How is Oracle licensed on Azure?
Oracle is licensed on Azure by vCPU, because Azure is one of the authorized clouds named in Oracle's cloud computing policy, and the policy counts two vCPUs with hyper threading enabled as one Oracle processor license. Where a virtual machine presents a vCPU as a full physical core rather than a thread, one vCPU counts as one processor. The on premises core factor table does not apply to deployments on Azure under this policy, so the count is driven directly by the size of the virtual machine you run.
That makes instance sizing the central licensing decision on Azure. An eight vCPU instance with hyper threading is four Oracle processor licenses, and a sixteen vCPU instance is eight. For Named User Plus the standard user minimums apply against the processor count those vCPUs generate. The discipline is to choose a virtual machine shape that matches the licenses already owned, because every additional vCPU adds to the count, and an oversized instance creates a shortfall that an audit will find at list price.
Does the Oracle and Microsoft partnership change licensing?
The Oracle and Microsoft partnership changes the picture only for Oracle Database services delivered through Oracle Cloud Infrastructure inside Azure, which follow OCI licensing rather than the bring your own license vCPU rule. Under that arrangement the database runs on Oracle managed infrastructure colocated with Azure, and the commercial terms are OCI terms. This is a distinct model from deploying Oracle software yourself onto standard Azure virtual machines, and it is important not to conflate the two when assessing a position.
For Oracle software that you install and run on ordinary Azure virtual machines, the cloud computing policy and its vCPU count still govern. The buyer move is to be precise about which model a given workload uses, because the licensing basis differs. A database on an OCI service inside Azure is counted one way, and a database you stood up on an Azure virtual machine is counted another. Mapping each workload to the correct model before an audit prevents Oracle from applying the reading that produces the larger number.
| Deployment | Licensing basis | Buyer move |
|---|---|---|
| Oracle on Azure VM, BYOL | 2 vCPUs per processor | Size to owned licenses |
| OCI database service in Azure | OCI terms | Read the OCI order |
| Autoscaling VM set | Peak vCPU counted | Cap and evidence it |
| Options enabled on the VM | Separately licensable | Confirm what is in use |
Can an Azure deployment trigger an Oracle audit?
An Azure deployment can trigger an Oracle audit, because cloud migration is one of Oracle's recognised audit triggers, alongside virtualization, Java downloads without a subscription, declining support spend and rejected sales proposals. A move to Azure is visible to Oracle and often coincides with the estate changes that draw attention. The response is not to avoid Azure but to settle the licensing position before the workload moves, with the vCPU count modelled and the licenses confirmed in advance.
Exposure concentrates where deployments outrun the records that explain them. An instance resized upward, an autoscaling set that peaks beyond plan, or options enabled on the virtual machine without a license all create findings that surface in an audit. Oracle's collection scripts can read cloud estates broadly, and preliminary findings arrive inflated at list price. The buyer move is to plan the Azure deployment with the count fixed, hold the configuration evidence, and review any script output before it is submitted.
What evidence defends an Azure finding?
The evidence that defends an Azure finding is the platform record of virtual machine sizes, vCPU counts, hyper threading state and autoscaling ceilings, with the dates each setting applied. Azure exposes this configuration data, which means the real footprint is recoverable rather than assumed. That record converts a broad claim into a precise count, because it shows the licensable basis instead of letting Oracle apply the maximum the subscription could in theory have run.
This evidence also feeds the contract argument, because the cloud computing policy is a policy document and not the contract, and the signed agreement governs where they conflict. When a buyer can show the exact vCPU footprint and the dates it held, a finding built on a wider assumption loses its base, and the contract reading carries the rest. An independent line by line review of findings typically cuts claims by 60 to 80 percent, and cloud findings built on peak or whole subscription assumptions are a frequent place for a large reduction.
The next step
This article is part of our Cloud and OCI Licensing cluster. Read the pillar, the Oracle virtualization licensing guide, for the full picture, and these related reads: Oracle on AWS, the licensing rules, and counting vCPUs on AWS and Azure.