Oracle's 2:1 vCPU ratio is identical on AWS, Azure, and Google, so the cheapest cloud is decided by Support Rewards at $0.25 per $1 of OCI spend, not by the license count

For a fixed Oracle Enterprise Edition workload, the license quantity you owe is the same on all three hyperscalers: 2 vCPUs equal 1 Processor license with hyperthreading on, and the 0.5 core factor does not apply. That means the price gap between platforms comes almost entirely from compute list rates, regional uniformity, and the Support Rewards accrual of $0.25 per dollar (or $0.33 for ULA customers), which now also accrues on Oracle Database@AWS. Decide platform on where your application tier sits and on how much of your support bill you can retire, not on a license count that does not move.

Prepared by Redress Compliance · August 18, 2026 · Oracle advisory. Multicloud platform selection and BYOL engagements, 2024 to 2026.

Start with the arithmetic, because it is the part vendors hope you get wrong.

On AWS, Azure, and Google Cloud, all three now sit on Oracle's Authorized Cloud Environment list, and the counting rule is identical: 2 vCPUs equal 1 Processor license when hyperthreading is on, 1 vCPU equals 1 Processor when it is off.

The Processor Core Factor Table does not apply in any Authorized Cloud Environment, so the 0.5 x86 multiplier you rely on for on-premises math disappears.

That single change is where migration budgets rupture: 32 physical x86 cores on-premises need 16 Processor licenses, while a 32 vCPU cloud shape also needs 16, but a like-for-like core rebuild at 64 vCPUs needs 32.

OCI counts on a different metric entirely, OCPUs, where 1 OCPU is one physical core with two threads that Oracle itself treats as 2 vCPUs, so a 32 core equivalent lands at 32 OCPUs. Two divergences deserve contract language rather than assumption.

First, Standard Edition 2: one published reading caps SE2 at 8 vCPUs on AWS (2 vCPUs to a core) but 4 vCPUs on Azure and Google (1 vCPU to a core), while a second reading gives 8 vCPUs on all three, with 4 vCPUs counting as one socket.

Pin the version of the policy PDF in force at your contract date. Second, SE2 carries a Named User Plus floor of 10 NUP per 8 vCPUs regardless of how few humans log in.

Third, and most expensive, the OCI BYOL conversion is published inconsistently: one source has 1 Processor covering 8 ECPUs or 2 OCPUs, while Exadata Database Service guidance has 2 Processor licenses consumed per OCPU, a 4x swing on the same estate.

Oracle's own source of record is the PaaS and IaaS Universal Credits Service Descriptions, and the conversion rate belongs in your ordering document, not in a slide.

In practice, the negotiating leverage is the ambiguity itself. Where two Oracle-sourced readings exist, you are entitled to ask for the favorable one in writing, and in our experience Oracle will concede a defined conversion in an ordering document far more readily than it will concede a discount.

Oracle has spent two product cycles removing price as a differentiator at the database layer, and it has largely succeeded. Oracle AI Database@AWS is transacted through the AWS Marketplace but priced identically to Exadata Database Service on OCI.

Database@Azure per-ECPU-hour rates land within 3% to 6% of equivalent ExaCS rates. Oracle states outright feature and list price parity for Oracle AI Database@Google Cloud.

That parity is deliberate: it pushes the platform decision onto where your application tier already runs, which is the one thing Oracle cannot monetize directly.

The anchors that do vary are the ones on OCI's own rate card, and they vary by BYOL status rather than by cloud. OCI compute lists at $0.025 per OCPU-hour. Autonomous Database Serverless runs $4.03 per OCPU-hour license-included against $1.34 with BYOL.

Exadata Cloud Service X10M runs $3.10 against $0.81 with BYOL. Autonomous on the ECPU metric circulates at roughly $0.2616 per ECPU-hour at list and about $0.336 license-included.

Published estimates of the BYOL saving span 40% to 80% depending on tier, so model your own crossover rather than quoting a headline; our BYOL versus License Included crossover analysis exists precisely because the answer is workload-specific.

The real price lever is the Universal Credits commitment. Negotiated discount runs roughly 10% at a $250K annual commit and past 35% at $10M or more.

That curve, plus regional uniformity on OCI against Frankfurt premiums of 20% to 29% and Sao Paulo premiums near 60% on the hyperscalers (vendor-sourced figures), is worth multiples of any per-hour rate difference you will find comparing marketplace listings side by side.

Look at what Oracle actually did to the comparison. The Enterprise Edition conversion is 2 vCPUs to 1 Processor license with hyperthreading on, 1 to 1 with it off, on Amazon EC2, Amazon RDS, Microsoft Azure, and Google Cloud alike, with the 0.5 core factor explicitly withdrawn.

