The decision is not technical and it does not need a finished inventory. It needs four numbers, five people in one room, and an honest answer about the estate shapes where staying with Oracle is still the right call.
This page is the gate, not the project plan. It answers one question: given your estate, your headcount and your calendar, should you leave Oracle Java at all, and if so, what is the latest week you can start and still land it on time.
For most enterprises, yes, and the reason is commercial rather than technical. Oracle's Java SE Universal Subscription prices Java against your total counted headcount, while every alternative prices it against the machines that actually run it. That gap is the whole argument.
But "most" is not "all", and a gate that only ever returns one answer is not a gate. There are four destinations, and this page exists to tell you which one you are actually in before you spend a quarter of engineering time finding out.
The four destinations, and who ends up in each
| Destination | Who lands here | Recurring Java cost basis |
|---|---|---|
| Subscribe to Oracle | Low headcount, high JVM count, or a hard Oracle certification dependency | Total counted employees |
| Free non Oracle build, no support contract | Estates with a strong platform team and no external SLA obligation | Zero license cost, internal run cost only |
| Commercially supported non Oracle build | Regulated, latency sensitive, or contractually obliged to name a support vendor | Cores, servers or desktops |
| Remove Java from the workload | Small legacy estates where the application is being retired anyway | None, after a one time project |
We lay the four out properly, including the option most buyers never price, in the alternative Java options landscape. The pure decision arithmetic, framed against Oracle specifically, sits in our Oracle Java versus OpenJDK decision guide.
You do not need an inventory to run the gate. You need four numbers, and a competent platform lead can produce credible ranges for all four in an afternoon.
Ask whether any Oracle branded Java binary is running in production today, and insist on evidence rather than belief. In our engagement file this question splits estates into three groups, and two of them do not need a migration project at all.
In roughly half the estates we open, the buyer either stopped using Oracle Java and kept paying, or believed they had stopped and had not. Both are decided by evidence, not opinion.
Six dated triggers, and only one of them is technical. Urgency in Java licensing comes from calendars you do not control: Oracle's free use windows, your own renewal anniversary, and the renewal of an adjacent Oracle product where Java becomes leverage.
The triggers that move a Java decision from important to urgent
| Trigger | Why it forces the date | What to verify first |
|---|---|---|
| No Fee Terms window closing on your LTS | Continued production use after the window defaults to a paid subscription | Oracle's Java SE Support Roadmap page for the current stated date |
| Java subscription anniversary | Missing the notice window buys another full term | The ordering document, not the invoice, for the notice period in days |
| Adjacent Oracle renewal inside 12 months | Java becomes a bargaining chip in a much larger negotiation | Every Oracle agreement end date in one table, not just Java |
| Download activity on an Oracle account | The download record is the most common opening for an Oracle approach | Which accounts can reach Oracle download endpoints from your network |
| Acquisition or a step change in headcount | The counted quantity moves, and it only ever moves up | Whether acquired entities are already inside the counted population |
| An ISV changes its support matrix | A named runtime can pin you to a build you were planning to remove | The current matrix, in writing, for your top ten vendor applications |
Oracle's No Fee Terms and Conditions license permits free production use of an Oracle JDK long term support release for a bounded window, historically running to one year after the following long term support release ships. It is a genuine free grant. It is also a clock.
On Oracle's published roadmap the JDK 21 window closes on 16 September 2026, with the final free update delivered in July 2026, and JDK 17 already closed in September 2024 after build 17.0.12. Treat both as stated positions to confirm, not as facts you can build a business case on unchecked.
Check the current wording yourself on the Oracle Java SE Support Roadmap and on the license text itself, and screenshot both with a date. Oracle has changed Java terms three times in six years, and the version you plan against should be the version you can produce later.
Three things arrive dressed as emergencies and are not. An unsolicited note from an Oracle license team offering a "review" is a sales motion with a compliance jacket on. A partner email warning that you are "at risk" is usually a lead generation exercise.
The third is a general market article about Java pricing. None of these change your ratio, your calendar or your certification dependencies, and none of them should compress a decision you have not modeled. When a real approach does land, run the sequence in our Oracle Java audit defense playbook rather than answering questions in the order Oracle asks them.
Because it severs price from consumption. Oracle's Java SE Universal Subscription is quoted per employee per month, and the counted population is your workforce rather than your Java users. One server running Java obliges you to license the entire company.
Oracle's definition reaches further than most buyers expect. It sweeps in full time, part time and temporary staff, and the equivalent populations at your agents, contractors, outsourcers and consultants where they support your internal business operations.
Divide your counted employee number by your running JVM count. That single figure predicts the outcome of the gate better than any inventory tool, because it tells you whether Oracle is charging you for what you use or for who you employ.
