Independent Research · Unvarnished Reviews
Unvarnished Reviews Research
The market has already been voting with its feet, and the trend is unusually well-documented for this category. Oracle supplied roughly 75% of the JDKs enterprises ran in 2020; that share fell to 34% in 2022, 29% in 2023, and 21% in 2024, tracking almost exactly with Oracle's January 2023 shift to per-employee Java SE pricing. Eclipse Temurin has been the primary beneficiary: One 2026 industry survey now places Temurin adoption at 31%, ahead of Oracle Java's 27%, a genuine reversal of the two vendors' historical positions. Azul's own 2025-2026 State of Java survey, a widely cited primary source in this space, found that 81% of organizations are migrating some or all of their Oracle Java environments to OpenJDK distributions, and 99% of surveyed organizations still actively use Java itself; this is a migration away from one vendor, not away from the language.
Java is free. It has been open source for nearly two decades, and identical, TCK-certified builds of it are available at zero cost from Amazon, Microsoft, IBM, the Eclipse Foundation, and Azul. Oracle's price for the same runtime: $15 per employee per month, every employee, including the receptionist, the warehouse crew, and per the contract's stated definition, your contractors and outsourcers, whether or not any of them has ever heard of Java. A 500-person company running Java on thirty servers owes Oracle for 500 people. This is the only major software subscription we have reviewed where the metric has no relationship whatsoever to usage.
| Distribution | G2 Rating | Reviews (G2) | Note |
|---|---|---|---|
| Azul Platform Core (Zulu) | 4.9 / 5 | ~58 | Highest-rated, smallest commercial vendor of the four |
| Amazon Corretto | 4.5 / 5 | ~116-123 | Free; largest G2 review base among the OpenJDK builds |
| Oracle Java (JDK) | 4.4 / 5 | Not consistently reported | The commercial, subscription-billed option this report is about |
| Eclipse Temurin | Not meaningfully available | 2 | G2 sample too thin to be representative; as a free, foundation-run distribution, most buyers don't formally review it the way they would a commercial product |
The rating pattern here reflects the category more than any real quality gap: Free, community-run distributions like Temurin are downloaded and deployed at massive scale but rarely formally reviewed, since there's no sales relationship prompting a review request. Treat the absence of a rating for Temurin as a data-availability gap, not a quality signal.
In January 2023, Oracle replaced its per-processor and per-user Java SE licensing with the Java SE Universal Subscription, priced per employee per month. Oracle's FAQ states it plainly: Pricing starts at $15 per employee per month, with published tiers falling to $5.25 for organizations above 50,000 employees. Total cost scales with headcount, never with Java footprint.
Two clauses do the real work. First, the definition of "employee": The price list counts all full-time, part-time, and temporary employees plus the employees of your agents, contractors, outsourcers, and consultants supporting your operations. Redress Compliance notes organizations routinely discover in audits that they undercounted: A 500-person company with 80 contractors and temps licenses 580.
Second, the enforcement. Java audit activity rose materially through 2024–2025, and Gartner has predicted that one in five Java-using organizations will face an Oracle audit by 2026. The tactics are distinctive: Palisade Compliance reports Oracle reps no longer ask how much Java you run. They arrive citing how many times your company downloaded Java from Oracle's site and your public headcount. A single developer pulling the Oracle JDK out of habit is the tripwire; your public employee count is the invoice.
Pre-2023 legacy contracts remain valid but cannot be expanded, a slow squeeze toward the employee metric at every renewal.
OpenJDK is the reference implementation, and the free distributions pass the same Technology Compatibility Kit as Oracle's build:
Migration, per Redress benchmarking across 30–40 Java estates, is rarely technical: Oracle-specific dependencies are uncommon outside WebLogic and a small set of JavaFX cases. The work is inventory and testing, not rewriting.
| Option | Metric | Per-Year | 3-Year |
|---|---|---|---|
| Oracle Java SE Universal (500 employees, $15 tier) | Headcount | $90,000 | $270,000 |
| Oracle, counted the way audits count (580 incl. contractors) | Headcount | $104,400 | $313,200 |
| Azul Platform Core (indicative, per Redress 20–40% benchmark) | JVMs/cores | ~$18,000–$36,000 | ~$54,000–$108,000 |
| Corretto or Temurin | N/A | $0 license | $0 license |
The Corretto/Temurin line is not free in the total sense: The real spend is a one-time migration: an estate inventory (which apps, which JDK versions, which download sources), swap and test, plus ongoing patch discipline your team now owns. For thirty standard workloads this is weeks of engineering time, not a program. There is no annual line item at all.
