Oracle counts every physical core in the live migration domain, not the virtual processors you assigned. The arithmetic, the failover allowance, and the architecture that bounds it.
Oracle classifies Microsoft Hyper V as soft partitioning, which means you license every physical core in every host the database could be scheduled onto, not the virtual processors you assigned. That single sentence drives almost all Hyper V audit exposure, and the ways to contain it are architectural rather than contractual.
This guide is for infrastructure and procurement leaders running Oracle Database or middleware on Microsoft Hyper V in 2026. Read it with the Oracle partitioning policy guide, the virtualized environments licensing guide, and the Oracle Knowledge Hub.
The primary sources behind this page are Oracle's own: the Oracle Partitioning Policy, the Processor Core Factor Table, Licensing Data Recovery Environments, the Oracle Technology Global Price List, and the Oracle Database Licensing Information manual.
Oracle counts physical cores in hardware, not virtual processors in the hypervisor. Hyper V sits in Oracle's soft partitioning category, so every cap you configure inside Windows is ignored for licensing purposes. The count follows wherever the database could run.
Because Oracle's approved list is closed and Hyper V is not on it. The policy names physical domains, capped Solaris containers, capped IBM partitions, and Oracle's own hypervisors configured with Oracle's own pinning method. Microsoft's hypervisor appears nowhere in that set.
Oracle's stated reasoning is that a hypervisor level limit is reversible by an administrator. Whether or not you find that persuasive is beside the point. The list is the test.
No, and this is worth stating because it is the most sophisticated argument buyers bring. Hyper V supports host CPU groups, which pin virtual processors to a defined set of logical processors at the hypervisor level, and that is genuinely stronger than a simple virtual processor cap.
It is still not on Oracle's list. Oracle does not run a technical assessment of how firm your restriction is. It checks a name against a document, and if the name is absent the answer is the whole migration domain.
You license every host the Oracle virtual machine can reach. In a Windows Failover Cluster with Cluster Shared Volumes, that is normally every node, because CSV makes the storage visible to all of them.
Since Windows Server 2012, Hyper V can live migrate a running virtual machine between two hosts with no cluster and no shared storage at all, using only a network connection. Many estates that believe they are running standalone are not.
Auditors have caught up with this. The question in a modern review is not whether you have a cluster, it is whether live migration is enabled on the host and which hosts it is constrained to. Check the setting, then document it.
Multiply the physical core count in scope by the factor for the processor. Most current x86 parts carry 0.5, so a 32 core host needs 16 Processor licenses before any cluster arithmetic. Our core factor reference lists the exceptions that matter.
The gap between a contained design and a shared cluster is rarely small, and it compounds annually through support. Enterprise Edition lists at $47,500 per Processor and support runs at 22 percent of the license fee every year.
Take four failover cluster nodes, two sockets each, 24 cores per socket. That is 48 physical cores per node and 192 across the cluster.
Same database, four Hyper V designs, four bills
| Design | Cores in scope | Processor licenses | EE at list | Annual support |
|---|---|---|---|---|
| Oracle VMs in the shared 4 node cluster, 48 cores per node | 192 | 96 | $4,560,000 | $1,003,200 |
| Dedicated 2 node Oracle cluster, 48 cores per node | 96 | 48 | $2,280,000 | $501,600 |
| Dedicated 2 node Oracle cluster, 16 cores per node | 32 | 16 | $760,000 | $167,200 |
| Single isolated 16 core host, migration disabled | 16 | 8 | $380,000 | $83,600 |
The line that matters is the second one against the third. Same node count, same resilience model, same operational pattern. The only difference is the processor part number, and it is worth $1.52M at list plus $334,400 every year.
Named User Plus carries a minimum of 25 users per Processor license. On the 16 Processor design that is a floor of 400 users at $950 each, or $380,000, half the Processor price.
The break even sits around 50 named users per Processor license. Below that, count your users properly and check whether an application front end forces you to count all downstream users. Above it, Processor licensing wins.
The Hyper V license requirement is set by the blast radius of a live migration, not by the size of the database. Shrink the radius and you shrink the bill.
Windows Server Datacenter is also licensed per physical core, with a minimum of 16 cores per server and 8 per processor. A dense Oracle node therefore gets charged twice, once by Oracle and once by Microsoft.
At roughly $6,000 to $7,000 per 16 core pack at open list, a 48 core node carries about $18,000 to $21,000 of Datacenter licensing before Oracle sees it. Model both vendors on the same node design, in the same spreadsheet, before you sign the hardware order.
Usually not. Oracle publishes a narrow failover allowance, and almost every Hyper V estate we review has read it more generously than it is written.
Read the detail in Oracle's Licensing Data Recovery Environments document. Then look at your actual failover history, because patching windows quietly consume the allowance.
A warm standby running Data Guard is fully licensed from day one, because the database is installed and running. A truly cold server holding only backup files, with no Oracle software running, needs no license until it is used.
The middle case is where the money goes. A disaster recovery cluster that is powered on, patched monthly and periodically tested is not a cold server, and an auditor will treat it as production.
Which secondary servers actually need a license
| Configuration | Oracle software installed? | License needed | The trap |
|---|---|---|---|
| Failover node, same cluster, one shared array | Yes, idle | Free for up to ten separate days per year | Patching and DR tests consume the allowance quietly |
| Second failover node in the same cluster | Yes, idle | Licensable | The allowance covers one node, not the spare pool |
| Data Guard physical standby, mounted or open | Yes, running | Licensable from day one | Active Data Guard adds $11,500 per Processor on top |
| Hyper V Replica target host | Yes, on disk | Treat as licensable | Replication is invisible to the database team that owns the license |
| Cold backup server, no Oracle software running | No | None until used | Powering it on for monthly patching ends the exemption |
Hyper V Replica copies a virtual machine to a secondary host and keeps it ready to start. The replica host holds installed Oracle software on disk, so treat it as licensable and confirm the position before you build it.
