A funded OpenJDK migration plan with named replacement builds and a dated 9 to 14 month schedule moves an Oracle Java number; an unfunded threat to leave moves nothing
GLAS prices your alternative, not your annoyance. When a 5,000-employee company facing $630,000 a year at list can show board-approved funding, a named vendor build, and a schedule that lands before the October 2026 JDK 21 patch cliff, the claim becomes a negotiation about transition length rather than a subscription for every employee. Without those three artifacts, Oracle correctly reads the exit talk as posturing and holds list.
Prepared by Redress Compliance · August 28, 2026 · Oracle Java advisory. GLAS audit and renewal engagements, 2024 to 2026.
Executive summary
The Employee metric is what makes an exit rational, and saying so out loud is your strongest single argument: a 5,000-employee company running Oracle Java on 40 servers pays $630,000 a year at list, or $15,750 per server.
No engineering leader defends $15,750 per server for a runtime with free binary-compatible substitutes, and GLAS knows the arithmetic is indefensible on the merits, which is precisely why it prefers to keep the conversation on download logs.
Credibility has three components and GLAS tests all three: a named replacement build, a dated schedule, and money already in a cost center.
Migration plans that name Amazon Corretto, Azul, BellSoft, Red Hat, Eclipse Temurin or Microsoft Build of OpenJDK, and attach an owner and a date per application tranche, get discounted differently from plans that say "we are evaluating alternatives."
The October 2026 Critical Patch Update is a two-sided deadline, and buyers who treat it as only Oracle's leverage give away the half that belongs to them.
Oracle JDK 21 updates move from the NFTC to the OTN license from that CPU, which raises your urgency; but it also means Oracle has a closing sales window, and a buyer whose migration lands in the same quarter has nothing left to buy.
Evidence, not intent, is what converts a plan into a lower number: substitution receipts, de-installation tickets, and a Java Usage Inventory that separates Oracle JDK from OpenJDK and embedded runtimes.
In practice this splits the claimed population, and since Oracle audits typically run three to nine months from formal letter to settlement, every tranche you finish inside that window is a tranche you never pay for.
How Oracle prices your alternative: the mechanics of a credible exit
GLAS is not scoring your irritation. It is running a private comparison between the subscription revenue at stake and its own estimate of two variables: the probability you actually leave, and how fast. Every artifact you put on the table moves one of those two numbers, and nothing else does.
Start with the arithmetic that defines the prize. The Employee metric runs seven published bands from $15.00 per employee per month at 1 to 999 down to $5.25 at 40,000 to 49,999, with no published rate above 50,000.
A 5,000-employee company running Oracle Java on 40 servers pays $630,000 a year at list, roughly $15,750 per server, because the price is decoupled from the footprint.
That decoupling is exactly why an exit is cheap for you and expensive for Oracle: the revenue Oracle loses when you migrate 40 servers is the same $630,000 it would lose if you had 400.
Note the band boundary too, because it tells you how mechanical this pricing is: 9,999 employees prices at $1,259,874 a year while 10,000 prices at $990,000. One more employee removes $269,874.
Also read the buried ceiling in the price list: the Employee metric permits installation on up to 50,000 processors (excluding desktops and laptops) before an additional license is required.
Where you sit against those numbers determines what tier of evidence you need to bring, and the four tiers below are what we see convert at the table.
| Credibility tier | What you are putting in front of GLAS | What it realistically achieves |
|---|---|---|
| 1. Verbal intent | "We are evaluating alternatives" on a call | Nothing. Oracle holds list and books the follow-up |
| 2. Technical assessment | Inventory plus a vendor shortlist, no dates, no money | Modest movement, typically discount inside the same band and term |
| 3. Funded plan, named builds, dated tranches | Board-approved cost center, named vendor build, 9 to 14 month schedule | The real shift: talks move from a perpetual subscription to a priced transition window |
| 4. In-flight migration, tranches completed | Change tickets and logs proving completed cutovers | Scope collapses to the residual estate; Oracle negotiates to keep a foothold |
This is why the trading floor and the WebLogic estate matter out of proportion to their size. Those are the workstreams buyers leave undated because they are hard, and they are precisely the ones an experienced GLAS rep asks about first.
