Scaling pgvector: From Thousands to Millions of Vectors

Scaling strategies: Small Scale (<100K) — single table with HNSW index. Medium Scale (100K-10M) — tune HNSW parameters, use connection pooling, consider partitioning. Large Scale (10M+) — horizontal partitioning, IVFFlat indexes, read replicas, caching layer. General tips: Bulk load before creating indexes, use VACUUM ANALYZE after large changes.