经济观察报·2025-07-22 12:38Core Viewpoint - The current development of financial IT can be summarized by two key terms: "domestication" and "intelligence," which are driving the industry into a full-chain upgrade phase [3][6]. Group 1: Trends in Financial IT Development - The demand for digital transformation among financial institutions has been increasing, extending from banks, securities, and insurance to consumer finance and leasing, with clear requirements for overall progress [5]. - The trend has led to an explosive growth of related projects by 2025, with a significant increase in the number of projects involving domestic database selection, cloud platform selection, hardware procurement, and new core system ISV collaborations among brokerages and insurance institutions [6]. - The number of domestic database vendors has decreased by over 60 in the past year, with financial institutions preferring mature products from leading vendors that have undergone extensive business verification and stability testing [7]. Group 2: AI and Intelligent Applications - The emergence of DeepSeek marks a significant turning point for the application of AI large models in the financial sector, enabling a broader range of institutions to develop applications at a low cost [8][10]. - AI applications are evolving through four stages: from "able to chat" to "able to work," then to "self-planning," and finally to "multi-Agent collaboration" [8]. - Current AI applications in financial institutions include code assistants that enhance development efficiency and knowledge base applications, while initial applications in trading and risk control are still in the early planning stages [10][11]. Group 3: Strategic Implementation and Challenges - Financial institutions are actively investing in AI, with many executives placing it at a strategic level, focusing on "phased usable scenarios" during implementation [13]. - A collaborative system of "small models + large models" is recommended, where institutions first establish usable datasets and then train small models for specific tasks while large models handle content generation [14]. - The implementation of intelligent architecture requires modular planning and stepwise advancement, with clear timelines, responsibilities, and acceptance criteria for each phase [14][15].