Zilliz Launches Vector Lakebase, Extending the World's Most Adopted Vector Database into a Unified Data Platform for AI

Zilliz, a prominent AI data infrastructure company and the creator of the popular open-source Milvus vector database, has announced the public preview of Zilliz Vector Lakebase. This new offering represents a major enhancement to Zilliz Cloud, transforming it into a unified data platform specifically tailored for artificial intelligence applications. The core innovation of Vector Lakebase lies in its ability to pair the high-performance vector database with a shared, lake-native data foundation, addressing the growing complexity of AI data management. Vector Lakebase introduces a comprehensive tiered storage system that strategically incorporates object storage alongside memory and NVMe. This multi-tier approach is designed to optimize both performance and cost-efficiency across a spectrum of AI workloads. For instance, high-performance tasks demanding single-digit millisecond latency utilize in-memory storage, while capacity-optimized operations leverage memory and NVMe. Critically, object storage forms the foundation for tiered storage, catering to less frequently accessed data at significantly reduced costs, making it ideal for large-scale data lakes. This intelligent tiering ensures that data is stored on the most appropriate medium based on access patterns and performance requirements. The platform's architecture allows for a single logical copy of data to serve multiple purposes, including production queries, interactive discovery sessions, and multi-petabyte training data pipelines. This eliminates the need for data duplication, migration, or maintaining parallel data stacks, which traditionally add complexity and expense to AI development. Zilliz CEO Charles Xie emphasized that Vector Lakebase is the next evolutionary step, providing a unified data foundation where vectors can seamlessly support various AI workflows. Furthermore, Vector Lakebase introduces an "On-Demand Search" capability, offering a pay-as-you-go model that bills directly for object storage and compute resources, rather than relying on potentially higher serverless markups. Zilliz benchmarks suggest this can lead to substantial cost savings, illustrating a 15-fold reduction compared to comparable serverless paths for specific workloads. The platform also supports external data lake search with a zero-copy mode, enabling state-of-the-art indexing and full-spectrum search directly on existing data formats like Lance, Iceberg, Parquet, and Vortex. This integration enhances the governability and analyzability of unstructured data, turning diverse data types into strategic assets for production AI.