Then it did the same thing to the rate cards: Oracle AI Database@AWS is sold through the AWS Marketplace at the same price as Exadata Database Service on OCI, Database@Azure lands within 3 to 6 percent of the equivalent ExaCS rate.

And Oracle publishes feature and list price parity for Database@Google Cloud.

Two of the three variables a buyer would normally use to rank platforms have been flattened on purpose. What remains is not a technical comparison at all.

It is a commercial one, and every variable left standing is one Oracle sets unilaterally: Universal Credits discount tiers, Support Rewards accrual rates, and the BYOL eligibility language in the Service Descriptions PDF that the ordering document merely points at.

The 2026 extension of Support Rewards to Database@AWS should be read as a concession on venue, not on money. For a decade Oracle's position was that the only cheap place to run Oracle was Oracle's own infrastructure, and the pricing policy enforced it. Rewards on Database@AWS ends that fight.

Oracle has decided it would rather host your database inside somebody else's data center, on a Universal Credits commit it controls, than lose the workload to PostgreSQL or to a competitor's managed engine. The annuity survives; the venue is negotiable.

That is a meaningful shift in leverage, and buyers who still frame the conversation as "we might have to move to OCI" are arguing against a position Oracle abandoned.

The trap sits in the arithmetic of the reward itself. At $0.25 per dollar of eligible OCI spend, retiring a $1M technology support bill requires roughly $4M of annual OCI consumption. At the ULA rate of $0.33 it still takes about $3M.

For most estates that number is three to four times larger than the cloud footprint the workload actually needs, which means the program does not reward migration, it rewards over-commitment.

Add the conditions: Universal Credits only (pay-as-you-go is excluded), the program has to be written into the cloud contract rather than assumed, credits accrue monthly and expire 12 months after issuance, they cannot be applied to tax.

They cannot be applied retroactively to past-due support invoices, and SaaS spend does not count.

In our negotiation experience the most expensive condition is the one nobody reads: an existing customer sitting on an above-standard volume discount can be reset to standard tier when a new order is placed to enable the program.

A 30 percent discount dropping to 20 percent on a $4M commit costs $400,000 a year, which is 40 percent of the reward you just signed up to earn.

This is why chasing the cheapest hourly rate is a losing strategy here. The hourly rate is the number Oracle has already equalized across venues and the number it is least willing to move on, because moving it would unwind the parity architecture.

The commit shape, by contrast, is where Oracle's own quota pressure lives, and it is where published discounting runs from roughly 10 percent at a $250K annual commit to 35 percent or more above $10M.

A buyer who negotiates hard on rate and accepts a commit sized to their support bill has traded the least valuable concession for the most expensive one.

There is a second-order effect worth naming. A commit sized to your support bill locks in the support bill.

If the reward is only harvested by spending 4x, then any program that reduces your support base (terminating unused Processor licenses, consolidating to fewer editions, moving a workload to PostgreSQL where the economics support it) also reduces the reward, which makes the commit look oversized.

Oracle has built a mechanism where support reduction and cloud commitment are financially antagonistic. That antagonism is the product.

The buyer-side conclusion follows directly. Treat the hourly rate as the least negotiable number on the page and stop spending negotiation capital there.

Spend it instead on the two things that are contractual rather than published: the conversion ratio written into your ordering document, particularly on OCI where OCPU and ECPU BYOL ratios vary by service, and the shape of the commit, meaning term length, ramp, rollover of unused credits.

And an explicit written protection that entering Support Rewards does not reset your existing discount tier.

Those clauses survive every rate card revision Oracle publishes.

Run the numbers before you let this program reorder your platform shortlist. Oracle's own published example is unambiguous: a $1M technology support bill is fully retired by $4M of total OCI spend at $0.25 per dollar. ULA customers accrue at $0.33, which brings the break-even to roughly $3M.

The program therefore only changes the ranking for estates where planned cloud spend already sits at three to four times annual support.

Below that, it is a discount on part of your support bill, not a reason to pick a platform, and it should be modeled as such alongside the rulebook differences across the three venues.

Then price the conditions, because each one shortens the effective value. Universal Credits is mandatory and pay-as-you-go earns nothing. The program has to be named in the cloud contract, so verify the clause rather than assuming your sales contact enabled it.

Credits accrue monthly and expire 12 months after issuance, so a support renewal that falls just outside the window strands them. They cannot pay tax, which on a European or Brazilian support invoice removes 17 to 25 percent of the invoice from coverage.

They cannot be applied retroactively to past-due invoices. SaaS spend is excluded entirely. And in our experience the reset risk is real: confirm in writing that placing the new order preserves your existing above-standard volume discount rather than dropping you to standard tier.

The license ratio is fixed at 2:1 everywhere, so the only compute variable left is where you run, and that variable is far larger than most platform business cases model.

Oracle's own pricing page (vendor-sourced, treat as directional and re-verify against current rate cards) puts a 4 vCPU AMD instance in Frankfurt at 20% above US East on AWS, 21% above on Azure, and 29% above on Google Cloud.