Date them, staff them, and if the answer is "these 12 apps stay on Oracle for 18 months," say so and price that residual explicitly. A named exception with an owner is a stronger artifact than a silent one, because it proves the rest of the plan was built by someone who counted.
The four artifacts GLAS actually reads
Four deliverables carry weight, and the rest is narrative. First, a Java Usage Inventory that lists every installation with version, vendor, host, business purpose, and a named owner.
This does double duty: it is the credibility floor for any migration claim, and it is your only defensible counter to Oracle's download-log arithmetic and inflated audit-script output.
Oracle cross-references its download records against its subscription database and treats an IP range in the JDK logs as hard evidence. An inventory that shows the host, the build vendor, and the de-installation ticket answers that; a spreadsheet of employee counts does not.
Second, a named-vendor decision, meaning a specific build and a specific support arrangement: Amazon Corretto, Azul, BellSoft Liberica, Eclipse Temurin, Red Hat, or Microsoft. "OpenJDK" as a category is not a decision and GLAS will read it as one.
Third, a tranche schedule with calendar dates and named owners across 9 to 14 months, which is the honest range for a mid-size estate, as set out in how long an OpenJDK migration actually takes.
Fourth, a funding line already sitting in an approved cost center, with the approval date and the approver.
Rank those four by how hard they are for Oracle to re-argue and the fourth wins outright. Your inventory can be challenged on completeness. Your vendor choice can be challenged on support depth and indemnity.
Your schedule can be challenged on optimism, and a good GLAS rep will do exactly that, application by application. But a board-approved budget line with a date and an approver's name is a fact about your company that Oracle has no standing to dispute.
It is the artifact that converts a stated intention into a sunk commitment, and it is the reason the CFO business case should be built before the first substantive GLAS call, not after the price arrives.
Oracle Java SE Renewal Exit Strategy. The buyer side brief.
Oracle Java SE renewal exit strategy buyer side brief. Independent buyer side advisory from Redress Compliance.
Get the white paper →Why the threat only works when Oracle believes you would rather pay someone else
Strip the emotion out of this and the exit conversation is not a threat at all. It is a valuation exercise Oracle runs on your behalf whether you participate or not. GLAS is modeling one number: the price at which you stop being a subscriber.
If your alternative is credible, that number becomes the ceiling on what Oracle can charge, and the negotiation collapses into a discussion about transition length and bridge coverage.
If your alternative is vague, Oracle's model returns "no viable substitute" and list price is not a starting position, it is the answer. Buyers who arrive expressing displeasure at $630,000 a year for 40 servers have communicated nothing Oracle can price.
Buyers who arrive with a named vendor build, funded engineering hours, and a cutover date have communicated a substitute cost, and the vendor is now competing against it.
The Employee metric is unusually vulnerable to this because Oracle deliberately decoupled price from footprint.
Oracle's own published example shows a 28,000 count (23,000 employees plus 5,000 contractors and consultants) at $6.75 per month producing $2,268,000 a year, and the price list is explicit that quantity is determined by the number of Employees, not the number who use the Programs.
The consequence is structural and works against Oracle at the table. Every acquisition, every seasonal hiring wave, every outsourced service desk you onboard raises the subscription bill.
The migration cost does not move, because the migration is scoped to your actual JVM estate, which for most enterprises is a few dozen to a few hundred runtimes.
When the recurring cost scales with headcount and the escape cost scales with servers, the substitute gets cheaper every year you delay, and both sides know it.
This is why the download log, which is GLAS's strongest opening card, is a weak closing card. Oracle cross-references its download records against the subscription database and treats an IP range appearing in JDK download logs without a matching subscription as hard evidence.
