Oracle meters cloud API Gateway at roughly $3.00 per million calls, but every on-premise gateway node is a WebLogic domain that can pull $45,000 per processor into scope
The same product name carries two irreconcilable metrics. In OCI, the meter is request volume with a 1 million call monthly free tier and no documented per-gateway charge, so scale-out is close to free. On-premise and hybrid gateway nodes are WebLogic domains, each with an admin server and a managed server, and the Processor definition (installed and/or running) captures every idle standby node behind your load balancer. Decide which side of that line your topology sits on before you add the next node.
Prepared by Redress Compliance · August 23, 2026 · Oracle middleware and integration advisory. Price list analysis and renewal engagements, 2024 to 2026.
Executive summary
The cloud meter is request volume at roughly $3.00 per million calls per month, with the first 1 million calls free, and horizontal gateway count is not the billed unit.
That means an OCI-native topology can run five gateways across three availability domains and pay the same as one gateway carrying identical traffic, which inverts the on-premise cost logic entirely.
One third-party guide asserts a small monthly per-deployment fee stacked on top of the per-request charge, and that single unverified claim decides whether your scale-out plan is free or metered.
Confirm it against the live Oracle Cloud Price List for your exact region before you commit to a deployment-per-microservice design pattern, because a per-deployment fee turns 200 API deployments into a fixed monthly floor.
Every self-hosted gateway node is a WebLogic domain running an admin server and a managed server, so a five-node load-balanced tier is five domains, and Processor counting applies to all of them at total cores multiplied by the core factor.
With WebLogic Suite at $45,000 per processor on the April 2026 list, a modest four-core-per-node tier at a 0.5 factor is ten processors, or $450,000 list before any gateway line item is added.
Processor is the only realistic metric for gateway-fronted APIs because named user counting collapses the moment external or machine consumers appear, and the crossover math confirms it past roughly 50 users per processor.
Oracle knows this, so do not spend negotiation capital arguing for Named User Plus on an internet-facing gateway tier; spend it on core factors, node count, and the standby exclusion instead.
Oracle is actively retiring standalone integration PaaS SKUs, with OCI Process Automation unavailable and unsupported after April 3, 2026 and Data Integration Platform Cloud dead as of December 12, 2024.
Any multi-year commitment to a standalone API management SKU should carry a written migration and price-protection clause, because Oracle's stated direction is to fold capability into Oracle Integration rather than sustain separate services.
The two meters: request volume in OCI, processor cores on-premise
Oracle sells two products that share a name and share almost nothing else.
In OCI, the API Gateway meter is request volume: roughly $3.00 per million API calls per month, with the first 1 million calls per month free, and (per the current third-party pricing guides) no separate charge for gateway instances, deployments, or idle runtime.
Oracle's own product page describes the gateway as a highly available virtual network appliance that receives API calls at scale inside a regional VCN subnet, either private or internet-facing. That framing matters because it invites a per-appliance mental model that the meter does not support.
One caveat I flag in every negotiation: at least one licensing guide asserts Oracle charges a small monthly per-deployment fee on top of the per-request rate. I have seen that claim repeated since 2020 and I have also seen it contradicted.
Verify it against the live regional price list before you design a topology around it, because it decides whether scale-out is free or metered. Usage plans are not a third meter. They are the quota and entitlement layer used to define access tiers, manage subscriptions, and monitor traffic per plan.
They govern who consumes, not what Oracle invoices. On-premise and hybrid gateway nodes obey the Processor definition instead: all processors where the programs are installed and/or running, counted as total cores multiplied by the core processor licensing factor.
Because gateway-fronted APIs make user counting impossible, Named User Plus is rarely defensible, and Processor becomes the default by elimination rather than by choice.
| Dimension | OCI API Gateway | On-premise / hybrid gateway node |
|---|---|---|
| Billable unit | API calls processed | Processors (cores x core factor) |
| Rate anchor | ~$3.00 per million calls/month | WebLogic Suite $45,000/proc (Apr 16, 2026 list) |
| Free allowance | First 1M calls/month | None |
| Cost of an idle node | Zero (no documented instance charge) | Full processor count, installed and/or running |
| Adjacent stack exposure | None | Service Bus $35,000, SOA Suite $57,500 per proc |
| Trigger for growth | Traffic volume | Node count and core count |
| Contract instrument | OCI Universal Credits / PAYG | Perpetual or subscription plus 22% support |
Read the last two rows together. Under the OCI meter, cost tracks the thing your business actually values, which is consumption. Under the Processor meter, cost tracks an architecture decision that your infrastructure team makes without ever opening a purchase order.
