Oracle Java Settlement Structures Compared: Subscribe, Migrate-Credit, or Walk

Oracle Global Licensing and Advisory Services (GLAS) offers three deal shapes to close a Java claim, and they are not priced equally. This analysis quantifies each structure and names the one that minimizes your total cost over the full term.

When a Java claim lands, GLAS does not present you with a single number. It presents a menu of deal shapes, each engineered to convert an uncertain past-use exposure into predictable, compounding future revenue. In 25 years of negotiating against this vendor, I have watched buyers accept the wrong structure not because the headline number was high, but because the five-year total was invisible at signing. This page compares the three structures GLAS actually offers, quantifies the long-tail cost of each, and tells you which minimizes total cost.

Before we compare, understand the machine you are negotiating against. The Java SE Universal Subscription is priced per employee, not per user and not per server. Oracle's own definition of Employee counts every full-time and part-time worker plus every agent, contractor, outsourcer, and consultant supporting internal operations, regardless of whether they ever touch Java. A 5,000-employee company running Oracle Java on 40 servers pays $630,000 a year at list; the same headcount on 100 servers pays exactly the same. That flat-against-deployment design is why the settlement structure you choose matters more than any technical remediation you perform.

The three structures on the GLAS menu

Every Java settlement collapses into one of three shapes, or a hybrid of them. Naming them precisely is the first act of buyer-side control, because GLAS blurs the boundaries deliberately.

• Structure 1: Back-fee waiver in exchange for term (migrate-credit). Oracle forgives the retroactive claim in return for a multi-year forward subscription. This is the structure GLAS pushes hardest because it converts a one-time, contestable penalty into guaranteed recurring revenue.
• Structure 2: Forward subscription with back fees priced in. Oracle bills historical unlicensed use as licenses, then attaches 22 percent annual support on that base, which then escalates. This is the most expensive structure over five years and the one buyers understand least.
• Structure 3: Walk (OpenJDK migration). You migrate off Oracle JDK entirely, terminate or never sign the subscription, and close the claim by removing the product. This is technically the lowest-risk path and financially the cheapest, but only if the timeline is managed before the renewal or audit clock forces your hand.

GLAS is not selling a penalty. It is buying a five-year annuity, and the back fee is merely the down payment it is willing to waive to close the sale.

Structure 1: Back-fee waiver in exchange for term

This is the structure GLAS offers first because it is the one Oracle wants. The pitch is consistent: "Sign a three-year Java subscription now, and we forgive the past unlicensed use." We have seen a company purchase a three-year subscription for all employees and watch Oracle drop a $1M back charge, and another negotiate a two-year agreement at a significant discount while Oracle waived historical fees entirely. Oracle's primary aim is future subscription revenue, not retribution, so the back claim is the softest line in the entire file and is routinely reduced or waived when the buyer defends properly.

The trap sits inside the term. The waiver looks like a gift, but the price of the gift is a multi-year, per-employee commitment against your entire headcount, priced against whatever discount you managed to negotiate at signing and locked for the full term. If you sign a three-year deal at 28,000 employees and your migration project could have retired Oracle Java in nine months, you have paid for roughly 39 months of subscription to erase a back fee that was probably contestable to near zero. The waiver is only a win if the forward number is genuinely lower than the walk-away number, and it almost never is once you count the full term. For the mechanics of dismantling the back-fee demand that GLAS uses as leverage, see how to push back on Oracle's retroactive Java back-fee demand.

When migrate-credit actually wins

Migrate-credit is the right structure in a narrow case: your estate genuinely cannot migrate inside the term window, your headcount is stable or falling, and you have secured a hard price hold plus an audit cap in writing. If any of those three conditions fail, the waiver is bait. The value of the forgiven back fee is finite and one-time; the value of a three-year commitment against a growing employee count is unbounded and compounding.

