Read more at:
I would say that, wouldn’t I? I mean, I work for Oracle, and Oracle AI Database 26ai can store vector embeddings alongside business data, and it supports HNSW and IVF vector indexes. But it’s not just Oracle. Literally every database that developers have used for years has vector support now. Microsoft has added a native VECTOR data type to SQL Server 2025, along with vector search and vector indexes. MongoDB has pushed automated embeddings into Atlas Vector Search, with embeddings generated in the database and synchronized as data changes. Postgres, through pgvector, also offers vector support. Etc., etc., etc.
That doesn’t mean Pinecone, Weaviate, Milvus, or the other purpose-built vector vendors are doomed, but it does call into question the premise behind their VC pitch decks. For most enterprise applications, vector support is a feature, one that should be tightly woven into an existing data estate.
This matters because the hardest part of production AI isn’t nearest-neighbor search: It’s context.
Proliferating data siloes
I’m not suggesting that vector search isn’t a thing: It’s very important. If you’re building retrieval-augmented generation (RAG), recommendation systems, personalization, agent memory, or anything that requires matching meaning rather than keywords, you need some way to compare vectors efficiently. And, credit where it’s due, the purpose-built vector vendors made that obvious before the incumbents did. I was working at MongoDB when Pinecone, Weaviate, Milvus, Qdrant, and others helped establish the patterns that everyone now treats as obvious. That’s real innovation.
