DB2 has a long history and is considered by many to have been the first database product to use SQL (also developed by IBM) although Oracle released a commercial SQL database product somewhat earlier than IBM did.

The name DB2 was first given to the Database Management System or DBMS in 1983 when IBM released DB2 on its MVS mainframe platform. Prior to this, a similar product was named SQL/DS on the VM platform. The earlier System 38 platform also contained a relational DBMS. System Relational, or System R, was a research prototype developed in the 1970s. DB2 has its roots back to the beginning of the seventies when E.F. Codd, working for IBM, described the theory of relational databases and in June 1970 published the model for data manipulation. To apply the model Codd needed a relational database language which he named Alpha. At the time IBM didn't believe in the potential of Codd's ideas, leaving the implementation to a group of programmers not under Codd's supervision, who violated several fundamentals of Codd's relational model; the result was Structured English QUEry Language or SEQUEL. When IBM released its first relational database product, they wanted to have a commercial-quality sublanguage as well, so it overhauled SEQUEL and renamed the basically new language(System Query Language) SQL to differentiate it from SEQUEL.

Historically, when Informix acquired Illustra and made their database engine an object-SQL DBMS by introducing their Universal Server, both Oracle and IBM followed suit by changing their database engines to be capable of object-relational extensions. Moreover, in 2001, IBM bought Informix and in the following years incorporated Informix technology into the DB2 product suite. Today, DB2 can technically be considered to be an object-SQL DBMS.

For some years DB2, as a full-function DBMS, was exclusively available on IBM mainframes. Later IBM brought DB2 to other platforms, including OS/2, UNIX and Windows servers, then Linux (including Linux on zSeries) and PDAs. This process was mainly undertaken through the 1990s. The inspiration for the mainframe version of DB2's architecture came in part from IBM DL/I and IBM IMS, both initially hierarchical and then later network (or CODASYL) databases. DB2 is also embedded in the i5/OS operating system for IBM System i (iSeries, formerly the AS/400), and versions are available for z/VSE and z/VM.

An earlier version of the code that would become DB2 LUW (Linux, Unix, Windows) was part of an Extended Edition component of OS/2 called Database Manager. IBM extended the functionality of Database Manager a number of times, including the addition of distributed database functionality that allowed shared access to a database in a remote location on a LAN. Eventually IBM declared that insurmountable complexity existed in the Database Manager code, and took the difficult decision to completely rewrite the software in their Toronto Lab. The new version of Database Manager, called DB2 like its mainframe parent, ran on the OS/2 and RS/6000 platforms, was called DB2/2 and DB2/6000 respectively. Other versions of DB2, with different code bases, followed the same '/' naming convention and became DB2/400 (for the AS/400), DB2/VSE (for the DOS/VSE environment) and DB2/VM (for the VM operating system). IBM lawyers stopped this handy naming convention from being used and decided that all products needed to be called "product FOR platform" (for example, DB2 for OS/390). The next iteration of the mainframe and the server-based products were named DB2 Universal Database (or DB2 UDB), a name that had already been used for the Linux-Unix-Windows version, with the introduction of widespread confusion over which version (mainframe or server) of the DBMS was being referred to. At this point, the mainframe version of DB2 and the server version of DB2 were coded in entirely different languages (PL/S for the mainframe and C++ for the server), but shared very similar functionality and used a common architecture for SQL optimization: the Starburst Optimizer.

Over the years DB2 has both exploited and driven numerous hardware enhancements, particularly on IBM System z with such features as Parallel Sysplex data sharing. In fact, DB2 UDB Version 8 for z/OS now requires a 64-bit system and cannot run on earlier processors, and DB2 for z/OS maintains certain unique software differences in order to serve its sophisticated customers. Although the ultimate expression of software-hardware co-evolution is the IBM mainframe, to some extent that phenomenon occurs on other platforms as well, as IBM's software engineers collaborate with their hardware counterparts.

In the mid-1990s, IBM released a clustered DB2 implementation called DB2 Parallel Edition, which initially ran on AIX. This edition allowed scalability by providing a shared nothing architecture, in which a single large database is partitioned across multiple DB2 servers that communicate over a high-speed interconnect. This DB2 edition was eventually ported to all Linux, UNIX, and Windows (LUW) platforms and was renamed to DB2 Extended Enterprise Edition (EEE). IBM now refers to this product as the Database Partitioning Feature (DPF) and sells it as an add-on to their flagship DB2 Enterprise product.

In mid 2006, IBM announced "Viper," which is the codename for DB2 9 on both distributed platforms and z/OS. DB2 9 for z/OS was announced in early 2007. IBM claims that the new DB2 will be the first relational database to store XML "natively". Other enhancements include OLTP-related improvements for distributed platforms, business intelligence/data warehousing-related improvements for z/OS, more self-tuning and self-managing features, additional 64-bit exploitation (especially for virtual storage on z/OS), stored procedure performance enhancements for z/OS, and continued convergence of the SQL vocabularies between z/OS and distributed platforms.

In October 2007, IBM announced "Viper 2" which is the codename for DB2 9.5 on the distributed platforms. There were three key themes for the release, Simplified Management, Business Critical Reliability and Agile XML development.

In June 2009, IBM announced "Cobra" which is the codename for DB2 9.7 for LUW. DB2 9.7 adds data compression for database indexes, temporary tables, and large objects. DB2 9.7 also supports native XML data in hash partitioning (database partitioning), range partitioning (table partitioning), and multi-dimensional clustering. These native XML features allows users to directly work with XML in data warehouse environments. DB2 9.7 also adds several features that make it easier for Oracle Database users to work with DB2. These include support for the most commonly-used SQL syntax, PL/SQL syntax, scripting syntax, and data types from Oracle Database. DB2 9.7 also enhanced its concurrency model to exhibit behavior that is familiar to users of Oracle Database and Microsoft SQL Server.

In October 2009, IBM introduced its second major release of the calendar year when it announced DB2 pureScale. DB2 pureScale is a database cluster solution for non-mainframe platforms and is ideal for Online Transaction Processing (OLTP) workloads. IBM based the design of DB2 pureScale on the Parallel Sysplex implementation of DB2 data sharing on the mainframe. DB2 pureScale provides a fault-tolerant architecture and shared-disk storage. A DB2 pureScale system can grow to 128 database servers, and provides continuous availability and automatic load balancing.

In February, 2010, IBM announced that DB2 10 for z/OS would enter beta testing with customers the following month.

In 2009, it was announced that DB2 can be an engine in MySQL. This allows users on the system i platform to natively access the DB2 under the i operating system (formerly called os/400), and for users on other platforms to access these files thru the MySQL interface. On the system i and its predecessors the As/400 and the System/38, DB2 is tightly intregrated into the operating system, and comes as part of the operating system. It provides journalling, triggers and many other features.