Reading the ratio before you build a business case
| Counted employees per JVM | Typical estate shape | What the gate usually returns |
|---|---|---|
| Under 3 | Software vendor, trading firm, small headcount with a dense Java platform | Subscribing is genuinely competitive. Model it properly before you leave. |
| 3 to 20 | Technology heavy enterprise, large engineering function | Genuinely close. The certification dependencies decide it, not the price. |
| 20 to 100 | Typical large enterprise with a mixed application portfolio | Leaving wins on cost by a wide margin in almost every model we run. |
| Over 100 | Retail, healthcare, logistics, manufacturing with a large frontline workforce | The subscription is close to indefensible. Treat the renewal as the deadline. |
A retailer with 40,000 counted employees and 300 JVMs sits at 133. A trading firm with 400 employees and 900 JVMs sits at 0.4. Those two organizations are having completely different conversations, and generic advice serves neither.
The full cost model, including the internal effort that never reaches a business case, is priced end to end in our three migration patterns cost model, and the finance framing sits in the CFO business case to leave Oracle Java.
The counted quantity is where the money is, and it is more negotiable than the rate. Oracle's opening position treats every contractor with any system access as a counted employee, which is broader than the contractual language supports.
Across the Java negotiations Fredrik Filipsson worked in 2024 and 2025, Oracle's opening counted quantity ran materially above the number a buyer could defend after a clean headcount reconciliation. A narrower definition held in roughly four cases out of five.
Build that reconciliation before the first quote, not after it. The procurement mechanics are in our 20 critical Java SE procurement insights.
A 30 percent discount on a headcount you never had to concede is not a win. Fight the counted quantity first, the rate second.
In four estate shapes, and they are more common than the migration literature admits. An advisory that never recommends staying is not being independent, it is being predictable.
This is the case almost nobody writes about. If you employ 300 people and run 2,000 JVMs, the employee metric is charging you for 300 units to cover 2,000 machines, which is the cheapest Java you will ever buy.
Software vendors, quantitative trading firms and specialist engineering businesses land here regularly. Run the arithmetic before you accept a migration recommendation, because in this shape a footprint priced support contract can cost more than Oracle does.
If your Java runs inside Oracle products, the Java decision is not really a Java decision. Oracle WebLogic, E Business Suite, Forms and Reports, Hyperion and the Retail applications are supported against Oracle's own runtime, and moving the JDK underneath them can move you outside the certification matrix for the product.
The same logic applies to non Oracle vendors that name a specific build. Get the matrix in writing before you assume it, because vendor sales teams and vendor support teams frequently disagree about what is certified. The overlapping middleware exposure is covered in our Oracle iAS licensing guide.
Two contracts commonly grant Java rights that buyers forget they hold. Pre 2023 perpetual Named User Plus and Processor licenses remain valid for what they cover, even though Oracle no longer sells them and you cannot expand them.
The second is Java bundled inside a middleware or applications agreement. We regularly find organizations paying a Java SE subscription for servers already covered by an Oracle product entitlement. Check the embedded and original equipment terms in the Java embedded and OEM licensing guide before you renew anything.
Some buyers carry a contractual or regulatory requirement for a named commercial runtime vendor with specified indemnities. That requirement is usually satisfiable by a non Oracle support contract, but not always, and not always inside the window you have.
When the deadline is closer than the approval cycle for a new supplier, subscribing for one controlled term while you do the work properly is a defensible commercial decision. It is not a permanent answer, and it should be documented as a bridge with an end date.
If you do stay, the four things to fix before you sign
| Item | Why it matters |
|---|---|
| The counted quantity | It is the multiplier on every other number in the deal, and it is arguable. |
| Term length | Never commit multiple years on a headcount you expect to fall. |
| The processor ceiling | The subscription carries an upper limit on processors covered. Confirm yours in the ordering document. |
| Notice mechanics | Record the notice date and owner the day you sign, not the month you want to leave. |
In ninety minutes, with five named people and six documents on the table. The reason Java decisions drift for a year is almost never analysis. It is that nobody convened the people who can actually say yes.
Score each answer as leave, stay or unknown. Any question that scores unknown becomes a named action with a date, not a reason to postpone the decision.
Question eight decides more programs than questions one through seven combined. A gate that ends without a named owner and a reporting date has produced an opinion, not a decision.
Work backwards from the deadline, not forwards from today. The deadline is whichever comes first: your free use window, your renewal notice date, or the date an adjacent Oracle negotiation begins and you want Java off the table before it does.