A team that doubles gateway nodes for resilience has just doubled the licensable footprint at $45,000 per processor list, and the traffic has not moved at all.
Anchor every discussion against the April 16, 2026 Technology Global Price List. WebLogic Suite sits at $45,000 per processor and is a prerequisite base license, not an optional add-on, wherever SOA-class options are in play. Service Bus is $35,000 and SOA Suite $57,500 per processor.
Support at 22% makes those numbers permanent, and Matching Service Levels blocks the partial drop you will eventually want. Model the support stream before you agree to the license line, not after.
Our Oracle Integration and SOA licensing buyer guide works the full middleware stack that a self-hosted gateway drags into scope.
Scale-out topology: why five nodes cost five times one on-premise and nothing extra in OCI
The inflation mechanism is documented by Oracle itself. Installing a gateway node creates a WebLogic domain with a single admin server and a single managed server backed by a lightweight Java database, and the reference topology explicitly places multiple lightweight nodes behind a load balancer.
A single API can be deployed to many gateways while retaining one identity, which is excellent engineering and terrible license economics. Your footprint is driven by horizontal node count, not by throughput.
That means an availability decision, made by a platform engineer at 4pm on a Tuesday, is a licensing decision with a six-figure tail.
Work the arithmetic. Five gateway nodes, four cores each, on Intel x86 with a 0.5 core processor licensing factor: 20 cores times 0.5 equals 10 processors.
At WebLogic Suite list of $45,000, that is $450,000 in license plus roughly $99,000 per year in support, before Service Bus or SOA Suite touch the estate. Collapse to a single node and the same math yields 2 processors. The traffic is identical in both cases.
Under the OCI request meter, the resilient configuration costs the same as the single-node one, because the meter never counts appliances.
The trap is that both patterns look like the same architecture diagram. Buyers routinely design a hybrid estate where the resilient tier runs on-premise for latency or data residency reasons, then discover the redundancy multiplier at audit. Decide the metric first, then the topology.
Our SOA Suite processor versus NUP analysis shows the same cluster-count problem in the adjacent product, and the counting logic transfers directly.
Avoid the Oracle Exadata core license trap
Oracle Exadata can lock you into full core licensing across X9M, X10M, and Cloud at Customer. The buyer side strategy to size the platform and cut the bill.
Get the white paper →The idle standby trap: installed and/or running captures nodes you never route traffic to
Four words in the Processor definition do all the damage: installed and/or running.
The Oracle Technology Global Price List (USA, dated August 3, 2026) defines Processor as all processors where the Oracle programs are installed and/or running, with the count derived from total cores multiplied by the core processor licensing factor.
There is no traffic clause, no active/passive clause, no "in production" clause. A gateway node that has never served a single request is licensable the moment the WebLogic domain sits on disk.
Since each gateway node is a full domain (one admin server, one managed server, a lightweight Java database), the binary footprint is identical on the node that carries 90% of your traffic and the node that exists purely so your DR runbook has somewhere to fail over to.
In 25 years of taking these apart, the pattern is depressingly consistent: the architecture diagram shows two active nodes, and the discovery scripts find eleven installations.
Passive standby pairs, blue-green deployment slots that sit idle between releases, DR site replicas kept warm for RTO reasons, and pre-baked container images all land on the same side of the definition.
A topology engineered for uptime routinely doubles or triples the licensable count against what the traffic graph implies.
Containers make it worse, and Oracle has already published the precedent. Verrazzano terms require licensing all processors in the nodes of the Kubernetes clusters where the images are pulled.
Read that literally and apply it to gateway pods: running self-hosted gateway nodes on a shared, autoscaling cluster means the licensable boundary is the cluster, not the pod, and the boundary moves every time the scheduler adds a worker.
That is the worst possible exposure shape at $45,000 per processor for the WebLogic Suite prerequisite. Three containment moves work. First, dedicated node pools with taints and tolerations so gateway images can only land on a fixed, small, documented set of workers.