Per employee, the Oracle math gets worse the more plainly you say it: At this company, Oracle charges $180 per year for every person in the building to run software that is free, in exchange for support most teams open a ticket against approximately never.
What does Oracle's $270,000 actually buy? Integrated support across the Java ecosystem, GraalVM entitlements, indemnification, and, for organizations deep in Oracle middleware, one throat to choke. If you run WebLogic, use Oracle-proprietary extensions, or your risk office requires the vendor of record for the runtime to be the runtime's steward, the subscription is defensible. And the Universal Subscription is, in fairness, simple: No counting processors, no true-up anxiety on the metric itself. You are buying an audit-shaped peace of mind, from the party conducting the audits.
The complexity picture here differs from a typical software migration. Because all four distributions build from the same OpenJDK source, independent sources consistently describe switching between them as largely a binary-compatible runtime swap, the harder work is auditing where Java actually runs across an estate, not the migration itself.
| Dimension | Oracle Java | Azul | Amazon Corretto | Eclipse Temurin |
|---|---|---|---|---|
| Migration/swap effort | N/A, the incumbent | Low; binary-compatible runtime swap per independent sources, with Azul's own migration tooling supporting the transition | Low; described consistently as a runtime swap and audit exercise rather than a rebuild, since Corretto tracks upstream OpenJDK closely | Low; the most widely deployed community OpenJDK build, independently described as the vendor-neutral default specifically because it introduces no new dependency |
| Support model | Commercial, per-employee subscription with Oracle Premier Support | Broadest support portability of the three alternatives: Commercial SLA-backed support available regardless of where workloads run (on-premises, any cloud, hybrid), the closest structural analog to Oracle's own support model | Real portability constraint independently documented across multiple sources: Commercial support is available only through AWS Support plans and only for AWS-hosted workloads; on-premises and multi-cloud estates fall entirely outside that boundary, and Amazon has stated no plans for a standalone Corretto support tier | Community-only from the Eclipse Foundation directly; commercial support is available, but only through third parties (including Azul, Red Hat, and IBM), an extra step organizations should budget planning time for if formal SLA coverage is required |
| Audit/discovery complexity | N/A, the incumbent | Same estate-discovery work applies regardless of destination distribution; not distribution-specific | Same estate-discovery work applies; the practical constraint is confirming which workloads are AWS-hosted before assuming Corretto's support boundary covers them | Same estate-discovery work applies; no distribution-specific complexity beyond the general audit effort every migration path requires |
The practical takeaway: The genuine complexity in this decision isn't the swap itself, independent sources are consistent that a binary-compatible runtime change is comparatively simple. It's confirming where Java runs across the estate first, and matching that footprint against each distribution's support boundary. Corretto's support model is the one with a real, structural constraint worth planning around: Commercial support exists only for AWS-hosted workloads, a hard boundary for any organization with on-premises or multi-cloud Java, not a limitation Azul or Temurin's third-party support model shares.
Running Oracle JDK out of habit, no Oracle-specific dependencies: Migrate to Temurin or Corretto. This is the highest-ROI licensing move in enterprise software right now, a six-figure recurring cost eliminated for a one-time testing effort. Sweep your estate for Oracle download activity first; that history is what audits are built on.
Legacy Java 6/7/8 estates that need supported patches: Azul's extended support is the purpose-built answer, at a metric tied to your JVMs instead of your org chart.
WebLogic and deep Oracle middleware shops: You are the customer this subscription was priced for. Negotiate the tier, cap the escalator (8% annual escalators appear in the wild, cap it in writing), and above all challenge the employee count: The cheaper move in most reviews is not a better rate but a smaller counted estate.
Everyone: Establish a JDK provenance policy today: which distribution, which download source, documented per install. The companies that get hurt are the ones that cannot answer "where did this JDK come from" when the letter arrives.
Oracle's Java pricing is a masterclass in decoupling price from value: The metric is your payroll, the enforcement is your download logs, and the product is available elsewhere for free, same standard, same compatibility kit. For a narrow band of Oracle-committed enterprises, the subscription buys real integration and cover. For everyone else, it is a tax on not having gotten around to a migration that, on the evidence, is mostly an inventory exercise. The 20% audit prediction is the deadline; do the inventory before Oracle does it for you.