I’d say that, wouldn’t I? I imply, I work for Oracle, and Oracle AI Database 26ai can retailer vector embeddings alongside enterprise information, and it helps HNSW and IVF vector indexes. Nevertheless it’s not simply Oracle. Actually each database that builders have used for years has vector assist now. Microsoft has added a local VECTOR information sort to SQL Server 2025, together with vector search and vector indexes. MongoDB has pushed automated embeddings into Atlas Vector Search, with embeddings generated within the database and synchronized as information modifications. Postgres, by way of pgvector, additionally affords vector assist. And many others., and so on., and so on.
That doesn’t imply Pinecone, Weaviate, Milvus, or the opposite purpose-built vector distributors are doomed, however it does name into query the premise behind their VC pitch decks. For many enterprise functions, vector assist is a characteristic, one which ought to be tightly woven into an present information property.
This issues as a result of the toughest a part of manufacturing AI isn’t nearest-neighbor search: It’s context.
Proliferating information siloes
I’m not suggesting that vector search isn’t a factor: It’s crucial. When you’re constructing retrieval-augmented technology (RAG), advice techniques, personalization, agent reminiscence, or something that requires matching which means reasonably than key phrases, you want some approach to evaluate vectors effectively. And, credit score the place it’s due, the purpose-built vector distributors made that apparent earlier than the incumbents did. I used to be working at MongoDB when Pinecone, Weaviate, Milvus, Qdrant, and others helped set up the patterns that everybody now treats as apparent. That’s actual innovation.