Second, image registry controls, because the Verrazzano wording turns on where images are pulled, so restrict pull credentials to the licensed pool and log every pull.
Third, a written topology attestation signed by your infrastructure lead each quarter, recording node count, core counts, and the pull boundary, so that when the audit letter arrives you produce a record rather than reconstruct a defense.
The same discipline applies to the WebLogic sprawl described in our SOA Suite processor versus NUP cluster count analysis.
The table your architects drew is a traffic diagram.
The table Oracle's audit team builds is an installation inventory, and the two only match by accident. Every uptime decision you made for free (a warm standby, a second region, a pre-pulled image) is a purchase order under the Processor definition.
Practically, this means the node count must become a governed number with an owner, not an emergent property of the deployment pipeline. Cap it in writing, review it before each release, and treat any request to add a gateway node as a procurement event rather than a ticket.
Analysis: Oracle sells you the cloud meter and audits you on the on-premise one
Treat the two meters as two different commercial instruments with two different jobs. The request meter in OCI (roughly $3.00 per million calls, with a 1 million call monthly free tier and no documented per-gateway charge) is a customer acquisition instrument.
It is priced so that no architect ever has to ask permission to try it. The Processor meter on the self-hosted side, anchored by WebLogic Suite at $45,000 per processor on the April 16, 2026 Technology Global Price List, is a revenue retention instrument.
It is priced so that leaving is expensive and so that growth in node count converts directly into perpetual license and 22% support. Neither price is wrong on its own terms. They are simply solving different problems for Oracle, and the enterprise that straddles them ends up funding both.
The straddle is not a mistake anyone makes; it is the operational default. The sequence repeats almost verbatim across engagements. A platform team pilots API management in OCI because it costs nothing meaningful and requires no procurement conversation. The pilot succeeds.
Then the constraints arrive: latency at a branch or factory edge, a data residency rule that will not permit request bodies to transit a region, a partner integration that must terminate inside the corporate network, or a regulator who wants the enforcement point on infrastructure you control.
The answer is self-hosted gateway nodes at the edge.
Only at that point, usually after the architecture is committed and the deployment automation is written, does the team discover that WebLogic Suite is a prerequisite base license rather than an optional add-on, in exactly the same way SOA Suite class options require it.
The commercial conversation now happens after the technical decision, which is the worst possible order.
The asymmetry between near zero per gateway in cloud and $45,000 per processor on-premise is not an artifact of costing. Nothing about running a JVM in an Oracle data center is thousands of times cheaper than running one in yours.
It is a gradient, deliberately shaped, designed to make the cloud path feel weightless and the self-hosted path feel like a capital project. Every gradient has a direction, and Oracle's account teams are compensated on movement along it.
Now invert it, because this is where the leverage sits. Oracle wants the consumption story more than it wants your perpetual renewal, and consumption revenue is what the field is measured on. That makes the perpetual side tradeable.
A buyer who can credibly say "we will commit annual OCI consumption at this level, and in exchange the self-hosted gateway estate is capped at N processors with fixed support and no true-up on node count" is offering the one thing the rep needs in the quarter he needs it.
We have seen that trade produce concessions on the perpetual side that no amount of discount argument would have moved on its own. The same dynamic governs the message-pack side of the portfolio, as covered in our Oracle Integration Cloud message-pack overage guide.
The failure mode is funding both meters simultaneously without a written boundary between them. That is what an audit finds: a healthy OCI consumption line, a scattered self-hosted estate that nobody counted, and a Processor definition that captures all of it.
The consumption spend buys you no defense whatsoever against the perpetual claim.
So do two things with the leverage. Refuse to pay both meters at once as an open-ended arrangement: any OCI commitment should be explicitly contingent on a capped, named self-hosted footprint.
And make the node count a negotiated, capped, documented number in the ordering document, with the cap expressed in processors after core factor, the permitted locations named, and standby and DR nodes either counted in or expressly excluded in writing.
A node count that is an architectural side effect is a liability. A node count that is a contract term is a budget line.
Retirement risk: what happens when Oracle folds the SKU into Oracle Integration
Treat product lifecycle as a licensing exposure, not a platform engineering inconvenience.
Oracle has already run this playbook twice in the adjacent integration portfolio. OCI Process Automation is unavailable and unsupported after April 3, 2026.
