June 2026 Summaries

At the 2026 Databricks Data + AI Summit, the focus shifted from individual announcements to the evolving significance of the data layer as AI systems move into production. While algorithms and compute have seen rapid market repricing, data remains undervalued due to its complexity and challenges in management, such as scattered and stale data, misaligned business semantics, and insufficient real-time capabilities. As AI systems require high-quality, timely, and well-governed data, the data layer is poised to become the next critical area of focus in the AI stack. Databricks is addressing this challenge with innovations like Lakebase and Lakehouse//RT, emphasizing the integration of real-time analytics and AI governance to ensure data quality and operational efficiency. The discussion highlights that the future of AI infrastructure relies on an AI-native data system capable of handling multimodal data, ensuring elasticity, and supporting agents' dynamic interactions, with a strong emphasis on governance, traceability, and auditability. This reflects a broader trend where databases are evolving into foundational systems for AI, requiring new architectures that support continuous, elastic, and agent-driven workloads.

VDBBench has introduced cost-aware benchmarking for vector databases, enhancing its ability to simulate real-world production workloads by incorporating factors such as ingestion, filtering, recall, latency, concurrency, and custom datasets. The latest release adds a cost dimension, allowing teams to evaluate the financial implications of achieving specific QPS targets, data readiness, and system behavior across multiple tenants. The tool was used to benchmark three managed vector database products—Zilliz Cloud, Turbopuffer, and Pinecone—highlighting their performance and cost trade-offs across various scenarios. The VDBBench Cost Leaderboard presents these findings, with open-source availability enabling teams to replicate the benchmarks or adapt them to their specific needs. By integrating cost into the evaluation process, VDBBench helps teams choose the most suitable vector database for their workload, performance goals, and budget, emphasizing the importance of balancing performance with economic considerations in decision-making.

AI data infrastructure should be tailored to the specific stage of a project's development, as misalignment can lead to costly rebuilds and inefficiencies. Initially, during the prototype stage, speed and functionality are prioritized over sophisticated infrastructure. As a product approaches market fit, there is a temptation to utilize multiple specialized databases, but this can lead to complexity and synchronization issues, suggesting a preference for a single, versatile database system. At the growth stage, cost management becomes crucial, requiring a shift to object storage solutions like S3, and employing targeted compute resources to handle specific workloads efficiently. In the enterprise scale stage, trust and structural considerations become paramount, with a need for secure, isolated, and geographically distributed data infrastructure. Successful teams anticipate future needs and make foundational infrastructure decisions that accommodate growth without necessitating disruptive changes, exemplified by the introduction of solutions like the Zilliz Vector Lakebase, which offers a unified semantic data platform designed for scalability and diverse application requirements.

Loon is a new storage engine designed for Milvus and Zilliz Vector Lakebase to address the challenges of managing constantly evolving vector datasets, which traditional storage systems struggle with due to frequent updates, schema changes, and diverse access patterns. Loon's architecture incorporates hybrid file formats, row ID alignment, and a versioned Manifest to efficiently support both analytical scanning and point reads, enabling seamless coordination across multiple systems involved in AI data workflows. This design allows for the physical separation of different column types while maintaining logical coherence, reducing the need for frequent data rewriting. The Manifest serves as a central source of truth, ensuring that multiple systems can read and update the dataset consistently without duplicating data, thereby facilitating efficient online search, offline analysis, and external computation. Loon is integrated into Milvus 3.0 beta and Zilliz Vector Lakebase, providing a versioned, lake-native foundation that enhances the performance and scalability of vector databases.

A Vector Lakebase is a comprehensive, lake-native data architecture designed for AI, integrating the functionality of vector databases with open lake storage, reusable indexes, and a shared semantic layer to enable both online serving and offline discovery without duplicating data across systems. It addresses the evolving needs of AI production teams by providing a unified foundation that supports retrieval, discovery, analytics, governance, and continuous improvement, thereby eliminating the fragmentation typically found in separate systems. Zilliz Vector Lakebase exemplifies this architecture by evolving from a managed vector database into a unified AI data platform capable of handling diverse AI workloads, from real-time serving to interactive discovery and batch processing. By focusing on principles like one data, one index, and one semantic layer, Vector Lakebase ensures data consistency, efficient index management, and meaningful organization of unstructured data. This architecture does not replace existing vector databases but rather enhances them within a broader data foundation, optimizing the lifecycle of unstructured data through continuous serving and discovery.