Notion scaled its vector database by 10x.

Notion scaled its vector search infrastructure to 10x capacity at one-tenth the original cost through several engineering optimizations. Documents are chunked via Apache Spark batch jobs and embeddings are stored in a vector database, with Kafka handling real-time updates. Key improvements include: chunk hashing to skip unchanged content on edits (reducing unnecessary updates), splitting chunks into metadata and textual content to avoid full re-embedding on permission changes (70% data volume reduction), tuning Apache Spark for tens of thousands of workspaces daily, moving to a serverless architecture (50% cost reduction), and migrating to Ray for unified CPU/GPU processing on the same node (anticipated 90% embedding cost reduction).