Covering SKUs B95504 (User) and B95505 (Execution Pack), with migration to Oracle Integration presented as the only forward path. Data Integration Platform Cloud reached End of Life on December 12, 2024, after which no new instances may be provisioned and no support is provided.
Oracle's contractual EOL definition is broader than most buyers assume: no enhancements, no patches, no standard Cloud Support, and no new deployments permitted. That last clause is the one that bites, because it converts a running estate into a frozen one on a date Oracle chose unilaterally.
The financial mechanism is straightforward. If you have signed a three or five year term commitment against an API Platform or gateway SKU and Oracle announces consolidation into Oracle Integration midway through, you are not credited into the successor at your negotiated unit economics.
You are re-quoted at then-current list, on a different metric (message packs rather than gateway nodes or calls), with a migration deadline that removes your ability to walk. In 25 years of negotiating with this vendor, the migration quote has never arrived cheaper than the product it replaced.
Our reading of how Oracle Integration meters message packs is that gateway traffic reclassified as messages is where the uplift hides.
The clause you need is migration price protection, and Oracle will not offer it unprompted.
Write into the ordering document that if Oracle discontinues, renames, or folds the ordered program into a successor during the committed term, you receive equivalent functional capacity in the successor at the same effective unit price.
For the remainder of the term plus one renewal cycle, with your existing discount percentage preserved against the successor's list.
Absent that language, an EOL notice is a forced repurchase event dressed as a product roadmap update. Price it now, before you sign, because a term commitment with no migration protection is an option Oracle holds and you paid for.
Evidence base and the patterns we see repeatedly
The sourcing here is deliberately mixed, because Oracle's own documentation does not resolve the question.
The legacy Fusion Middleware section of the Technology Global Price List carried API Gateway, API Manager, and API Catalog as separate Named User Plus and Processor line items, sitting alongside WebLogic and SOA products, which is why so many buyers hold paper on metrics Oracle no longer markets.
Non-USD lists diverge materially: the Japan price list dated July 15, 2026 shows Service Bus at 3,565,000 yen per processor and SOA Suite for Oracle Middleware at 8,912,500 yen per processor, with annual support at 784,300 yen and 1,960,750 yen respectively.
And the per-deployment fee claim on the OCI side is genuinely contested across sources dated 2020 through 2026, which is a reason to verify against your regional price list rather than a reason to assume it does not exist.
Request volume pricing with a monthly free tier and no documented per-gateway instance charge.
Every self-hosted gateway node is a WebLogic domain, and the prerequisite base license is not optional.
| Recurring finding | Where it surfaces | Buyer exposure |
|---|---|---|
| Undocumented standby nodes | Passive DR and blue/green gateway pairs | "Installed and/or running" counts them at full Processor rate |
| Gateway pods on shared Kubernetes | Nodes shared with unrelated workloads | Verrazzano precedent licenses all processors in the cluster |
| Missing WebLogic Suite prerequisite | Gateway installed without the base license on paper | Retrospective purchase at $45,000 per processor plus back support |
| NUP bought where Processor is required | Externally consumed or multi-tenant APIs | Users are uncountable, so Oracle reclassifies to Processor |
| Matching Service Levels | Any attempt to drop a middleware line | Support repricing cancels most of the intended saving |
The pattern behind all five is the same: buyers reason about gateways as network appliances and Oracle counts them as WebLogic estates.
Standby nodes and Kubernetes co-tenancy are the two that most often appear for the first time in an audit script rather than in an architecture review, and both are fixable in advance.
The Processor versus NUP decision on clustered middleware follows identical logic and is worth reconciling in the same exercise.
- Percentile standing for your exact deal size and industry, from real closed transactions
- Scenario simulation before the call: test alternative terms and see the financial impact of each
- A negotiation playbook, talking points, and a two page executive brief on day one
Your first five moves
- Get the per-deployment question answered in writing (week 1, sourcing lead). Third-party guides disagree on whether OCI charges a monthly per-deployment fee on top of roughly $3.00 per million calls, so demand the live regional price list extract and an email confirmation from your Oracle rep, because that single line decides whether your OCI scale-out is free or metered.
- Build a node inventory that separates active, standby, DR, and container images (week 2, platform architecture). Each on-premise gateway node is a WebLogic domain with an admin server and a managed server, and Oracle's Processor definition counts every processor where programs are "installed and/or running," so a spreadsheet listing hostname, core count, core factor, traffic status, and image source is your only defense against a five-times multiplier.
