Oracle has three layers, and they feed each other.

At the bottom is Oracle AI Database, the relational database many of the world's banks, telecoms and airlines keep their transactions in, with Exadata machines built to run it and Java and MySQL beside it. On top sit the applications: Fusion Cloud ERP, HCM, SCM and CX for large companies, NetSuite for smaller ones, Oracle Health (the former Cerner) for hospitals, and the older E-Business Suite, PeopleSoft and JD Edwards that many customers still run. Under both, and now the largest change in the business, is Oracle Cloud Infrastructure (OCI): a public cloud with compute, GPUs, storage and managed databases.

For most of its history Oracle sold a licence once and a support contract every year after. Substantially all licence buyers take support and substantially all renew it, and that stream carried the company. The cloud moves the same customers to subscriptions and usage bills: a database customer moves to Autonomous AI Database or Exadata on OCI, or to Oracle Database@Azure; an E-Business Suite customer moves to Fusion.

OCI is the newer story. Oracle built it as a second-generation cloud, launched in 2016 with engineers from AWS on the team, around a flat RDMA network, bare-metal servers and control code kept off the customer's machines. Those choices suit large AI training jobs, which need tens of thousands of GPUs talking to each other with little delay. AI labs, OpenAI among them, have signed multi-year contracts for capacity, and Oracle is building data centers at a pace it has never attempted.

The customers are large enterprises and governments, now joined by AI companies, and they pick Oracle for different reasons in each layer: the database for systems that cannot fail, the applications for one data model across finance, HR and supply chain, the cloud for GPU capacity, for Oracle workloads and for the option of running the cloud in their own buildings or countries. Since September 2025 two chief executives run the company: Clay Magouyrk, who led OCI, and Mike Sicilia, who led the industry applications, with Larry Ellison as Executive Chair and Chief Technology Officer.

An Oracle customer's core systems are written for Oracle. A bank's ledger carries years of PL/SQL code, reports and integrations built against Oracle Database; moving it to PostgreSQL or SQL Server means rewriting and retesting all of it, retraining the team and running both systems side by side for months, with the risk landing on systems that move money. Packaged software adds to it: many industry applications are certified on Oracle, so leaving the database can mean leaving the application too. That is why substantially all support customers renew, year after year.

The applications hold customers the same way. Fusion and NetSuite carry a company's ledger, payroll and supply chain, and the processes and training built around them; replacing an ERP is a multi-year project most companies attempt once a decade. Oracle Health customers face the same arithmetic, and some have paid it to move to Epic.

OCI is the least sticky layer. Compute and GPUs can be rented from AWS, Azure, Google Cloud or GPU specialists such as CoreWeave. What holds a customer there is the work of moving data and pipelines, multi-year capacity contracts, and for Oracle workloads, Exadata, which runs only on Oracle's own hardware, whether in OCI, in a customer's building or inside another cloud.

What customers most often try to escape is Oracle's licence terms and audits. The long migrations to PostgreSQL and Aurora that many companies run show that leaving is possible, and slow.

Oracle's shape changed in fiscal 2025 and 2026 (its years end in May).

Revenue was $67.4B in fiscal 2026, up 17.3%, after growth of 6.0% in fiscal 2024 and 8.4% in fiscal 2025. For years before that it barely moved, from $37.8B in fiscal 2017 to $40.5B in fiscal 2021. Over the five years to fiscal 2026 it grew about 10.7% a year, helped by Cerner from fiscal 2023.

Software support and licences cost little to deliver, which is why the operating margin held around a third of revenue: 37.6% in fiscal 2021 and 30.6% in fiscal 2026. Net income was $17.1B in fiscal 2026, a 25.4% net margin.

The cloud is what changed the cash. Building OCI data centers and filling them with GPUs took capital spending from $6.9B in fiscal 2024 to $21.2B in fiscal 2025 and $55.7B in fiscal 2026, 82.6% of revenue. Operating cash flow rose to $32.0B, but free cash flow was negative $23.7B in fiscal 2026, after negative $0.4B in fiscal 2025: the only two years in the decade when spending outran the cash from operations. After equity pay of $4.8B (7.1% of revenue), it was negative $28.5B.

The servers are bought before the contracts they serve start paying, so the revenue from them arrives over the years that follow. Oracle ended fiscal 2026 with $31.3B of cash.

Oracle is turning from a software company into an infrastructure company that also makes software. The plan has three parts.

First, OCI capacity at very large scale: GPU clusters for AI labs and for companies training and running models, with multi-year contracts already signed for much of what is being built.

Second, the multicloud database. Oracle Database@Azure, @AWS and @Google Cloud put Exadata inside the other clouds, so customers can move Oracle workloads to the cloud of their choice and still run them on Oracle.

Third, AI inside the applications and the database: agents in Fusion and NetSuite that do routine finance and HR work, the Clinical AI Agent and a new EHR in Oracle Health, and vector search in Oracle AI Database so companies can point models at their own business data without copying it elsewhere.

The applications keep moving to the cloud as E-Business Suite, PeopleSoft and on-premise database customers migrate, a slow and steady shift. The fast shift is OCI's, and the business it produces looks different from the old Oracle: heavy spending on buildings, power and chips up front, revenue over many years, and a few very large customers. How well Oracle runs that buildout, on time and at the cost it planned, will shape the company more than any single product.

• Building at this scale. Data centers, power, cooling and GPUs on a timetable set by AI customers, with capital spending of $55.7B in fiscal 2026 against operating cash flow of $32.0B. A delay in power or chips holds back revenue the contracts already promise.
• A few very large customers. Much of OCI's new capacity is committed to a small number of AI companies whose own plans and funding can change.
• Competing with the larger clouds. AWS, Azure and Google Cloud have deeper service catalogs and far more developers who know them; GPU specialists such as CoreWeave compete for raw capacity, and NVIDIA sells the same chips to all of them.
• Keeping the database chosen for new apps. PostgreSQL is the default for new projects, and many customers are moving off Oracle Database to get away from its licensing.
• Oracle Health against Epic, with a federal rollout that has to win back clinicians' trust.
• Two chief executives leading a company whose figures were built on software margins through its most capital-heavy years.