HuaLong Securities: Agent commercialization accelerating, application scenarios expected to blossom in multiple ways.

Huolong Securities released a research report stating that AI Agent or Relay Chat bot will become the mainstream form of AI products in the next stage. From the perspective of the ultimate form of AI, the development of AGI requires AI to be able to participate in decision-making and execute operations based on recognition, understanding, and reasoning. Agent is driving the transformation of enterprise budgets from "buying tools" to "buying results". Global computing Capex continues to rise, basic models accelerate iteration, and create favorable conditions for the prosperity of the Agent ecosystem. The computer industry is maintaining a "recommended" rating. Huolong Securities main points are as follows: Transition from process delivery to result delivery, potential increase in enterprise willingness to pay The brokerage believes that from Chat bot to Agent, there has been about three generations of evolution in the form of AI products. In the process of evolution, the interaction between users and AI becomes deeper, and the delivery of task results becomes more complete. Currently, Agent is a better carrier for transitioning from "process delivery" to "result delivery". Looking ahead, the evolution of AI products will further enhance the attribute of "productivity" rather than purely as a "tool". For enterprises, labor operating expenses are being replaced by GPU capital expenditures. Specifically, for B-end customers, the demand for AI applications to help companies reduce costs and increase efficiency is clear. AI applications can help companies increase productivity more than the cost of investment, i.e. ROI> 1, which will stimulate the willingness of enterprise customers to pay. Additionally, since Agent can assist humans in generating economic benefits, enterprise expenditures on AI will shift from Capex to Opex. Capex, or capital expenditures, refers to large investments used to acquire or upgrade long-term assets (such as equipment, factories, technology) and gradually convert them into expenses through depreciation/amortization. Opex, or operating expenses, refers to the recurring costs of maintaining day-to-day operations of the enterprise (such as wages, rent, utilities). As Agent can partially replace labor functions and bring in revenue, the funds invested in Agent are equivalent to paying for labor, so this part of the funds will shift from pure Capex to Opex, further enhancing enterprise willingness to pay. AI Infra construction is rapidly developing, creating prerequisites for the prosperity of the Agent ecosystem Under AI-driven strategies, overseas cloud giants such as Microsoft, Google, Amazon, Meta, and other North American cloud giants have continuously increased capital expenditures in recent years, intensifying their investments in AI and cloud infrastructure. At the beginning of 2025, Alibaba stated in its performance conference that it plans to invest in AI and cloud computing infrastructure in the next three years, exceeding the sum of the past ten years. Based on Alibaba's increase in capital expenditures, it is expected that in the short term, the capital expenditures of domestic major factories will continue to rise. On the other hand, domestic large model architectures continue to be optimized, with significant improvements in inference efficiency. For example, DeepSeek-R1 and Step-3 are continuously innovating in the underlying model, and model performance competition remains fierce, laying a solid foundation for Agent. Specialized Agents are in full bloom, with multi-Agent collaboration becoming a trend Multi-Agent Collaboration refers to multiple autonomous intelligent agents collaborating through communication, coordination, and task division to jointly accomplish complex goals that a single Agent cannot efficiently or independently achieve. Its core characteristics are "decentralization, interactivity, and complementarity", not relying on a single supermodel, but emerging through collective intelligence that surpasses individual capabilities. From a technical perspective, the transition from models to multi-Agent scenarios is expected to form a closed loop. From an ecosystem perspective, the path of AI applications may evolve from single-point tools to Agent platforms, ultimately forming vertical industry solutions based on multiple intelligent agents. In terms of current business models, end-point Agents embed with consumers predominantly rely on one-time hardware revenue. Non-endpoint Agents typically adopt: (1) subscription models (SaaS models). Users pay to use the AIAgent functionality on a periodic (monthly/yearly) basis, suitable for standardized demand scenarios (such as intelligent customer service, data analysis). Enterprises subscribing to this service can receive stable service and continuous updates. (2) Pay-per-call, that is, billing based on the number of calls or task volumes. This is suitable for scenarios with large fluctuations in demand (such as cloud computing, big data analysis). Enterprises pay based on the actual number of API calls or task complexity, reducing upfront investment risks. (3) Customized services charged in specialized scenarios (pay per intelligent agent). For example, provide customized AIAgent for specific needs in industries such as finance, healthcare, logistics, and charge development and deployment fees. Currently, SaaS and pay-per-call models are the mainstream choices, and the model where the first party pays only for "actual savings in manpower/increased revenues" will accelerate the commercialization of the Agent. Risk warnings: (1) Increased market competition. (2) Risks of errors in the quoted data. (3) Slower than expected implementation of AI applications. (4) Slower than expected pace of technological iterations. (5) Key focus on company performance not meeting expectations. (6) Slower than expected speed of policy standards.