Large models need to balance technology, human resources, and capital elements

The development of the AIGC industry is a comprehensive test field for technical conditions, talent conditions and capital conditions. Among them, technical conditions are undoubtedly at the core. The main technologies of AIGC include algorithms, computing power, and data, which are mutually causal.

The current AIGC industry chain consists of data supply, model development and customization, application, and distribution. Currently, the model layer is a key factor; secondly, there is great room for development in the application layer. The upstream data supply collects large amounts of raw data, pre-processes it, and provides it to models for training, with high investment certainty. In the middle, using annotated data to develop and train AI models to generate content, developing and refining them in vertical sub-sectors to adapt to customized needs; downstream assists users to use models and algorithms to generate content such as text, images, and videos. Based on different value creation logic, generated content is distributed through various channels.

In the short term, general-purpose large models have a high market demand and long-term multiple model combination is the future development direction. According to model size, AIGC models can be divided into large models, small models, and micromodels. Large models have more parameters and strong computing power, and are capable of solving various problems in general. Small models focus on specific industries and often possess sufficient data and problem-solving capabilities in certain scenarios. Micromodels are more personalized and trained using personal user data.

AI is mainly divided into infrastructure layer, framework & model layer, and application layer at three levels. The core of the infrastructure layer is to provide computing power, including CPU, GPU servers, etc. The model layer focuses on AI model products, with a longer investment cycle and certain technical barriers. The application layer is the downstream part of the AI industry chain, directly serving customers and users, mainly including consumer-end terminals for end users and industry solutions for businesses. There are relatively low entry barriers in the application layer while China has broad application scenarios for AI applications, so there are many opportunities here.