When databases are no longer just passive data warehouses, but intelligent cores that can actively understand, reason, and participate in AI decision-making, the paradigm of data infrastructure is being completely redefined. At the 2026 Alibaba Cloud PolarDB Developer Conference, Alibaba Cloud officially launched the AI Data Lakehouse (Lakebase) and a series of new capabilities for the Agent era, marking the global launch of the first "AI-ready database" system. It not only unifies the management of structured, semi-structured, and unstructured multi-modal data, but also directly performs semantic retrieval, model inference, and intelligent decision-making within the database, truly realizing "data as intelligence."
The released Lakebase represents a core breakthrough for PolarDB. It breaks down the barriers between traditional data lakes and databases, building an "integrated lake and warehouse" architecture, and for the first time achieves efficient access and consistent management of multi-modal data (text, images, logs, vectors, etc.) under a unified logic. With an innovative caching acceleration mechanism, Lakebase dynamically optimizes IO and bandwidth for different scenarios, ensuring high-speed data flow for AI training and inference.
More importantly, PolarDB deeply embeds AI capabilities into the database kernel. Through the Model-as-Operator technology, developers can directly call AI models within SQL to perform tasks such as semantic search, sentiment analysis, or anomaly detection without exporting data to external systems—significantly improving efficiency while ensuring the privacy compliance requirement of "data staying within the domain." The system also integrates KVCache, graph database, and vector retrieval technologies to build an intelligent retrieval solution that balances long-term and short-term memory with low computing power consumption, providing continuous context support for Agent applications.
Alibaba Cloud further defines the four pillars of the "AI-ready database":
1. Multi-modal AI Data Lakehouse: Unified management of all types of data, supporting efficient caching across media;
2. Efficient integrated search capabilities: Seamlessly integrating vector retrieval and full-text search within SQL, achieving dual matching of keywords and semantics;
3. Model operator service: In-database inference + Agent-Ready architecture, making the database the "memory and brain" of intelligent agents;
4. Backend services for Agent applications: Rapidly supporting vertical industry intelligent agent development through Serverless and multi-tenancy packaging.
Currently, PolarDB has deployed over 3 million computing cores, covering 86 availability zones worldwide, and has been scaled in key fields such as finance, automotive, and government. A major commercial bank uses it to build a real-time risk control system, while Li Auto and XPeng use it for autonomous driving data loops. MiniMax and Mihoyo also use it as the underlying engine for large model training and inference.