- Run the crossover model before you approve the next node (week 3, FinOps with licensing). Compute total cores times core factor times $45,000 for WebLogic Suite, then compare that against twelve months of actual call volume priced at the cloud rate; in our experience the on-premise footprint loses this comparison badly at anything under a few hundred million calls per month, and the gateway's user population is uncountable, so Named User Plus is not an escape route.
- Write the self-hosted node ceiling into the ordering document (at contract signature, legal and sourcing). Name the topology, the maximum node count, and the treatment of passive standby and DR nodes; without that language Oracle prices the topology it finds, not the one you designed.
- Model Matching Service Levels and repricing before you touch any support line (before any reduction, licensing lead). Support is the permanent number, and terminating a gateway line can reprice the surviving SOA and integration entitlements around it.
The sequence matters more than any single move.
Steps one and two are evidence gathering, and they are cheap.
Steps three through five are commitments that Oracle will hold you to for the life of the agreement. Buyers who reverse this order sign a topology cap before they know their real node count, then discover the container images pulled into a Kubernetes cluster were never in the inventory.
Finish the inventory first, then negotiate.
Frequently asked questions
Does Oracle charge per API Gateway instance in OCI?
The consistent published position, dating from 2020 through 2026 third-party guides, is that OCI API Gateway bills only on request volume at roughly $3.00 per million calls per month after a 1 million call free tier, with no charge for gateway instances, deployments, or idle time.
One 2025 guide contradicts this and asserts a small monthly per-deployment fee.
Because that single claim decides whether a deployment-per-microservice pattern is free or metered, get the answer confirmed in writing against the live Oracle Cloud Price List for your specific region before you finalize the design.
How is an on-premise or hybrid Oracle API gateway node licensed?
By Processor. Each gateway node installs as a WebLogic domain with a single admin server and a single managed server, and the Processor definition covers all processors where the programs are installed and/or running.
The count is total cores multiplied by the applicable core processor licensing factor, so a five-node tier at four cores per node with a 0.5 factor is ten processors.
Do I have to license WebLogic Suite underneath the gateway?
If the gateway node runs as a WebLogic domain, yes, WebLogic Suite is a prerequisite base license rather than an optional add-on, following the same pattern Oracle applies to SOA Suite-class options.
At $45,000 per processor on the April 16, 2026 Technology Global Price List, that prerequisite frequently exceeds the cost of the gateway line item itself. Confirm which base is required for your exact release and record the answer in the ordering document.
Can I use Named User Plus instead of Processor for API gateways?
In practice, no. Processor is the default metric for any deployment where users cannot be counted, which includes web-facing applications, multi-tenant systems, and integration platforms, and gateway-fronted APIs almost always serve external or machine consumers.
The crossover math confirms it: with NUP typically priced around one fiftieth of the Processor rate, Processor wins past roughly 50 real users per processor, and minimum user floors apply regardless.
Are idle standby gateway nodes licensable?
Yes, under the standard reading. The Processor definition captures programs that are installed and/or running, so a passive standby node with the software installed is in scope even if it never receives traffic.
Blue-green pairs, DR replicas, and pre-baked container images carry the same risk, and Oracle's Verrazzano terms show the pattern extended to all processors in Kubernetes cluster nodes where images are pulled.
What happens if Oracle retires the API management SKU I bought?
Oracle's contractual EOL definition means no enhancements, no patches, no standard Cloud Support for that version, and no new deployments permitted, with all workloads required to migrate to a GA release before the EOL date.
Recent precedent includes OCI Process Automation unavailable and unsupported after April 3, 2026 with mandatory migration to Oracle Integration, and Data Integration Platform Cloud at End of Life on December 12, 2024.
Negotiate a migration and price-protection clause into any multi-year term so a retirement does not become a forced repurchase at then-current rates.
How do I compare the cloud meter against on-premise licensing?
Take 12 months of actual API call volume, price it at roughly $3.00 per million after the free tier, and set that against the Processor count your self-hosted topology produces at total cores times core factor multiplied by the applicable list price plus 22 percent annual support.
In most mid-volume estates the call-based number is dramatically smaller, which is exactly why Oracle prices it that way. Run the comparison before you approve the next gateway node, not after.
