Insights into the Commercial Application Scenarios of AI Agents in China 2025

The Evolution and Development Background of AI Technology

According to the analysis of Gartner’s Emerging Technology Hype Cycle, generative AI entered its explosive growth phase at the end of 2022, followed by rapid development. By the end of 2024, thanks to DeepSeek’s open-source model, low-cost localized large model solutions are being widely implemented in commercial applications, significantly reducing the deployment difficulty and privacy risks of AI agents. This allows most enterprises to access affordable and usable local models.

At the same time, generative AI can achieve multimodal generation of long texts, long codes, images, and videos. AI agents have begun pilot projects and are gradually moving towards commercialization.

Definition of AI Agent Concept

An AI agent is an intelligent system capable of perceiving its environment, making autonomous decisions, and executing tasks to achieve specific goals, commonly referred to as “AI agents” in China. It can understand and perform corresponding operations based on user needs to complete complex tasks. Compared to traditional large models, AI agents have the ability to call various large models, APIs, and plugins, integrating visual and auditory recognition functions, and possess multimodal characteristics.

Currently, mainstream models like Chat-GPT, Kimi, and DeepSeek already possess some agent capabilities and are also the main models called by agents.

Localized Deployment of Large Models Benefits AI Agents, Expanding the Market Scale of AI Agents

The market scale of AI agents continues to expand, expected to grow from 57.4 billion yuan in 2023 to 3.3009 trillion yuan by 2028. Thanks to DeepSeek’s open-source model, low-cost localized large model solutions are being widely implemented in commercial applications, while AI agents combining large models and RPA are being piloted.

Benefiting from open-source models and localized large model solutions, the deployment difficulty and privacy risks of agents have significantly decreased. Currently, 23% of enterprises have confirmed localized deployment, with the market scale reaching 64 billion yuan, and the proportion is expected to grow to 90% by 2028. With the performance improvements brought by the iteration of GPUs and AI processing units, the cost per token will continue to decline.

It is expected that by 2025, the deployment scale of large model integrated machines in state-owned enterprises and government departments in China could reach 123.6 billion yuan, with considerable future industry penetration. The localized deployment of large model integrated machines presents a favorable relationship for the development of AI agents, where data security and compliance become core driving forces. Technological collaboration will lower the deployment and operation maintenance thresholds, enhance performance optimization and scenario adaptation capabilities, and ultimately reduce costs to promote large-scale application implementation.