Sao Paulo widens the spread to 59% on AWS, 60% on Azure, and 59% on Google. OCI charges the same rate for a service regardless of region. For a single-region US estate that uniformity is worth nothing.

For a bank running Frankfurt plus Sao Paulo plus Singapore under data residency rules, it is the dominant term in the model, and it compounds every year of the contract while the license count sits still.

The second cost that outlives the decision is exit. A platform selected on a 5% hourly advantage is routinely locked in by a migration bill that exceeds three years of that advantage, and in our engagements the two largest line items are almost never compute.

They are egress on the initial and reverse data movement, and the dual-running period where the source and target both hold licensable cores.

Read the egress cost trap on cross-cloud Oracle migrations before you sign, and model the overlap window explicitly using the dual-running cutover guidance, because Oracle does not grant a free parallel-run entitlement on ACE platforms.

Price the exit at contract signature, not at the point you want to leave. A 90-day overlap on a 200-Processor estate is a real second license bill, and it is the number that turns a rate-card win into a three-year loss.

Four patterns recur in engagements often enough that we now test for them by default. First, teams size by thread count and double the license need, a mistake the Authorized Cloud Environment core-counting rules make avoidable in an afternoon.

Second, BYOL is claimed on services the underlying license was never eligible for, which surfaces only at audit.

Third, Support Rewards is assumed rather than contracted: it requires a Universal Credits plan, explicit inclusion in the cloud agreement, and credits expire 12 months after issuance and cannot offset tax.

Fourth, discount tiers reset on new orders, so the 35% you won at $10M commit does not automatically carry to the next order form. Pin all four in writing before signature.

Related advisory: the full topic guide.

Is Oracle Database genuinely cheaper to license on OCI than on AWS or Azure?

Not on license quantity. The Authorized Cloud Environment policy applies the same 2 vCPU to 1 Processor ratio on AWS, Azure, and Google Cloud, so the same Enterprise Edition workload needs the same number of licenses on all three.

OCI differs because it counts OCPUs rather than vCPUs and because it accrues Support Rewards at $0.25 per dollar of spend, and those two mechanics, not the license count, are what produce any cost advantage.

Does the Processor Core Factor Table apply in the cloud?

No. Oracle's cloud licensing policy states the Processor Core Factor Table is not applicable in Authorized Cloud Environments, and you cannot apply the 0.5 x86 multiplier on top of the vCPU rule.

This is the single most common budget error we see: 32 x86 cores on-premise need 16 Processor licenses, but a like-for-like cloud instance sized at 64 vCPUs to match thread count needs 32.

How many vCPUs can Standard Edition 2 run on in the cloud?

Published readings conflict. One reading caps SE2 at 8 vCPUs on AWS (2 vCPUs per core) and 4 vCPUs on Azure and Google Cloud (1 vCPU per core), while another reads 4 vCPUs as one socket with an 8 vCPU maximum on all three.

Verify against the version of Oracle's cloud licensing policy PDF in force on your contract date, and remember the separate floor of 10 Named User Plus per 8 vCPUs applies regardless of user count.

Can I earn Oracle Support Rewards on AWS?

Yes, for Oracle Database@AWS specifically. Oracle and AWS confirmed in July 2025 that Database@AWS consumption qualifies for AWS commitments and for Oracle Support Rewards, which removes the historical assumption that rewards only accrue on OCI.

General EC2 or RDS spend does not qualify, so scope the eligible services precisely in the ordering document.

How much OCI spend do I need to eliminate my Oracle support bill?

Roughly four times your annual technology support bill at the standard rate of $0.25 per dollar of OCI spend, so a $1M support bill requires about $4M of qualifying OCI spend. ULA customers accrue $0.33 per dollar, bringing the requirement to roughly three times.

Credits expire 12 months after issuance, cannot pay tax, and cannot be applied to past-due invoices, so the offset only works if consumption and renewal timing line up.

How much does BYOL actually save versus License Included?

Published figures range from about 40% to roughly 76% depending on the service tier. Autonomous Database Serverless lists at $4.03 per OCPU-hour without BYOL and $1.34 with it, and Exadata Cloud Service X10M lists at $3.10 versus $0.81, both near the top of that range.

Lower-tier services show smaller gaps, so model your specific service rather than applying a blanket percentage.

Do regional price differences matter more than the license ratio?

For multi-region estates, often yes. Oracle publishes that a 4 vCPU AMD instance in Frankfurt costs 20% more on AWS, 21% more on Azure, and 29% more on Google than in US East, and that Sao Paulo runs 59% to 60% higher on all three, while OCI prices uniformly by region.

Those figures are vendor-sourced and should be treated as directional, but the dispersion is large enough to swamp any hourly rate difference you negotiate.