Structure 2: Forward subscription with back fees priced in

This is the structure that looks the same as migrate-credit but costs materially more, because the back fees are not waived; they are converted into licenses that carry support. When a settlement includes backdated license fees, Oracle applies its standard 22 percent annual support charge on those licenses, and that support base then escalates at 8 percent per year compounding. A $500,000 license settlement today translates to $700,000 or more in support obligations over five years. The headline settlement is the smallest number you will ever pay under this structure.

Two contractual traps ride along with this structure and both survive the negotiation if you do not surface them. First, order documents can carry a hidden $50K to $100K annual floor that survives headcount drops, so even if you migrate and shrink your Java footprint to zero, the floor keeps billing. Second, several 2026 renewal letters quietly drop the prior 36-month audit cap and replace it with language permitting Oracle to audit at any cadence. You must negotiate both the floor out and the audit cap back in writing. For the floor mechanics specifically, see how to avoid the hidden $50K Java subscription floor in the order.

Read the illustrative support figure as the point of the entire structure. The 22 percent support charge and the 8 percent compounding escalation are not negotiable concessions Oracle grants; they are the design of the annuity. This is why we advise buyers to refuse any structure that converts back fees into a support-bearing license base. The legal argument is clean: during the retrospective period the organisation received no actual support services, so there is no service-for-payment justification for backdated support. Do not pay a maintenance stream on a period during which nothing was maintained.

Structure 3: Walk (OpenJDK migration)

Walking is not a threat you use to improve a deal; it is a deal structure in its own right, and across the five exit paths we have run, walking and its variants delivered reductions of 60 to 95 percent against the Oracle baseline. Migration off Oracle JDK to any major OpenJDK distribution is typically a drop-in replacement for the same major version (8, 11, 17, 21). It is a packaging change, not a code change. All four major distributions (Temurin, Corretto, Zulu, and Liberica) are free for production commercial use. The technical risk that GLAS wants you to fear is mostly imaginary. See the OpenJDK alternatives compared on cost, SLA, and migration risk for the distribution decision.

The one structural catch that makes walking an all-or-nothing decision: the per-employee subscription does not permit a split. The metric counts your entire employee base regardless of which estate runs Oracle Java versus OpenJDK. Once the subscription is in place, you pay per employee regardless of installed footprint. So a half-migration buys you nothing; the buyer-side response is to migrate the entire estate and terminate the subscription. The exception is entity scoping after partial migration, where you license only a defined legal entity and migrate the rest, which is a genuine middle path but requires disciplined organizational boundaries and Oracle's agreement to scope.

You cannot partially subscribe your way to savings. The metric counts every employee whether they touch Java or not, so the only version of walk that works is the complete one.

The walk window is defined by the clock, not the code

OpenJDK cutovers finished inside 90 to 180 days wherever the application vendor certification conversation started before the renewal clock, and slipped past the renewal wherever it did not. That single sentence is the whole game. Your ability to walk is bounded by whether you started the vendor certification conversations early enough to complete migration before the renewal or audit deadline forces a signature. This is why a credible, evidenced migration plan is also your strongest negotiating lever even if you never intend to complete it: it caps what GLAS can plausibly demand. See using a credible OpenJDK exit to lower your Oracle Java number.

The version cliffs that turn a passive estate into a converting one

Any structure comparison is meaningless if your estate is silently converting to paid status under you. JDK 17's NFTC license expired in September 2024; build 17.0.12 from July 2024 was the last free update, and every release since falls under the OTN agreement requiring a subscription for production use. JDK 21 is next: all updates through September 2026 are free under the NFTC, but the October 2026 Critical Patch Update moves to OTN, the same paid license as Java 8, 11, and 17. The danger is that patch pipelines pull post-cliff updates automatically, so the estate that does not actively manage its Java versions is buying subscriptions one automated update at a time. Freeze your version inventory before you negotiate any structure, or you will be arguing about a footprint that is growing while you talk.

Which structure minimizes total cost

On total five-year cost, the ranking is consistent and it rarely matches the order GLAS presents. Walk (full OpenJDK migration) is cheapest, delivering 60 to 95 percent reductions against the Oracle baseline and eliminating support escalation, the hidden floor, and future audit exposure in one move. Migrate-credit is second-best and only when migration is genuinely impossible inside the term and you have locked price protection. Forward subscription with back fees priced in is the most expensive structure and should be treated as the floor you negotiate away from, never the deal you accept.