That works beautifully against a buyer who can only argue. It works poorly against a buyer who produces receipts from an alternative build source and archived change-management tickets showing a company-wide uninstall date. A download is evidence of an event in the past.
A substitution receipt is evidence of the present state, and it also happens to be evidence of the future, which is what the pricing model actually depends on. Argument invites more questions.
Documentation ends them, which is the recurring theme in the buyer-side approach to negotiating a GLAS Java claim.
The October 2026 patch cliff is widely described as your problem. Read it again.
Oracle has stated that JDK 21 updates through and including September 2026 are available under the NFTC and that from the October 2026 Critical Patch Update, further updates move to the OTN license, the same license used for Java 8, 11 and 17.
Oracle's own advice at that cliff is to consider the Universal Subscription, with commercial support for Java 21 running to at least September 2031. That is a sales deadline as much as a compliance deadline.
GLAS has a quarter in which a specific, dated event is supposed to convert reluctant prospects. If your migration plan lands before that date, the event converts nothing from you, and the account team's forecast has a hole in it.
The deadline is symmetric, and only one party usually behaves as though it is.
Disciplined buyers also hold a second lane. Oracle has stated that JDK 25 updates are planned to remain under the NFTC until October 2028, one year after the next planned LTS.
That gives a technically competent organization a legitimate route: move to JDK 25 on free-license terms, buy roughly two years of runway, and complete the vendor OpenJDK transition on your own schedule rather than Oracle's.
Whether that fits your estate is an engineering question addressed in the 2026 migration decision gate. What matters at the table is that two independent exit routes are harder to price against than one, and a buyer who can describe both is visibly not bluffing.
The most common failure in our practice is not weak evidence, it is bad sequencing. Buyers commission the migration analysis, approve the funding, select the vendor, and then tell Oracle they are still evaluating options.
Every artifact that could have compressed the number is sitting in a folder nobody at Oracle has seen, and the account team is quite reasonably pricing a customer with no alternative. Confidentiality about your internals is sound practice.
Confidentiality about your alternative is unilateral disarmament.
Finally, weigh the asymmetry honestly, because it should govern how hard you are willing to push. If Oracle loses your subscription, it books the revenue miss in one quarter and moves on to the next of the hundreds of organizations GLAS contacts per quarter.
If you sign a five-year Employee-metric commit at a bad rate, you carry it through every headcount change and every reorganization until 2031, with no mechanism to recover it. Oracle's downside is temporary. Yours is contractual.
What to say at the table, and what not to
Language discipline is where good preparation gets thrown away.
Introduce the plan as a decision already taken, not as a warning: "Our target-state Java platform is Vendor X's build, funded and scheduled for completion in Q3.
And I can share the milestone dates." Say nothing about population in the same breath, because any employee number you volunteer becomes the baseline for every subsequent proposal.
When the download claim arrives, do not argue it. Produce the alternative vendor's download or repository logs alongside archived change-management tickets carrying the uninstall date, and let the documents answer.
On the 45-day questionnaire covering global employee counts, deployments by version, installation inventories, and virtualization and cloud environments, refuse the scope without refusing the relationship: "We are happy to address the specific concern you have identified.
That request goes beyond it, so tell us which downloads you are asking about and we will respond to those." If the contact came as soft outreach, keep it there deliberately, since a soft letter is not a contractual event: reply from procurement rather than the named CIO or General Counsel.
Respond in writing, do not accept a formal audit clock, and never use the word audit yourself.
Three sentences are unrecoverable. "We cannot migrate before renewal" tells Oracle the substitute is fiction and hands back list. "We have budget approved for the subscription" is a signed order in everything but form.
"Our developers prefer Oracle builds" concedes that the substitute is technically inferior, which is the only defense Oracle has left. The first GLAS call script covers the withholding rules in detail.
