Vectorization in mainstream databases: Oracle’s push and the design choice dilemma for new systems

Oracle is vectorizing its customers' data, a bold push to bake vector features into mainstream databases. It’s fueling a hot debate: should vector work live inside the DB or sit in a dedicated vector store? ^[1]

From the discourse in Build Your Own Database, the core question isn’t merely storage but retrieval. The post asks, "How do we store data persistently and then efficiently look it up later?"—and argues that persistence and lookup may be two problems. ^[2]

This setup tees up a design-choice dilemma: embed vector features in a database, with potential consistency and tooling benefits, or lean on specialized vector tooling outside the DB. The post even riffs on the fantasy of a 'write-only database' that’s lightning fast. ^[2]

• Inside-DB vector features — the discussion frames the tradeoffs as part of the broader data-stack design inquiry. ^[2] • External vector tooling — the argument is that dedicated vector stores can optimize embeddings and similarity search without bloating the core DB. ^[2]

Bottom line: Oracle's move intensifies a long-running debate about where vectorization belongs in the data stack. Expect more architectures to test the inside-DB vs external-vector-store split. The conversation highlights real-world implications for performance, tooling, and how fast new features land. ^[1][2]

References