If you cannot walk cleanly, the two clauses that decide whether a subscription structure is survivable are price protection and the audit cap. Negotiate a fixed per-employee rate for the full term, or cap escalation at 2 to 3 percent annually rather than the 8 percent default. For 2026 contracts, request the lower of CPI or a fixed 3 percent. Both are only achievable at a contractual event: an Ordering Document amendment, co-term, true-up, renewal, or ULA exit. Outside one of those events, Oracle has no obligation to entertain the redline, so time your ask. For the full set of price movers, see the levers that move Oracle Java off list price, and for the overarching engagement approach, start with the buyer-side strategy for negotiating with Oracle GLAS on a Java claim.

What to do next

• Freeze your Java version inventory now and block automatic patching past the NFTC cliffs (JDK 17 already converted; JDK 21 converts at the October 2026 CPU), so your negotiated footprint does not grow mid-negotiation.
• Model all three structures on a five-year total-cost basis, not the headline settlement. Explicitly compute the 22 percent support plus 8 percent compounding on any converted back-fee licenses.
• Treat the back fee as the most negotiable line in the file. There is no service-for-payment justification for backdated support, so anchor at zero.
• If any subscription structure survives, capture the price hold (lower of CPI or 3 percent), remove the $50K to $100K floor, and reinstate the 36-month audit cap in writing at the contractual event.
• Score a full OpenJDK walk against the deal. Start application vendor certification conversations before the renewal clock, because the 90-to-180-day window only opens if you begin early. Never accept a partial subscription, because the metric bills the whole employee base regardless.

Related advisory: the full topic guide.

Frequently asked questions

Which Oracle Java settlement structure is cheapest over five years?

A full OpenJDK migration (the walk) is cheapest, delivering 60 to 95 percent reductions against the Oracle baseline and eliminating support escalation, the hidden minimum floor, and future audit exposure. Migrate-credit ranks second only when migration is impossible inside the term. Forward subscription with back fees priced in is the most expensive because it adds 22 percent support escalating 8 percent per year.

Is Oracle's Java back fee really negotiable to zero?

Yes, in defended settlements the back claim is routinely reduced or waived entirely. Oracle's aim is future subscription revenue, not punishment, and there is no service-for-payment justification for backdated support because no support was actually delivered during the retrospective period. Anchor the back fee at zero and negotiate the forward deal on its own merits.

Can I subscribe for only the part of my estate that uses Oracle Java?

No. The Java SE Universal Subscription is priced per employee and does not permit a split. The metric counts your entire employee base including contractors and consultants regardless of which systems run Oracle Java versus OpenJDK. The only way to save money is to migrate the entire estate and terminate the subscription, or use entity scoping to license a single legal entity after migrating the rest.

What is the migrate-credit deal Oracle offers?

Migrate-credit is a back-fee waiver in exchange for term: Oracle forgives the historical claim if you sign a multi-year forward subscription. It converts a one-time, contestable penalty into guaranteed recurring revenue against your full headcount. It only wins when migration is genuinely impossible inside the term and you have secured a fixed-price hold and an audit cap in writing.

How long does an OpenJDK migration take?

Cutovers to distributions like Temurin, Corretto, Zulu, or Liberica finished inside 90 to 180 days wherever application vendor certification conversations started before the renewal clock. Migration is a drop-in packaging change for the same major version, not a code rewrite. The constraint is timing, not technical risk, so start certification conversations before the renewal or audit deadline.

What clauses must I add if I have to sign a Java subscription?

Capture a price hold that freezes your per-employee rate or caps escalation at the lower of CPI or 3 percent instead of the 8 percent default. Remove the hidden $50K to $100K annual floor that otherwise survives headcount drops. Reinstate the 36-month audit cap that several 2026 renewals quietly delete. All of these are only achievable at a contractual event such as a renewal, amendment, or true-up.