The pattern we see repeatedly is that buyers lose on tone, not on facts. A cooperative refusal ("here is what we will answer, and here is why the rest is out of scope") holds up through months of GLAS pressure.
A combative refusal invites escalation to a formal notice and the three-to-nine-month process that follows. A cooperative full disclosure is the worst of all outcomes, because it arms the other side.
Assume every sentence you speak is written down and reused in the next proposal, because it is.
Prepare the four or five sentences you are willing to say about scope and schedule, agree them with legal in advance, and let anything outside that set go to a written follow-up rather than an improvised answer on the call.
Structuring the settlement your exit earns
A credible exit does not usually buy you a discount on a three-year term. It buys you a different shape of deal, and shape is worth more than percentage.
The first structure is a transition subscription: 12 to 24 months, priced to a scoped population rather than the full employee count, bought explicitly to cover the runway between today and your migration completion date.
The second is a migrate-credit structure, where a defined portion of what you pay in the transition period offsets a reduced tail on any residual Oracle estate you cannot move (typically WebLogic-coupled Java SE, which carries its own trap covered in the Java SE bill hiding inside your WebLogic migration).
Both structures only exist because you have a named end state; Oracle does not offer them to a customer whose alternative is hypothetical.
In twenty-five years of these negotiations, the clause set matters more than the headline rate, because a good rate on evergreen paper becomes a bad rate at month 37.
Insist on four things in writing.
A fixed term end date with the words "expires without renewal," not "term of 24 months." Explicit removal of evergreen and auto-renewal language, including any notice-period clause that requires you to cancel; if there is no renewal, there is nothing to cancel. Written closure of the audit period.
Naming the years covered and confirming Oracle releases all claims for commercial use prior to the effective date, because an unclosed period is a repeat demand waiting for your renewal cycle.
And confirmation that legacy perpetual Java SE Advanced holdings remain valid as audit cover for the estate they were bought for, even though Oracle will refuse to credit them against the subscription price. Get that last point in the ordering document, not in an email from a sales rep.
You do not need an uplift cap in a non-renewing agreement, and asking for one signals you expect to renew. Compare the trade-offs across the three settlement structures side by side before you pick one.
The evidence base: what we see repeatedly in 2026 GLAS files
Gartner estimates one in five organizations will receive a formal Oracle Java audit notice by the end of 2026.
Formal notices now name a window commonly of forty-five days and set an expansive scope covering global employee counts, deployments by version, installation inventories, and virtualization and cloud environments.
The 2026 files share a signature. The letter is no longer soft outreach from a sales rep; it is addressed to a named CIO, CFO, or General Counsel and signed by a GLAS representative, which converts it from a commercial conversation into a contractual event with a running clock.
Before that letter arrives, Oracle already holds download telemetry: it cross-references JDK download logs against its subscription database and flags IP ranges appearing in one and not the other, treating the download record as hard evidence rather than an opening question.
Two traps recur in almost every file we see. The first is embedded JREs shipped inside third-party applications with no papered OEM agreement behind them, which Oracle counts against you unless the ISV can produce its distribution license.
The second is the 8u211 auto-update boundary, where machines that legitimately ran free 8u202 silently patched forward across the paid threshold, converting a compliant estate into a claim without anyone making a decision.
Both are answerable, but only with logs you collected before the notice arrived. Read the first GLAS call guidance before anyone in your organization responds.
The pattern worth the most money is the one buyers overlook: pre-2023 contracts sometimes carry legacy usage-based pricing that survives, and Oracle will describe it as obsolete because describing it that way is free.
Pull the original ordering documents and the governing agreement before you accept that the employee metric is your only option. We have seen files where the old processor-based paper covered the majority of the disputed estate at a fraction of the Universal Subscription number.
Read the notice against the clock it actually starts. A soft letter creates no obligation; a formal notice starts the timelines written in your agreement, and the average file runs three to nine months from letter to settlement, which is roughly the window a funded migration needs to become visible.
