Frost & Sullivan in collaboration with LeadLeo releases the '2024 China Industry Large Model Market Report'

Before the advent of large AI models, the development of AI models mainly followed a traditional path of 'customization and scenario binding'. This meant that for each specific application scenario, a small model had to be built and trained separately. This development model has significant drawbacks: model resources are difficult to reuse and accumulate effectively, resulting in a high threshold for practical applications of AI technology. Additionally, since each model has to be built from scratch, it is not only costly but also inefficient to implement.

However, the emergence of general-purpose large models has completely changed this situation. It has successfully built a model-based platform with broad applicability and outstanding generalization capabilities. On this platform, vertical industries can quickly build industry-specific large models by fine-tuning or customization to meet specific industry needs. This transformation has brought multiple positive impacts: firstly, it significantly reduces the computing power and data volume requirements of vertical models during training, making model development more economical and efficient; secondly, it greatly shortens the model development cycle and accelerates the application and implementation of AI technology in vertical fields; finally, it strongly promotes application innovation and development efficiency improvement in corresponding vertical fields, laying a solid foundation for the widespread application of AI technology.

In 2023, China's large model market witnessed significant growth, reaching a market size of 105 billion yuan. This growth was mainly driven by the strong demand for intelligent transformation across various industries. With the continuous advancement of artificial intelligence technology and its increasingly widespread application in all sectors, enterprises' demand for large model products and services that can improve efficiency, optimize decision-making, and enhance user experience is also increasing. It is expected that by 2024, the scale of China's industry large model market will further expand to 165 billion yuan, a growth rate of 57% compared to 2023. This rapid growth reflects the market's high recognition and broad application prospects for large model technology. In multiple fields such as finance, manufacturing, and healthcare, large models are gradually becoming a key force driving industrial upgrading and innovation.

Looking ahead, it is expected that by 2028, the market scale of industry large models in China will reach 624 billion yuan. This not only reflects the important role that large model technology plays in the transformation and upgrading of the Chinese economy but also indicates that the field will continue to maintain a high-speed development trend over the next few years. With continuous maturation and improvement of technology, as well as support from the policy environment, the application scenarios of large models will become more diverse and rich, providing strong technical support for the digital transformation of different industries.

Prompt Engineering, Retrieval Augmented Generation (RAG), and Fine-tuning are the three core approaches to enabling large language models (LLMs) to be efficiently deployed in diverse application scenarios.These technologies each have their own characteristics and work together to improve the performance and adaptability of the model, ensuring that LLM can better serve various practical needs.

Prompt Engineering: By inputting text (i.e., prompts), the pre-trained model is guided to generate outputs that better meet industry requirements. Advantages include task normalization, simplification of diverse task processing procedures, and the ability to flexibly adapt to a wide range of needs. However, finding efficient prompts often relies on repeated experimentation and is not always stable in results.

Retrieval-enhanced Content GenerationThis technology combines the advantages of large language models and external knowledge bases, enhancing the model's generation capabilities by retrieving relevant information from the knowledge base. When generating text, the RAG model can not only rely on its own knowledge reserves but also access and integrate external knowledge in real-time, thereby generating richer, more accurate content with context coherence.

fine-tuningFine-tuning is the process of further training a pre-trained large model for specific tasks. By continuing to train the model on small-scale task-related datasets, it can become more adaptable to the task requirements of specific domains, such as sentiment analysis, named entity recognition, etc. The key to fine-tuning lies in selecting appropriate training data, adjusting hyperparameters such as learning rate, and ensuring that the model fully absorbs knowledge from specific tasks while maintaining its generalization ability. Fine-tuned models often achieve higher accuracy on specific tasks while maintaining good generalization performance.

At the application end, understanding and integrating domain-specific expertise is key to driving the application of large models. Models need to be able to accurately reflect and solve complex problems in industries, which requires teams to have in-depth industry understanding and practical experience. Cost-effectiveness analysis is crucial for determining the balance between model investment and actual business benefits. High development and deployment costs may require long-term investment, while the efficiency and returns of models need to be clearly quantified and proven. The development, deployment, and maintenance costs of large models are often high. This includes expenditures on hardware equipment, data management, human resources, and security measures, which require effective cost management and control at the initial stages of projects and during long-term operations. Recruiting and training AI professionals with deep learning, data science, and industry backgrounds is a key challenge. These talents not only need technical capabilities but also understand the data characteristics and challenges of specific industries and can effectively apply models to solve problems in practice. The accuracy and adaptability of models directly affect their application effectiveness in actual business operations. Industry-specific data and requirements have a significant impact on model predictions and recommendations, so ensuring that models can provide reliable and operational results in specific industry environments is an important challenge.

At the technical level, training and running large-scale models require enormous computational resources. Cloud service providers typically charge based on the computing resources used, and long-term and large-scale usage can lead to significant economic burdens that necessitate effective resource planning and optimization strategies. Optimizing large models involves reducing complexity, improving computing efficiency, and optimizing prediction speed. Ensuring that the optimization algorithm has sufficient response speed and real-time performance in practical applications is a key technical challenge. In addition, ensuring data quality is also crucial; the quality of data directly affects the model's performance and predictive capabilities. Industry data may be diverse, incomplete, or of poor quality, which need to be addressed through effective data cleaning, preprocessing, and validation steps to ensure the reliability and accuracy of model training.

