智通财经网·2025-11-26 01:37Core Viewpoint - China Galaxy Securities emphasizes the importance of selecting leading AI application companies under the AI-First paradigm, focusing on five key areas: edge AI, AI creative and content generation tools, enterprise-level AI agents, vertical industry expert AI agents, and AI-native social and companionship applications. The report highlights the expanding supply-demand gap in domestic computing power and continues to recommend leading domestic AI computing power companies [1]. Industry Overview - The computer industry in 2025 shows a trend of "initial rise followed by decline and oscillation," with the industry index up 19.57% year-to-date as of November 18, outperforming the CSI 300 index. The first three quarters of 2025 saw a slight revenue recovery with a 9.13% year-on-year revenue growth and a 38.25% increase in net profit attributable to shareholders. However, the gross margin decreased by 1.19%, while the net margin increased by 0.50%. The accounts receivable turnover rate grew by 8.17%. The current valuation of the computer industry index is above the historical ten-year average, with a PE (TTM) of 87.97 times and a PS (TTM) of 3.62 times [1]. 2026 Industry Outlook - The trend of model parity is accelerating, with the penetration rate of domestic computing power continuing to rise. Domestic large models are increasingly adopting open-source routes, with 9 out of the top 20 large models being domestic, reflecting a nearly 50% share. The average API price for domestic models is 3.88 yuan per million tokens, significantly lower than the 20.46 yuan per million tokens for overseas models. The demand for domestic AI chips is expected to grow rapidly due to strong demand for inference computing power and U.S. restrictions on high-end chips, further widening the supply-demand gap [2]. AI Agent Development - The AI Agent sector is expected to see a structural acceleration in implementation by 2026. Current challenges include a mismatch between technological maturity and commercialization pace, with 90% of AI Agents achieving over 70% accuracy but only 66% meeting autonomy standards. The high cost of enterprise-level agents compared to general large models limits large-scale deployment. However, structural opportunities are emerging, with enterprise-level agents transitioning from "L2 popularization" to "L3 integration," and the implementation rate in financial risk control and customer service quality inspection exceeding 40% [3]. AI-First Paradigm - The shift towards AI-First applications is driven by the need to internalize AI capabilities into product architectures rather than treating them as additional features. This approach will determine the survival of AI application companies in the next competitive round. A three-dimensional verification system is proposed to identify AI-First companies, focusing on qualitative aspects (data loop capabilities, AI R&D investment over 15%, AI technical background of core teams), quantitative health of AI business (over 50% revenue growth, gross margin above 40%, customer repurchase rate over 80%), and barriers to entry (exclusive scenario data, patent reserves, deep industry know-how) [4].