Oracle V2 Ships — The First Commercial SQL Relational Database

In June 1979 a small Silicon Valley startup shipped version 2 of a relational DBMS called Oracle. By the company's own reckoning it was the first commercial SQL relational database management system on the market.

The strange part is who did not ship one. IBM had invented the relational model and built System R to prove it, and IBM's first commercial relational product, SQL/DS, did not arrive until 1981.

Three Names for One Company

The corporate chain is routinely garbled, so it is worth setting down against Oracle's own thirtieth-anniversary timeline:

Oracle's own account dates the change to Oracle Systems to 1982; a number of secondary histories put it at 1983. The one thing that is not in dispute is the direction: the database was named first and the company took its name afterwards.

Where the Name Came From

The three founders met at Ampex Corporation, and by Oracle's telling the origin was a piece of reading. Ellison, "routinely scanning the IBM Journal of Research and Development," found a paper describing a working prototype of a relational database management system, showed it to Miner and Oates, "and he soon learned that no company had committed to commercializing the technology."

The name came from elsewhere: Oracle was the code name of a CIA project the three had worked on at Ampex. The company history in the International Directory of Company Histories records that the founders "persuaded the Central Intelligence Agency (CIA) to let them pick up a lapsed, $50,000 contract to build a special database program." Oracle's anniversary issue reproduces a photograph of the Project Oracle user guide, captioned as a manual with "a limited but highly important audience—computer users at the CIA."

V1 Was Never Shipped

The version completed in 1978 was Oracle Version 1: written in assembly language, running on a PDP-11 under RSX, in 128K of memory, with an implementation that separated Oracle code from user code. It was never officially released.

Oracle V2, in June 1979, was the first version customers received — a decision usually explained as not wanting a version 1 with a reputation for bugs. It ran on the DEC PDP-11, was written in PDP-11 assembler, and used a syntax close to SEQUEL, SQL's precursor.

What the Relational Model Let Oracle Sell

The competing systems of 1979 were hierarchical (IBM's IMS) or network/CODASYL. Both required the application program to know how records were physically arranged and which pointers to follow. That coupling had a commercial consequence as much as an engineering one: a database was something you bought from the company that sold you the machine, tuned to that machine's file structures, and the queries you wrote were an asset of that installation and no other.

A relational system moves the description of what data is wanted above the description of how to reach it. The query names relations and predicates; the DBMS compiles that into access paths. Change the storage layout and the application does not move. Once the query language is that far above the metal, nothing in it is specific to a vendor's hardware — which is what let a startup with no machines to sell walk into accounts owned by companies that had them. Ken Jacobs, later Oracle's vice-president of product strategy, put the period plainly: "Back in the old days, databases came from the hardware vendor. Oracle was among the first to offer a DBMS that would run on different hardware and operating systems."

The first customer was the CIA, followed by other US government agencies. Institutions that needed to ask arbitrary questions of large classified holdings had the strongest possible reason to want the questions decoupled from the file layout.

The Portability Decision

Version 2 could only prove the argument on one machine. Version 3, in 1983, was rebuilt in C — the single decision that turned the theoretical portability of the relational model into a shipping product line. It became, by Oracle's account, the first RDBMS to run on mainframes, minicomputers and PCs.

"Version 3 of Oracle was the portable Oracle. It was written in C, ran on hundreds of operating systems and hardware platforms, and it really allowed us to expand our marketplace dramatically." — Ken Jacobs, vice-president of product strategy, Oracle

The discipline behind it mattered as much as the language. Andy Mendelsohn, then a senior vice-president: "At Oracle we did things differently. We had one version of the code that was portable everywhere." One source tree, many ports — against IBM, whose SQL/DS (1981) and DB2 on MVS (1983) ran on IBM operating systems and nowhere else. By the time IBM's products were established on the mainframe, Oracle had an installed base off it, on DEC, Sun and HP UNIX systems.

From IPO to Enterprise Standard

Oracle went public on 15 March 1986, offering 2.1 million shares on NASDAQ. It had 450 employees and annual revenue of US$55 million.

The versions that made it the enterprise default came next, and the credit is often misassigned. Oracle Version 6 (1988) brought row-level locking, hot backup — administrators could archive data while users kept working — and PL/SQL. Oracle7 (1992) added stored procedures, triggers and declarative referential integrity, and is the release usually named as the point at which Oracle became the de facto enterprise RDBMS. Oracle 9i (2001) added Real Application Clusters, 12c (2013) multitenant pluggable databases, and 18c/19c (2018–) the Autonomous Database line.

Sun, and the Cloud

In 2010 Oracle acquired Sun Microsystems for US$7.4 billion, taking MySQL, Java, Solaris and SPARC in a single move: the leading open-source relational rival brought inside, and the server, storage and operating system owned outright.

The 2010s were harder — AWS, Google and Microsoft took the cloud market Oracle wanted — but Oracle Cloud Infrastructure and a turn toward AI infrastructure changed the shape of the business, including participation in the Stargate project announced in January 2025. In the fiscal year ended 31 May 2026 Oracle reported revenue of US$67.4 billion. The number that describes the turn is not revenue but backlog: remaining performance obligations of US$638 billion, against US$138 billion a year earlier, which the company attributes to "certain significant cloud contracts that were entered into during the period."

The Lab Publishes, the Startup Ships

Codd's eleven pages appeared in 1970. Nine years later they were a product; the interval is short by the standards of academic ideas reaching a market.

And the party that made the crossing was not the laboratory that produced the paper. IBM published the research openly, and a company three people had started two years earlier read it and shipped first. That pattern — the theory described inside a large firm, the market taken by a venture outside it — recurs through the database industry, from MySQL and PostgreSQL to MongoDB and Snowflake.