Oracle Doesn't Sell Software. It Sells the Inability to Leave.

A licensing advisor draws up the math for a small firm: 250 employees, 20 people who occasionally open a Java desktop app. Under the old rules, that company paid about $3,000 a year. Under the rules Oracle introduced in January 2023, the same usage costs $45,000 a year — because the new meter doesn't count the 20 users, it counts all 250 employees, contractors included.⁴¹⁰ A roughly 15-fold increase at list price, and the software did not change. That gap between what you use and what you pay is not a billing quirk. It is the whole business.

The official story is that Oracle is a database company that wins because its database is excellent. The more useful story is that Oracle long ago stopped needing to win. In fiscal 2025, cloud services and license support — recurring fees from customers it already has — were $44.0 billion of $57.4 billion in revenue, about 77%, while new license sales were a comparatively tiny $5.2 billion.¹ When three-quarters of your money comes from people who already can't leave, product merit becomes optional. The moat isn't the technology. The moat is the exit.

Why ripping out 90% of Oracle saves you almost nothing

Lock-in starts in the code, not the contract. Decades of enterprise logic get written in Oracle's own dialect: PL/SQL stored procedures, functions like DECODE, CONNECT BY and ROWNUM that have no clean equivalent elsewhere, plus Oracle Text, Spatial, Forms and APEX layered on top.⁹ To leave, you don't migrate data — you rewrite the business rules baked into the database itself, line by line, while it runs. That's the technical switching cost, and it's real. But the genius is what Oracle does with it commercially. Cut your Oracle footprint by 90% and you'd expect to cut the bill by something close to that. You don't. Advisors who do this for a living report that Oracle simply reprices what's left, so the remaining estate carries nearly the same total spend it did before.⁹ You can shrink your usage all you like; the invoice was never tied to it.

The 'unlimited' license that takes away your freedom

Oracle's answer to a nervous, growing customer is the Unlimited License Agreement — typically a three-year deal with an upfront fee often in the millions, sold as the freedom to deploy a listed set of products without counting.⁷ The word 'unlimited' does a lot of quiet work. The unlimited rights apply only to the products explicitly named; entering the agreement folds your existing perpetual licenses into it, so you can't revert to what you owned before; and once you certify at the end of the term, your annual support fees — running around 22% of license cost — are frozen at that level and cannot be reduced even if your usage later collapses.⁷ There's a sharper hook: get acquired, and the ULA triggers accelerated certification, forcing a renegotiation at exactly the moment you have the least leverage.⁷ It is a contract designed to convert a moment of optimism into a permanent floor under your spending.

When the audit is really a sales call

The third layer is the one Oracle would least like described plainly. Licensing firm Palisade Compliance — whose founder spent 16 years at Oracle as Global VP of Contracts and Global Process Owner of Oracle's audit team — describes a tactic using what it calls Oracle's own vernacular: 'Audit, Bargain, Close' — ABC — in which a compliance audit is not a check on the rules but a lever to push the customer into buying Oracle Cloud.⁸ The pattern shows up in court: a securities class-action alleging Oracle threatened costly audits unless customers shifted to its cloud settled in January 2023, and in an earlier dispute with Mars, Oracle chose a confidential settlement in late 2015 over letting a judge scrutinize its licensing stance.⁸ The audit's effectiveness depends on uncertainty. House of Brick notes Oracle's audit scripts overstate exposure by not separating what a customer is entitled to use from features that were simply enabled by default — and, more aggressively, that Oracle has presented Java deficiency claims using the per-employee metric for 2021 and 2022, before that metric even existed.⁶ The bill arrives first; the customer argues it down second; the close is the cloud contract.

Isn't this just a strong product earning what it's worth?

The honest objection is that lock-in and value can look identical from the outside. Oracle's database genuinely is engineered to run mission-critical workloads, the switching costs are real partly because the platform is deep and capable, and customers signed every contract with their eyes open. A lot of what reads as a trap is just the ordinary friction of any serious enterprise system — and Oracle's own cash flow, $18.7 billion from operations in fiscal 2024, is what a useful product looks like.² Fair. But there is a tell that separates earned loyalty from engineered captivity, and Oracle just tripped it. A vendor confident in its value lowers friction to keep you; a vendor relying on captivity raises the cost of staying because it knows you can't go. The per-employee Java meter — Gartner pegs it at 2–5x the old cost, with a Dimensional Research survey finding 73% of Oracle Java users audited in the prior three years — is the second move, not the first.⁵ And Gartner forecast that by 2026, more than 90% of Java applications would run on third-party runtimes instead — a migration at a scale the installed-base model was never built to survive.¹¹ The moat held because leaving was expensive. Oracle has now made staying expensive too — and that is the one thing the model was never built to survive.

Oracle built one of the most durable revenue annuities in software by understanding something most vendors never accept: the sale that matters is not the first one, it's the impossibility of the last. Proprietary code makes leaving expensive. The ULA makes shrinking pointless. The audit turns uncertainty into a cloud contract. Stacked together, they convert technical switching costs into a near-permanent stream that no longer depends on the product being best — 77 cents of every revenue dollar comes from people who already can't walk away.¹ The threat was never a faster database. It was Oracle forgetting that a moat protects you only as long as the water is cheaper than the bridge. Charge enough to cross, and people will learn to swim.