As a key area for large model applications in the financial industry, it is moving from pilot exploration to engineering deployment. Financial large models are efficiently empowering customer service, knowledge management, process assistance, and other areas. At the same time, different types of financial institutions are exploring differentiated paths for large models. For example, banks mainly focus on private projects under compliance and regulatory requirements, while leading banks are developing their own models or collaborating with large model vendors to set industry benchmarks; the insurance industry has highly structured business processes, naturally suited for large model capabilities, and tends to focus on the large-scale implementation of standard processes; securities firms are more inclusive of technological innovation on the business side, especially in areas such as research assistance, content summarization, and sentiment analysis; internet finance is actively promoting the deployment of standardized financial large model native products through cloud-native architecture.
Based on research and analysis of financial large models in China for the entire year 2024, Frost & Sullivan (Frost & Sullivan, hereinafter referred to as "Frost & Sullivan") in collaboration with LeadLeo Research Institute released the "China Financial Large Model Market Tracking Report 2024". This report deeply analyzes the market share and competitive landscape of financial large models in China, elaborates on business models, product delivery, and major participants under private deployment and MaaS deployment forms, and also deeply analyzes the deployment characteristics and scenario needs of different financial institutions, as well as makes predictions about the future development trend of large models in the financial industry.
PART.01
In 2024, the market scale of financial large models in China reached 2.866 billion yuan, but due to the increased requirements of traditional financial institutions for SLA (Service Level Agreement) and delivery stability, the market growth rate in the second half of 2024 fell short of expectations
In 2024, the AI demand of financial institutions in traditional scenarios such as customer service, marketing, and compliance continued to rise, and the market as a whole showed a year-on-year growth trend. The market scale did not significantly exceed that of the first half in the second half of the year, reflecting behind the high threshold and technical rigor of AI implementation in the financial industry. On one hand, financial customers have put forward higher standards for model response speed, SLA achievement rate, and delivery quality; on the other hand, some projects originally deployed on public clouds were forced to migrate to local hardware due to the inability to meet financial-grade security and compliance requirements, resulting in extended delivery cycles and slowed project progress.
PART.02
In 2024, the market competition focus of financial large model MaaS shifted from model performance to platform capabilities. Platform cloud vendors are accelerating the construction of "financial AI operating systems" to seize strategic high ground
In 2024, the MaaS market has moved from model competition to platform construction capability competition. Cloud vendors create a model capability scheduling middle platform by integrating multiple open-source/closed-source and self-developed models, unifying SDK encapsulation, model store distribution, etc., to help financial institutions call different models with low thresholds to handle customer service, marketing, risk control, and other scenarios, significantly reducing migration and integration costs. In contrast, third-party inference clouds and single model vendors are limited by their ecosystem construction and standardization capabilities and find it difficult to meet the long-term expectations of financial institutions for "large model middle platforms
PART.03
In 2024, standardized financial models (71%) dominated the market. After 2025, hardware-software integrated machines are expected to open up a new round of growth space in private deployment
In 2024, standardized products became the preferred path for financial institutions to build large model capabilities due to their high delivery efficiency, low deployment threshold, and wide coverage of scenarios. Entering 2025, with high-quality open-source models such as DeepSeek and Tongyi promoting the lightweighting of large models and reducing inference costs, hardware-software integrated machines will further unleash their potential in scenarios with prominent localization and compliance requirements. Their strong replicability, flexible delivery, and rapid iteration are expected to accelerate expansion in private deployment and become an important selection direction for financial institutions under the goals of "cost reduction, security, and closed-loop
PART.04
In 2024, financial large models were most widely implemented in customer service scenarios, and in 2025, the willingness to deploy in office efficiency scenarios (especially code) increased the most significantly
In 2024, customer service scenarios became the most widely covered scenario for large models in financial institutions due to their low implementation cost and obvious response benefits. Entering 2025, office efficiency scenarios (such as code assistants) are listed by most financial institutions as key promotion directions. At the same time, financial institutions are currently deploying multiple models in different scenarios (customer service, risk control, operations, compliance, investment research), facing problems such as data silos, computing power waste, and fragmented inference capabilities. The industry generally has a gradually increasing demand for "unified intelligent middle platforms". In the future, large models with general inference capabilities and industry-specific expertise will become the foundation to support the horizontal integration of multiple businesses and the vertical coordination of tasks.
PART.05
It is expected that 2025 will be the first year for AI Agents to be widely implemented in the financial industry, and the MOA architecture will become a key support form for breaking through technical standardization and collaborative capabilities
The financial industry is taking the lead in implementing AI Agents in scenarios with clear tasks and rules such as customer service and code. The current technical path is also gradually moving from single-point tool calls to multi-agent collaboration. With the maturity of ReAct prompt words, Planner schedulers, and external API call capabilities, Agents already have the potential to complete closed-loop tasks such as analysis, generation, and compliance detection. In this process, the multi-agent collaborative execution mode represented by the MOA (Mixture of Agents) architecture is gradually replacing single models and becoming the mainstream engineering form of financial AI systems. The MOA architecture emphasizes modularity, controllability, and task decoupling, adapting to the high process complexity and strong compliance requirements of the financial industry. Entering 2025, with vendors such as Anthropic launching general protocols such as MCP, AI Agents are accelerating their evolution towards cross-system collaboration, laying the foundation for "multi-agent distributed collaboration + financial AI operating system
In the wave of digital and intelligent transformation of financial institutions, enterprises such as Alibaba Cloud, Baidu Smart Cloud, SenseTime Technology, and Huawei Cloud have dominated the financial large model market with their profound technical accumulation, accurate grasp of the industry, and rich project implementation experience. These leading financial large model enterprises not only deeply cultivate their own technical fields but also, driven by strong technical advantages, precisely meet market demands, injecting strong impetus into the intelligentization and digitalization process of the entire industry.
