GF Securities: AI Agent Inference Costs Continue to Drop Globally, AI Applications Poised for New Catalytic Phase

GF Securities released a research report stating that with the recent intensive launch of AI Agents both domestically and internationally, coupled with continuously declining inference costs, AI applications are expected to enter a new phase of catalysis. Opportunities are seen from both industrial logic and catalyst mapping perspectives, and the Agent industry trend will directly drive demand growth in cybersecurity, IT operations, and data governance. For the first half of 2026, domestic AI models are expected to accelerate their catch-up efforts, with short-term optimism focused on the iterative updates of domestic models like DeepSeek. Additionally, the recent earnings season for US stock giants has reinforced the global AI growth narrative. It is recommended to monitor overseas cloud giant Google (GOOGL.US); the era of large models favors leading players with concentrated resources, suggesting attention on internet leaders such as Alibaba-W (09988). The main views of GF Securities are as follows:

Recently, MoltBot (formerly ClawdBot) has gained phenomenal popularity in overseas communities. As an AI Agent automation hub for developers, MoltBot's features—including 24/7 background operation, local long-term memory, direct invocation via chat software, and integration across multiple terminals—have rapidly increased its visibility among overseas developers. As of January 27, 2026, MoltBot's GitHub page had received 62.4k stars and 740 forks. MoltBot demonstrates high flexibility and scalability but currently has several drawbacks: (1) It primarily targets professional developers, presenting a high deployment barrier and significant learning costs for non-technical personnel; (2) Development and optimization of the Agent itself are insufficient, relying heavily on model performance and third-party tool APIs; (3) MoltBot has numerous network security risks, including data leaks and exposure of sensitive information.

Claude Cowork: An Office Agent Integrated into Local Workflows. The primary function of Claude Cowork is to directly host desktop operations and perform system-level file management. Users only need to specify the target folder and the final task objective, and the product can autonomously break down complex requirements, plan execution paths, and execute step-by-step. The Claude for Excel plugin can call local Excel files to analyze data, create visual charts, and perform complex modeling.

Domestic Agent products and functionalities are rapidly catching up. (1) MiniMax Agent 2.0 has become a productivity tool deeply integrated into user workflows. Its context is derived from the user-authorized local work environment, allowing it to autonomously read folder contents, common file formats, local tool configurations, and material structures without requiring repeated user input. By authorized access to the user's work environment, it can autonomously understand file content, web information, and workflows. (2) Coze 2.0 introduces four main capabilities: Skill, long-term planning, programming, and office tasks.

Investment Recommendations: With the recent intensive launch of AI Agents globally and continuously declining inference costs, AI applications are poised for a new catalytic phase, offering opportunities from both industrial logic and catalyst perspectives. The Agent industry trend will directly drive demand growth in cybersecurity, IT operations, and data governance. For US stocks, it is recommended to focus on: (1) Cybersecurity: Cloudflare (NET.US), CrowdStrike (CRWD.US), Palo Alto Networks (PANW.US); (2) IT Operations: Datadog (DDOG.US); (3) Data Governance: Snowflake (SNOW.US), MongoDB (MDB.US). For the first half of 2026, domestic AI models are expected to accelerate their catch-up; short-term optimism is placed on iterative updates of domestic models like DeepSeek. The recent earnings season for US giants reinforces the global AI growth narrative, suggesting attention on overseas cloud giants Google (GOOGL.US) and Amazon (AMZN.US). The large model era benefits leading players with concentrated resources; it is recommended to monitor internet leaders Alibaba-W (09988) and Tencent Holdings (00700). For leading players in specific application scenarios, consider Kuaishou-W (01024), Meitu (01357), Shunwang Technology (300113.SZ), and Fenbi (02469).

For the IP + AI video industry chain, focus on Yuewen Group (00772), Chinese Online (300364.SZ), Shanghai Film (601595.SH), Alpha Group (002292.SZ), Huace Film & TV (300133.SZ), H&R Century (000892.SZ), and Zhangyue Technology (603533.SH). For AI content rights confirmation, monitor Vobile Group (03738). For AI marketing, consider Mobvista (01860), Yeahka (301171.SZ), BlueFocus (300058.SZ), and Tianxiasiu (600556.SH). For the AI e-commerce direction, focus on SMZDM (300785.SZ). For AI customer service agents, monitor Ultrapower (300002.SZ). For AI gaming, watch Kingnet Network (002517.SZ). For AI healthcare, follow JD Health (06618) and AliHealth (00241).

Risk warnings include model iteration effects falling short of expectations, slower-than-expected commercialization and application deployment, and risks related to copyright, ethics, and content quality.