Milvus was built from the start for a scale most vector databases don't need to worry about: billions of vectors, distributed across a cluster, with the kind of cloud-native architecture that separates compute from storage the way modern data warehouses do. That makes it a common choice for larger organizations running vector search at genuinely massive scale rather than the RAG-prototype scale a lot of newer vector databases are optimized for.
It's part of the LF AI & Data Foundation, giving it neutral open-source governance rather than being controlled entirely by one commercial entity, even though Zilliz, the company behind much of its development, offers a managed cloud version called Zilliz Cloud for teams that don't want to run a distributed system themselves.
Being free, open source, and specifically built for distributed scale, Milvus tends to come up in conversations where Pinecone, Qdrant, or Weaviate feel like they're hitting real scaling limits, rather than as a first choice for a small RAG prototype where the operational overhead of a distributed system isn't worth it yet.