- 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
- Build the Java Usage Inventory before you answer anything, separating Oracle JDK from OpenJDK distributions and embedded runtimes (WebLogic, database clients, third-party ISV packaging), because Oracle already holds your download logs and cross-references IP ranges against its subscription database, so an inventory you did not build is one GLAS builds for you.
- Name one replacement vendor build in writing this quarter, a single distribution (Azul, Adoptium, Amazon Corretto, BellSoft, Red Hat) with a signed support agreement or quote attached, because "we are evaluating OpenJDK" reads as posturing while a named build with a purchase order reads as a priced alternative Oracle must underbid.
- Get a funding line approved and a tranche schedule signed by an application owner, not by IT alone: board or CFO approval of the migration budget plus dated tranches owned by named application leads is the artifact that converts intent into a schedule, and it is the core of the CFO business case to leave Oracle Java.
- Archive substitution receipts and de-installation tickets against every historical download, so a 2018 download record is answered with a change-management ticket showing the company-wide uninstall date and a vendor log showing where your updates actually come from; GLAS respects receipts and dismisses intent.
- Open the settlement conversation only after tranche one has actually completed, not when it is planned, and target a fixed-term transition subscription sized to the remaining runway rather than an evergreen employee-metric commitment, using the October 2026 JDK 21 patch cliff as your date and reviewing the settlement structures compared before you name a number.
Frequently asked questions
Does threatening to leave Oracle Java actually reduce the price?
Only when the threat is backed by artifacts Oracle can verify: a named replacement build, a dated tranche schedule, and an approved funding line. Verbal intent and "we are evaluating alternatives" are discounted to zero because GLAS sees the same language on hundreds of accounts each quarter.
A migration with even one completed tranche moves the number more than any argument about fairness.
What happens to Oracle JDK 21 in October 2026?
Oracle JDK 21 updates through and including September 2026 are available under the free No-Fee Terms and Conditions license.
Beginning with the October 2026 Critical Patch Update, further Oracle JDK 21 updates are planned under the Java SE OTN license, the same paid-for-production license used for Java 8, 11 and 17. Java 21 builds from Java.com are free to use until 16 September 2026.
Is there a later free window I can migrate into instead?
Yes. Oracle has stated that JDK 25 updates are planned to remain under the NFTC until October 2028, one year after the next planned LTS, Java 29.
That gives disciplined buyers a second lane, but it is a plan-and-fund decision, not a reason to delay, because the OTN switch on 21 still applies from the October 2026 CPU.
How long does an OpenJDK migration realistically take?
Plan 9 to 14 months for a multi-application estate, with the bulk of the effort in inventory, third-party vendor certification, and change control rather than in the JVM itself. Compress it by running tranches in parallel and starting with the applications Oracle's download logs point at.
Publish the schedule with named owners, because an undated plan reads as intent, not commitment.
What evidence does GLAS accept as proof I am no longer using Oracle Java?
Substitution proof and de-installation proof. Substitution proof means receipts, repository logs or support contracts from the alternative vendor showing you take your updates from them.
De-installation proof means archived change-management tickets and logs establishing a dated, company-wide uninstall, so a 2018 download record can be answered with a documented removal date rather than an argument.
Do my legacy perpetual Java SE Advanced licenses help my negotiating position?
They help as audit cover and migration runway for the deployments they lawfully support, and they should be produced in a claim. They do not earn any credit against the Universal Subscription price, so do not build a discount case on them.
Also read any pre-2023 contract in full before Oracle calls it obsolete, because usage-based pricing terms sometimes survive in that paper.
Why does the Employee metric make an exit more attractive than it looks?
Because price is decoupled from footprint. A 5,000-employee company running Oracle Java on 40 servers still pays for 5,000 employees, roughly $630,000 a year at list or $15,750 per server, and the bill grows with hiring even if the deployment shrinks.
Migration cost is a one-time, largely fixed number, so the payback case improves every year you defer it.