Latest safe start, measured back from the date Oracle Java must be gone
| Estate size | Elapsed time | Start no later than | What sets the pace |
|---|---|---|---|
| Single platform team, fewer than 200 JVMs | 6 to 10 weeks | 3 months before | A single base image rebuild inside one approved window |
| Departmental scale, 200 to 1,000 JVMs | 4 to 7 months | 9 months before | Waiting on application owners to accept the change |
| Enterprise scale, 1,000 to 5,000 JVMs | 9 to 14 months | 16 months before | The release calendar, and the freezes inside it |
| Above 5,000 JVMs, or a validated environment | 12 to 20 months | 24 months before | Producing validation evidence for every wave |
| Any material desktop Java population | Add 3 to 6 months | Add to the above | Repackaging, a Web Start substitute, user acceptance |
Swapping a runtime is a configuration change. Getting two hundred application owners to accept a configuration change to a production system they are accountable for is an organizational problem, and it moves at the speed of your governance, not your build pipeline.
The phased schedule and the overruns we see most often are broken down in how long an OpenJDK migration actually takes. Treat that page as the delivery plan once this gate has returned an answer.
Add those up honestly before you promise a date. A twelve month plan with three freeze windows inside it is a nine month plan wearing a costume.
Source: Redress Compliance advisory engagement file, 2024 to 2025.
Three things, and only one of them appears on an invoice. Late decisions are expensive because they remove your options one at a time until the only remaining option is the one Oracle prefers.
The migration itself has to reach the places nobody counts. Build pipelines and developer laptops are covered in cleaning Oracle JDK out of CI/CD and developer workstations, and vendor products carrying their own runtime in when third party applications bundle Oracle Java.
The runtime is the same upstream code. Every mainstream distribution is built from the OpenJDK project and tracks the same quarterly security cycle, so the technical loss on a version for version swap is close to zero.
What you give up is the Oracle support desk, Oracle specific commercial features, and Oracle's position in your product certification matrices. Whether that matters is an application by application question, and it is the one worth spending real time on. The support and rollback exposure is examined in OpenJDK support and rollback risk after leaving Oracle Java.
The standard advice is to inventory first and decide afterwards. We disagree, and the reason is arithmetic. A full inventory takes two to three months in a real enterprise, notice windows are typically thirty days, and in nine estates out of ten the inventory does not change the destination because the employee to JVM ratio already decided it. What the inventory changes is the plan, the sequence and the cost, all of which you need after the decision, not before it. Decide the destination on four numbers this month, appoint an owner, then let the inventory shape the program. Buyers who wait for perfect data pay for another year of the thing they were trying to stop buying.
Run the ratio first: counted employees divided by running JVMs. Above roughly 20 the subscription is the most expensive route in almost every model we build, and below about 3 it is genuinely competitive. Between those two figures the deciding factor is how many applications carry a vendor support matrix that names an Oracle build.
Oracle's published roadmap states 16 September 2026, with the last free update delivered in July 2026. Verify the current wording on Oracle's Java SE Support Roadmap page and on the No Fee Terms license text, and keep a dated screenshot. Oracle has revised Java terms more than once.
No, not while a subscription agreement is in force. The No Fee Terms are a route for organizations that never subscribed, and an existing Java SE agreement governs your use regardless of which build you install. Ending the subscription cleanly at the notice date is the only route out of it.
Ninety minutes, if the right five people are in the room with the four inputs prepared. What stretches the decision to a year is convening rather than analyzing, and the single most commonly missing person is whoever chairs the change advisory board. Book that session before you commission any tooling.
Yes, in four shapes. A low headcount with a dense Java footprint, a certification matrix you cannot leave, Java rights you already hold through another Oracle agreement, and an obligation you cannot transfer to a new supplier inside the time available. Each of those is defensible, and each should carry a review date rather than being treated as permanent.
Waiting for a complete inventory before deciding the destination. The inventory takes two to three months, the notice window is typically thirty days, and the destination is usually already determined by the ratio. Decide first, then let the inventory shape the sequence and the cost.
It removes the ongoing obligation, but only once the binaries are gone and you can prove it. Oracle approaches are commonly opened from download records rather than from installed software, so the evidence pack matters as much as the migration. Keep dated inventories, block Oracle download endpoints, and retain proof of the date each environment was cleared.
Oracle Java SE Universal Subscription bills every employee, not just developers. The 2026 buyer guide to the cost math, audit exposure, and OpenJDK migration.
Gated with a work email on the download page. No sales follow up you did not ask for.
Get the White Paper →
500+ enterprise clients. 11 vendor practices. Industry recognized. One conversation can change what you pay for the next three years.
One buyer side briefing a week. Renewal signals, audit moves, and the levers that work. No vendor spin.
