阿里云李飞飞：AI原生是数据库演进必然方向，大模型Token调用量激增百倍

Core Viewpoint - The concept of "AI Native" is evolving, and the focus should be on "AI Ready" as a more practical definition for the current stage of development in databases [1][4][6]. Group 1: Definition and Standards - The definition of "AI Native" is still under development, and there is no unified standard in the database industry [1][5]. - Two key standards for measuring the maturity of AI Native databases are proposed: at least 50% of instances should be used by Agents, and at least 50% of outputs should be measured in Bytes, specifically Tokens [5][6]. Group 2: Current Positioning - Alibaba Cloud's PolarDB is positioned as "AI Ready" rather than "AI Native," emphasizing a gradual upgrade in technology architecture to lay the foundation for becoming AI Native [6][7]. - The company acknowledges the rapid evolution of AI and advocates for a measured approach to adopting AI technologies, avoiding radical overhauls [6][7]. Group 3: Future User Dynamics - Future database users are expected to be primarily Agents rather than programmers, with predictions that 80% to 90% of new databases will be created autonomously by Agents [7]. - The interaction between users and databases will shift towards natural language for users, while Agents will continue to use command lines and scripts [7]. Group 4: Data Management and Cost Considerations - The concept of "hot," "warm," and "cold" data is emphasized, with hot data being critical for real-time transactions that large models cannot process [8]. - Recent increases in memory costs (30% to 40%) are anticipated to continue for 3-5 years due to unprecedented demand driven by AI [8][9]. Group 5: Architectural Changes - PolarDB is adopting a serverless model that allows for minimal computational resources until requests are made, reflecting a shift in database design logic [7][9]. - The architecture is evolving to support cost reduction through techniques like data tiering and resource pooling [9]. Group 6: Future Vision - The future vision includes databases acting as the central hub for actions, with Agents directly issuing commands to databases, transforming them from mere storage to decision-making entities [10]. - The company believes that a steady and pragmatic approach to becoming AI Native is essential for long-term competitiveness in the industry [10].