Tianfeng: High-quality landing scenarios for computer AI applications, securities are expected to become the core benefiting scenes.

Tianfeng released a research report stating that DS leads the open source ecosystem of large models. With the improvement of product capabilities in various vertical industries, the value differentiation mainly depends on three major factors: user stickiness, data moat, and potential for conversion to paid services. Based on the above conditions, the firm believes that AI securities are likely to become a core beneficiary of AI empowerment in computer application software. They recommend focusing on companies such as Hithink RoyalFlush Information Network (300033.SZ), Beijing Compass Technology Development (300803.SZ), Shenzhen Fortune Trend Technology (688318.SH), as well as securities asset management IT service providers such as Hundsun Technologies Inc.(600570.SH), Fujian Apex Software(603383.SH), and Shenzhen Kingdom Sci-Tech(600446.SH). Key points from Tianfeng include: 1) Strong user stickiness: Traffic advantages build a natural commercial foundation According to Analysys Qianfan data, in January 2025, Hithink RoyalFlush Information Network had 34.18 million MAU, East Money Information had 16.79 million MAU, and Shanghai DZH Limited had 11.53 million MAU, with active users increasing by 14.23%/10.01%/24.07% respectively. A high stickiness user base establishes a scarce foundation for AI service payment conversion. 2) Vertical data barriers: Full-chain data closed loop from B to C Stock trading software as the core entry point for stock traders is expected to form a "data-product-user" positive flywheel with large models. Taking Hithink RoyalFlush Information Network as an example, the firm believes that its iFind (institutional end) and Wencai (C end) are expected to form a vertical flywheel of "data-product-user". iFind has time series data of over 3,000 indicators, tracking daily information from over 2,000 listed companies, 400 industry information websites, and 50 financial media. Wencai interacts through natural language and converts professional data into research conclusions that retail investors can understand (e.g. "companies with ROE>15% in the past three years with R&D expense growth>30%"). 3) High conversion in paid scenarios: Commercial space driven by rigid demand can be quantified The firm believes that retail investor decision-making naturally fits the scenario of paid tools, and financial large models significantly improve product capabilities after deep thinking. On February 7th, Hithink RoyalFlush Information Network updated the deep thinking function on Wencai, showing capabilities similar to DS, greatly improving user experience. On February 14th, East Money Information introduced the "AI Researcher" feature and the "Deep Search/Multi-source Analysis" question and answer mode, allowing the large model to achieve deep thinking based on a complete financial database. Based on data as of February 23, 2025, the advanced and professional versions of Wencai from Hithink RoyalFlush Information Network had a total of 253,400 reservations (16,260 reservations on February 5th, an increase of about 90,000 in 18 days). Assuming that the number of new Wencai users increases at this rate each month, with an additional 150,000 users per month, the total number of Wencai users is expected to reach 1.82 million annually (about 5.3% of MAU). Assuming that the proportion of paid users among them is 30% (conservative)/60% (neutral)/90% (optimistic), this corresponds to incremental income of 857 million/1.714 billion/2.571 billion (assuming the usage of the advanced and professional versions is similar, with an ARPU of 1572 yuan per year). Risk warning: AI product development progress is lower than expected, market sentiment is lower than expected, subjective risk in calculations, etc.