OCI Database Service Pricing: Base, ADB, and Exadata Compared

When organizations model OCI costs, the database line usually carries half or more of the total, and it is the line where placement decisions matter most. OCI offers three main homes for an Oracle database: Base Database Service on virtual machines, Autonomous Database in serverless or dedicated form, and Exadata Database Service on dedicated engineered infrastructure. All three run real Oracle Database, all three bill by consumption, and the difference between putting a workload on the right tier and the wrong tier is frequently larger than every other optimization on the bill combined. This article explains how each tier prices, how the license included and bring your own license models change the math, and how to decide where each database belongs.

This article is part of our complete guide to OCI pricing and TCO. The compute side of the bill is covered in OCI compute pricing and the storage underneath it in OCI storage pricing.

The three tiers and what they cost

The honest way to compare the tiers is by their billing unit and their floor. Base Database Service bills like compute: you choose a VM shape, pay per OCPU hour, and pay for the storage you provision. Autonomous Database bills by the ECPU, an abstracted unit of compute, with the serverless variant able to scale to zero infrastructure commitment and the dedicated variant adding an Exadata infrastructure floor. Exadata Database Service has a genuine infrastructure floor, the dedicated engineered system, plus a per core charge for the database compute you enable on it.

At list, an Autonomous Database ECPU runs around $0.336 per hour with the license included, and Base Database Service OCPUs price in a similar band depending on edition. The numbers move with Oracle's price list, so treat them as orientation rather than quotation, but the relationships hold: serverless autonomous is the cheapest way to run a small or intermittent Oracle database, Base is the predictable workhorse, and Exadata trades a high floor for the lowest cost per unit of serious work, as we showed in the cost of Exadata Cloud Service.

License included vs BYOL: the multiplier underneath everything

Every tier offers two license postures. License included bundles the Oracle Database license into the hourly rate, which is simple and carries no compliance exposure. Bring your own license, BYOL, applies licenses you already own and cuts the hourly rate dramatically, commonly to around a quarter of the license included figure for Enterprise Edition workloads. On a database estate of any size, that multiplier dwarfs every infrastructure optimization available.

The catch is that BYOL is a licensing position, not just a checkbox. The licenses must actually cover the OCPUs or ECPUs consumed, option usage must match what is owned, and the on premises estate the licenses came from must genuinely release them. Getting this wrong converts a saving into an audit finding. This is exactly the boundary between our work and licensing specialism: we size and place the workloads, and for the license entitlement and negotiation analysis, independent Oracle licensing advice should be secured before any BYOL commitment is signed.

Base Database Service: the predictable middle

Base Database Service is the closest thing to running Oracle on a well managed VM. You pick a shape, Enterprise or Standard Edition, license included or BYOL, and pay per OCPU hour plus block storage. Its economics reward right sizing: OCPUs scale online, so a database that needs four cores during the day and one overnight can be scripted to follow that curve. Standard Edition on Base is one of the quiet bargains on the platform for smaller databases, because the license included rate is modest and the edition limits rarely bite at that scale. The common cost mistake is habit sizing, carrying the on premises core count into the cloud without measuring, which is why every migration we run starts from observed AWR utilization rather than the existing server spec.

Autonomous Database: paying by the ECPU

Autonomous Database serverless bills per ECPU hour plus database storage, and two behaviors make it cheap or expensive. Auto scaling lets the database breathe with demand, billing actual consumption rather than peak provision, which suits spiky and analytical workloads well. And the ability to stop a database entirely, or schedule it to run business hours only, means dev and test estates can shed most of their cost outside working hours. The expensive pattern is treating ADB like a fixed appliance: provisioning for peak, never enabling auto scaling, and leaving every environment running around the clock. The platform will happily bill for all of it. Storage on ADB bills separately per terabyte, and the same tiering honesty applies as everywhere else.

Autonomous dedicated adds an Exadata infrastructure rental underneath, which only makes sense when isolation, custom maintenance windows, or regulatory posture demand it. The infrastructure floor needs enough database consolidation on top of it to earn its keep, the same consolidation logic as Exadata itself.

Exadata Database Service: a floor that has to earn its keep

Exadata Database Service prices in two layers: the infrastructure, rented by the month from a quarter rack upward, and the database server cores you enable, billed per core with the BYOL discount available. The floor is real five figure money every month, and the equation only works when the platform carries enough work, either one genuinely demanding system or a consolidation of many databases that would otherwise each carry their own infrastructure. When it works, the cost per transaction is the lowest on the platform, because the engineered system does more work per core and the licensing tied to enabled cores follows the same curve. The full breakdown, including the enabled core discipline that controls the bill, is in our Exadata cost article.

The lines around the database

Database services carry their own storage and backup meters, and they deserve a line in the model rather than a shrug. Base Database Service bills block storage at the standard rates with performance settings that should match the workload, ADB bills database storage per terabyte per month with backup storage on top, and Exadata includes its storage grid in the infrastructure fee but bills backups to object storage separately. Backup retention policy is the quiet driver: a 60 day retention on a busy database can hold more terabytes of backup than the database itself, billed every month, and trimming retention to the genuine recovery requirement is free money in most estates. Data Guard standbys are the other commonly unmodeled line. A standby is a second database billing full infrastructure rates around the clock, and while tier one systems justify it without argument, estates that reflexively standby every database are paying production rates for protection that a backup based recovery would deliver at a tenth of the cost for the systems that can tolerate hours of recovery time.

A worked example: one database, three price tags

Consider a single 8 core Oracle Enterprise Edition database with a few terabytes of storage, a steady daytime load, and quiet nights. On Base Database Service with license included it lands in the mid four figures per month. The same database on Base with clean BYOL entitlement drops to the low four figures. Moved to Autonomous serverless with auto scaling and a business hours schedule, the effective consumption falls again, often to half the always on figure, because the nights and weekends stop billing. And consolidated onto an existing Exadata platform with spare enabled cores, its marginal cost approaches the storage it occupies. Four legitimate homes, a fourfold spread in cost, and the only differences are placement and posture. Multiply that spread across an estate of fifty databases and the placement review stops looking optional.

A placement framework

This is the sequence we use in assessments to put each database on the cheapest tier that actually fits it.

What a wrong placement costs

The failure modes are symmetrical. A small departmental database on dedicated Exadata pays an engineered system premium for work a Base VM would do at a tenth of the cost. A tier one system scattered across oversized Base VMs pays more in aggregate than a consolidated Exadata floor while delivering worse performance. And any tier running license included on workloads with clean BYOL entitlement pays roughly four times the necessary rate. In our optimization engagements, where the fee is a percentage of verified savings, database placement and license posture corrections routinely contribute the largest share of the 40 percent average reduction we see across engagements, ahead of compute right sizing and storage tiering.

Bringing it together

OCI database pricing rewards measurement and deliberate placement. The three tiers are honest about their economics: no floor and modest rates at the bottom, a real floor and the best unit economics at the top, and an abstracted consumption model in the middle that is as cheap or as expensive as your scaling discipline. The license posture multiplies whichever choice you make. If your database estate moved to OCI by lift and shift, or grew by default rather than design, a placement review is usually the single highest value exercise available, and it is the first thing we model in an OCI assessment, on a fixed project fee, a managed monthly retainer, or an optimization fee paid only on verified savings.