The development trend of the implementation of industry large models in China is demonstrating multi-dimensional and profound changes and innovations. Firstly, the increase in model scale and complexity has become an inevitable trend. With continuous technological progress, the scale and complexity of large models will continue to expand to meet increasingly complex and diverse business needs. Secondly, interpretability and fairness have become important considerations in model development. To enhance model reliability and fairness, the future trend will be to improve model interpretability, making the behavior and decision-making process of models more transparent and understandable. At the same time, fairness will also become an important principle in model design and application, ensuring that models remain fair and unbiased when handling different groups and scenarios. Thirdly, deployment strategies and efficiency optimization have become the focus of industry attention. To reduce model resource consumption and improve response speed, the industry will continuously explore and optimize model deployment strategies, such as adopting distributed computing, edge computing and other technical means. At the same time, continuous optimization of model operation efficiency will also become an important direction for industry development, aiming to enhance the real-time performance and stability of models. Fourthly, adaptation to specific fields and customization have become trends. Developing more targeted models and solutions according to the characteristics of different industries and application scenarios will become an important direction for industry development. This customization is reflected not only in the choice of model structure and algorithm but also in various aspects such as data preprocessing, feature engineering, model evaluation, aiming to improve the performance and applicability of models in different scenarios. Fifthly, the rise of self-supervised learning provides new ideas for model training. Using unlabeled data for self-learning and reducing dependence on large amounts of labeled data have become important ways to improve model learning ability and generalization ability. Finally, the integration ability of multimodal data has become an important trend in model development. With technological progress and the continuous expansion of application scenarios, single-modal data can no longer meet the needs of complex tasks. Therefore, integrating multiple types of data sources, such as images, speech, text, etc., to support more diverse application scenarios has become an important direction for model development.

In the wave of digital transformation, enterprises such as Huawei, Alibaba, Baidu, and SenseTime have dominated the industry's large model market with their profound technical accumulation, precise grasp of the industry, and rich experience. Leading large model companies not only delve deeply into their respective technical fields but also drive precise market demand with their strong technological advantages, injecting powerful momentum into the intelligentization and digitalization process of the entire industry.

• financial industryThrough in-depth research on large models of the Chinese financial industry, Frost & Sullivan and LeadLeo recommend focusing on Huawei Cloud, Alibaba Cloud, SenseTime Technologies, and Baidu Intelligent Cloud. These companies excel in technological innovation, market share, customer service, and industry solutions.

   

• government affairs industryAfter in-depth research on the large models for China's government affairs industry, Frost & Sullivan and LeadLeo recommend focusing on Huawei Cloud, Inspur Cloud, Baidu Smart Cloud, and Alibaba Cloud. These enterprises have become key forces driving industry development through their outstanding performance in technical strength, market layout, and solutions for government affairs.

   

• Telecommunications industryAfter in-depth research on large models in the telecommunications industry, Frost & Sullivan and LeadLeo recommend focusing on Tianyi Cloud, Baidu Smart Cloud, China Unicom, and Zhipu AI. These enterprises have demonstrated outstanding performance in the telecommunications field with their technical strength, industry focus, and innovation capabilities, and are the core driving force for digital upgrading of the industry.

   

• education industryAfter in-depth research on large models in the education industry, Frost & Sullivan and LeadLeo recommend focusing on Inspur Cloud, iFlytek, Baidu Intelligent Cloud, and Huawei Cloud. These enterprises have outstanding performance in technological innovation, educational scenario applications, and intelligent solutions, providing strong support for promoting the digital and intelligent transformation of the education industry.

   

• Industrial sectorAfter in-depth research on large industrial industry models, Frost & Sullivan and LeadLeo recommend focusing on Huawei Cloud, Baidu Intelligent Cloud, Alibaba Cloud, and iFlytek. These enterprises demonstrate outstanding capabilities in technology research and development, industrial scenario applications, and intelligent solutions, providing important support for the digital upgrading of the industrial industry and the development of intelligent manufacturing.

   

• automobile industryAfter in-depth research on large models in the automotive industry, Frost & Sullivan and LeadLeo recommend focusing on Huawei, Baidu, Byte beating, and Alibaba. These companies are leading the automotive industry towards intelligentization and digitization by leveraging their leading advantages in intelligent driving technology, data processing capabilities, and industry solutions.

   

• meteorological industryAfter in-depth research on large models in the meteorological industry, Frost & Sullivan and LeadLeo recommend focusing on Huawei, Tsinghua University, Shanghai Artificial Intelligence Laboratory, and CRRC Dawning. The outstanding performance of these institutions in meteorological data processing, model development, and intelligent application scenarios provides strong technical support for promoting the intelligent and precise development of the meteorological industry.

   

• medical industryAfter in-depth research on large models in the healthcare industry, Frost & Sullivan and LeadLeo recommend focusing on Inspur, Huawei, Baidu, and Tencent. These companies excel in healthcare data processing, intelligent diagnostic technology, and industry solutions, providing key support for promoting the digital upgrade and intelligent services of the healthcare industry.

   

• Pharmaceutical industryAfter in-depth research on large models in the pharmaceutical industry, Frost & Sullivan and LeadLeo recommend focusing on Huawei, Baidu, and Alibaba. These companies have demonstrated outstanding capabilities in data analysis, model optimization, and intelligent solutions for drug research and development, providing important support for improving the efficiency and precision of drug development.