its own audit evidence.
What if an Oracle renewal is close?
Redesign the boundary before the true up conversation starts. Retiring shelfware and rebalancing editions, covered in optimizing your Oracle footprint before renewal, works far better once the cluster boundary is tight.
What to do next
- Before the VCF design is final. Treat the VMware redesign and the Oracle count as one decision, with a dedicated Oracle cluster outside the general purpose VCF cluster.
- Before you order hosts. Model the Oracle count for the target host shape. A 64 core per socket server that helps the VMware bill can quadruple the Oracle count on any host it joins that runs a database.
- This quarter. Pull your ordering documents and check whether the partitioning policy is referenced anywhere.
- During the build. Close the storage, vMotion network and vCenter paths that reach beyond the Oracle cluster, and add a licensing check to change control for any host added or cluster merged.
- Before anyone asks. Assemble the evidence pack: host inventory, core counts, boundary diagrams and exported migration history.
- At the next Oracle order or renewal. Ask for the named environment, host replacement and audit evidence terms while Oracle wants the signature.
- If an exit is on the table. Price OCI and hard partitioning against the dedicated cluster before you commit to either.
Frequently asked questions
Does moving Oracle to a bigger vSphere 8 cluster increase my license count?
Yes. Under Oracle's soft partitioning position every host the database VM can reach is licensable, so the count grows linearly with cluster size. Thirty identical 32 core hosts carry three times the Oracle obligation of ten, even if every database stays where it was, and each host added later raises it again.
Can DRS host rules or CPU pinning limit which hosts Oracle counts?
Not under Oracle's stated policy, which treats affinity controls inside VMware as soft partitioning that cannot limit licenses. Keep the rules as supporting evidence, but rely on a physically isolated Oracle cluster with its own vMotion network and datastores, so no migration path to other hosts exists.
Does Broadcom's 72 core minimum still apply, and why did it matter for Oracle?
No. Distributors announced a 72 core minimum per order from April 10, 2025, and Broadcom withdrew it around the date it was due to start. It still mattered because the advice to buy 32 to 64 cores per socket shaped many refresh plans. Every extra physical core on an Oracle host adds half a processor license.
Does switching to Named User Plus avoid the core inflation?
No, although it can lower the bill. The Enterprise Edition minimum of 25 NUP per processor costs about half the processor price, which helps small user populations. The minimum is still calculated from physical cores, so a cluster that triples in size triples the NUP floor as well.
Is Oracle's partitioning policy legally binding on me?
Not by itself. The document describes itself as educational and says it may not be incorporated into any contract, so the processor definition in your agreement governs. Check whether any ordering document references the policy, because a reference changes the analysis and weakens your negotiating position in an audit.
Should I migrate off VMware to solve the Oracle problem?
Not as a first step. A migration relocates the exposure rather than removing it unless the target platform is licensed differently, and OCI savings depend on your metric and workload. Price a dedicated vSphere cluster alongside the exit options, since it is often the cheapest way to cap the count.
Does Oracle count hosts in other clusters managed by the same vCenter?
It can try. Auditors have argued reachability beyond a cluster where vMotion networks, distributed switches or datastores are shared, or where cross vCenter vMotion is configured. Separate networking and storage for the Oracle cluster, and records showing those paths do not exist, keep the discussion at the cluster you designed.