Containment is an architecture problem before it is a contract problem. The objective is to make the physical boundary small, permanent and easy for a stranger to verify. Four moves carry most of the saving.
Yes, and isolate them properly. A dedicated Hyper V cluster for Oracle workloads with its own hosts, its own storage and no migration path outward is the only design Oracle reviewers consistently accept as a boundary.
Half measures do not survive. Keeping Oracle in the shared cluster and relying on preferred owner settings puts you back at the full node count the moment an auditor asks what enforces it.
The most valuable decision on this page is the part number. Oracle bills by core, so a 16 core part on an Oracle node costs a third of a 48 core part for the same node count.
Keep configuration exports that prove the migration scope as it stood, dated and stored outside the cluster. A correct position with no evidence loses to a wrong position with evidence.
Sometimes, and it is dismissed too quickly. Standard Edition 2 is licensed per socket at $17,500, so a two node cluster of two socket hosts is four sockets and $70,000 rather than $760,000.
The limits are real. SE2 requires servers with a maximum of two sockets, caps a single database at 16 CPU threads, and excludes Partitioning, Advanced Compression, Diagnostics and Tuning. Check the workload against those before you plan a migration.
The standard advice is to cap virtual processors, set anti affinity rules, document the intent and hope the auditor accepts it. We disagree, and the disagreement is about what is actually being tested. Oracle is not assessing how firmly your restriction is enforced, it is checking whether the technology appears on a closed list, and Hyper V does not. Every hour spent perfecting a soft control is an hour not spent moving the database onto hardware that is genuinely bounded. Buy the smallest Oracle cluster that meets the workload, disable live migration outward, prove it with storage zoning, and treat every hypervisor level setting as an operational convenience rather than a licensing defense.
Documented soft controls still have negotiating value. They move the discussion from an opening claim across the whole estate to a factual argument about where the database ran, and that argument settles materially lower. Treat them as a discount lever, not a defense.
No, and that cuts both ways. The document states on its face that it is provided for educational purposes and is not incorporated into any agreement, which means neither side can simply point at it and declare the matter closed.
Oracle has to argue that your ordering document's definition of Processor, read against your deployment, produces the cluster count. That is an arguable position rather than an automatic one, and it is where an independent reading of your contract earns its fee.
You cannot rely on the policy's non contractual status as a shield while relying on the same document's approved list when it suits you. Pick one reading and hold it consistently, because auditors notice when a customer switches.
Most disputes settle on evidence rather than on interpretation. The buyer with dated proof of a bounded migration domain settles low, and the buyer arguing policy status without evidence settles high, whichever reading of the document is technically correct.
Source: Redress Compliance advisory engagement file, 2024 to 2025.
If you are weighing platforms rather than fixing one, the same rule lands the same way on Nutanix AHV and on VMware. If the answer is to leave the data center, read the Oracle cloud licensing policy and how bring your own license works before you size anything.
Yes for support, no for licensing relief. Oracle supports Oracle Database and many other products running in Windows guests on Hyper V, but it does not recognize Hyper V as a hard partitioning technology. Technical support and license counting are entirely separate questions.
No. Oracle treats Hyper V as soft partitioning, so virtual processor caps do not limit the count. You license every physical core in every host the database could be scheduled onto, after applying the core factor.
Every node the virtual machine can reach, which with Cluster Shared Volumes is normally all of them. The defensible exception is a node the machine is technically prevented from reaching, evidenced by storage zoning and migration configuration rather than by policy or intent.
No. CPU groups pin virtual processors to specific logical processors at the hypervisor level, which is technically stronger than a simple cap, but Oracle's approved list does not include Hyper V in any configuration. The strength of the control is not the test being applied.
Only within a narrow allowance. Oracle permits one unlicensed node in the same cluster sharing one disk array, for a total of ten separate 24 hour periods in a calendar year. Past that the node must be fully licensed, and Data Guard standby databases are licensable from day one.
The standard Processor Core Factor Table applies unchanged. Most current x86 processors carry a factor of 0.5, so two physical cores equal one Processor license, and that factor is applied after you have counted every core in the migration domain.
It changes the unit of counting rather than the classification. SE2 is licensed per socket at $17,500 with a two socket maximum per server, which often produces a far smaller bill, but the deployment is still soft partitioned and the feature limits are strict.
No. Those are operational controls that any administrator can change, and Oracle does not accept them as a licensing boundary. They are useful evidence of where the database actually ran, which helps in negotiation, but they do not shrink the countable estate on their own.
Provide dated configuration evidence that Oracle virtual machines cannot reach unlicensed hosts. Cluster node membership exports, live migration settings, storage zoning and System Center host group topology are the artifacts that hold up, and they must be contemporaneous with the review period.
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.
The standard advice on Oracle and Hyper V is to cap virtual CPUs and hope. We disagree. In the virtualization reviews we have run, vCPU caps never held against Oracle. The buyer side move is to isolate the hosts and prove the boundary with configuration evidence.
500+ enterprise clients. 11 vendor practices. Industry recognized. One conversation can change what you pay for the next three years.
One short note on Oracle licensing moves, audit posture, partitioning policy, and the buyer side levers we are running in client engagements. No noise.
