Why 23ai Creates New Licensing Complexity

Oracle Database 23ai introduced AI Vector Search, True Cache, and Globally Distributed Autonomous Database — three features that can trigger additional licence obligations the moment they are activated, regardless of whether you plan to use them commercially. With AI Vector Search, JSON Relational Duality, True Cache, SQL Firewall, and dozens of other new capabilities landing simultaneously, procurement and architecture teams face the challenge of mapping each new feature against the licence entitlements they hold.

The situation is made more complex by 23ai's limited on-premises availability as of early 2026. General availability for non-Exadata, non-ODA commodity hardware remains pending. This means most enterprise customers considering 23ai are evaluating either a cloud deployment or an Exadata-based upgrade — both of which carry their own distinct licensing rules. Understanding those rules before committing to a deployment architecture is essential.

Oracle's licensing framework is designed to capture value from every capability enhancement. Understanding which 23ai features are included in your existing entitlements and which require new licence purchases determines whether a 23ai migration delivers net cost savings or unexpected compliance cost. See our comprehensive Oracle Database 23ai Licensing Guide for a full analysis of editions, metrics, and options.

Implication 1: AI Features Are Included — But Options Are Not

Oracle made a strategic decision to include the headline AI capabilities of 23ai — AI Vector Search, Select AI, and JSON Relational Duality — within the base Enterprise Edition and Standard Edition 2 licences at no additional cost. This is a deliberate competitive positioning move against cloud-native vector databases and AI data platforms, and it is genuinely favourable for customers who already hold Enterprise Edition licences.

However, the key word is "headline." The AI features themselves are included, but the architectural patterns required to make AI workloads performant at scale often rely on paid options. Consider the following scenarios:

  • Partitioned vector tables: Large-scale AI Vector Search implementations frequently use table partitioning to manage embedding stores efficiently. If your AI workload requires Oracle Partitioning to deliver acceptable query performance, a Partitioning option licence ($11,500 per processor) is required — the AI Vector Search feature does not include Partitioning.
  • In-Memory analytical acceleration: Customers combining vector search with analytical queries may enable the In-Memory column store for performance. That activation requires the Database In-Memory option ($23,000 per processor) regardless of whether the use case is AI-driven.
  • Diagnostic Pack activation: Performance monitoring of AI workloads through Oracle Enterprise Manager's AWR and ADDM facilities requires the Diagnostic Pack licence. If your DBA team monitors 23ai performance through OEM without holding the Diagnostic Pack licence, Oracle's audit scripts will surface that as a violation.

Practical Implication

Before designing your 23ai AI workload architecture, map every Oracle feature your implementation will use — not just the AI features — against your current licence entitlements. Options triggered by AI performance patterns are the most common source of unbudgeted compliance cost in 23ai deployments.

Implication 2: Upgrade Path Restrictions Create Interim Licence Risk

Oracle Database 23ai accepts direct upgrades only from Oracle Database 19c (version 19.3 or higher) and Oracle Database 21c (version 21.3 or higher). Customers running Oracle 12c, 18c, or older versions must first upgrade to an intermediate release before migrating to 23ai. This creates an extended migration timeline and, critically, a period where both the source and interim database versions are running simultaneously.

Oracle's standard licence terms do not include an automatic parallel-running allowance. If the source database remains operational during the 23ai deployment — even in a read-only or standby state — Oracle can treat those processors as separately required licences unless specific migration rights have been negotiated in your contract. Organisations that have not addressed this in their master agreement face the prospect of needing to purchase duplicate licences for the duration of the migration window, which for complex enterprise deployments can run from three to twelve months.

The practical action is to review your Oracle master agreement for any parallel-use or migration rights clauses before beginning any 23ai upgrade project, and to negotiate those rights explicitly if they are absent. Redress Compliance has negotiated migration rights provisions for clients that provide a defined parallel-running window without requiring additional processor licences — but these provisions must be secured before the project starts, not after Oracle has observed the configuration change.

Implication 3: Cloud Deployment Licensing Is Materially Different

Because general on-premises availability of 23ai for commodity hardware remains pending, many enterprise customers considering 23ai are evaluating Oracle Cloud Infrastructure (OCI) or Oracle's co-location services (Database@Azure, Database@AWS, Database@Google Cloud). Each option carries distinct licensing implications.

On OCI, customers can deploy 23ai under Bring Your Own Licence (BYOL), applying existing perpetual on-premises entitlements. OCI measures compute in OCPUs (one OCPU = one processor licence under Oracle's rules), which aligns with the on-premises processor metric. This makes OCI the least commercially risky cloud deployment path for customers with existing perpetual licences. Under License Included pricing on OCI, the database licence cost is embedded in the hourly or monthly consumption rate — this is simpler to budget but typically more expensive per unit than BYOL if you hold adequate on-premises entitlements.

For AWS, Azure, and Google Cloud deployments outside of Oracle's co-location services, Oracle's Cloud Licensing Policy applies. This policy requires that customers licence all vCPUs on the physical host running Oracle unless approved hard partitioning is in use. Most standard cloud instance types do not constitute approved hard partitions under Oracle's policy. The result is that organisations running 23ai on standard EC2 or Azure VM instances may be substantially under-licensed relative to Oracle's counting methodology — a risk that can translate into seven-figure audit claims in large deployments.

Implication 4: Support Costs Compound Faster Than Most Budgets Assume

Oracle's annual support fee increases by 8% per year. This is compounded, not simple, and applies to the base licence support fee as well as to any options you hold. For organisations that have accumulated a substantial Oracle Database estate over many years — multiple processor licences across several EE deployments, with RAC, Partitioning, and other options — the compounding effect of 8% annual support increases is one of the largest unbudgeted costs in their IT spend.

A 23ai migration is an appropriate moment to conduct a full Oracle licence and support cost review, because it represents a natural re-evaluation of your Oracle estate. The questions to ask are: Are we still using all the options we are paying support for? Can we consolidate multiple smaller Oracle deployments into fewer, larger CDBs on 23ai and reduce our total processor count? Are there workloads that could migrate off Oracle entirely, reducing the support base on which 8% annual increases compound?

"Oracle's 8% annual support increase is not a negotiated outcome — it is a standard contract term. The only way to reduce its impact is to reduce the size of the support-bearing licence estate."

Implication 5: The Multitenant Free Limit Requires Governance

Oracle Database 23ai, like its predecessors since 19c, permits up to three user-created PDBs per CDB without requiring the Multitenant option licence. This is a genuine licence inclusion that many customers under-exploit during consolidation projects. However, the free limit requires active governance to maintain: Oracle does not enforce a hard technical block on creating a fourth PDB without the licence, meaning that the compliance violation is silent until an audit surfaces it.

For organisations planning 23ai consolidation projects — migrating multiple standalone Oracle databases into shared CDBs to reduce processor licence counts — the three-PDB free limit means careful planning of CDB topology is required. Three PDBs per CDB is often insufficient for large consolidation projects, and the decision to exceed that limit must be accompanied by a Multitenant option purchase and budget approval, not left to individual DBAs creating PDBs on demand. See our detailed article on Oracle Database container licensing for the full governance framework.

Implication 6: 23ai Free Edition Is Not a Production Option

Oracle Database 23ai Free is available for download without licence fees and includes the full set of 23ai features including AI Vector Search. The free edition is subject to hard resource limits: 2 CPUs for foreground processes, 2 GB of RAM, and 12 GB of user data. More critically, the free edition is licensed for development, testing, and educational use only. Deploying 23ai Free in a production environment — even a small one — is a licensing violation. This is a point worth reinforcing to development teams that may deploy 23ai Free for prototyping and then leave it in place as a "lightweight" production instance. Oracle has no technical block on this usage pattern, but it is out of licence scope and represents audit exposure.

What Enterprise Teams Should Do Now

Before committing to a 23ai deployment architecture, enterprise licensing and procurement teams should take the following steps. First, conduct an entitlement review — document all Oracle Database licences currently held, the editions and options they cover, and whether they include migration rights or parallel-use clauses. Second, map your planned 23ai architecture against those entitlements — identify which new features and options will be in use and whether they are covered. Third, model the upgrade path — assess parallel-running duration, whether an intermediate upgrade hop is required, and what licence implications that creates. Fourth, engage independent advice — Oracle's sales team will frame every commercial conversation around expansion and new purchases; having independent analysis before those conversations ensures you are not paying for licences you do not need or signing contracts with unfavourable terms.

Redress Compliance provides independent, buyer-side Oracle licensing advisory with 20+ years of Oracle Database experience. Contact our Oracle advisory team to discuss your 23ai deployment and licence position.

Planning a 23ai deployment or migration?

Get independent Oracle licensing analysis before you commit to an architecture or a contract.
Talk to an Advisor →