According to the Zhixin Finance app, Pacific Securities released research reports stating that in the coming year, it is necessary to focus on three main themes of AI: 1) Focus on the development of AI applications for B-end users and end-side deployment: To cope with the high cost of model calls and achieve commercialization of applications, applications targeting higher-value B-end users will be implemented first. At the same time, end-side deployment is an important path for AI application to achieve application outbreak, and end-side deployment can reduce the demand for cloud computing power and inference costs. 2) Focus on progress in multimodal technology: The commercial value of the content form generated by multimodal models corresponding to application scenes is greater. In addition, applications developed based on multimodal models have more diverse interaction methods, which can further increase the volume of applications. 3) Focus on the demand for large-scale models with independent controllability: OpenAI disconnecting service is expected to increase the attention of domestic large-scale models, which is bullish for companies with universal and vertical large-scale models.

Please use your Asia Vets account to access the feature.

3) Pay attention to the demand for large-scale autonomous and controllable models: OpenAI's breakthrough is expected to increase the attention to domestic large-scale models, which is bullish for companies with general and professional large-scale models.

Regarding individual stocks, the bank suggests paying attention to: 1) General large models: Kunlun Tech (300418.SZ); 2) Vertical-class large models: Guangdong Insight Brand Marketing Group (300781.SZ), COL Group Co., Ltd. (300364.SZ), and so on.

The main points of Pacific Securities are as follows:

Theme 1: Focus on the development of AI applications for B-end users and end-side deployment

At present, the overseas mainstream large model call cost is still high, such as GPT-4o, which costs $5 for every million tokens input and $15 for every million tokens output. Although domestic large-scale model manufacturers have experienced collective price reductions of APIs, the price adjustment is for the low-performance parameter models. High-performance large-scale model API prices remain high, such as Baidu ERNIE-4.0, which costs 120 yuan for every million tokens input and output. We believe that high-performance underlying large-scale models are the basis for ensuring the degree of intelligence and user experience of applications. Therefore, under the circumstances of high-cost model call for high-performance models, AI applications targeting higher-value B-end users will be implemented first. At the same time, end-side deployment is an important development path for AI applications to reach mass users and promote the popularization of AI and the realization of application outbreak. In addition, end-side deployment can reduce the demand for cloud computing power and inference costs. Bullish on the development of AI applications targeting B-end users and end-side deployment.

Theme 2: Focus on progress in multimodal technology

The commercial value of the content form generated by multimodal models corresponding to application scenes is much greater than that of large language models. In addition, applications developed based on large language models have relatively single interaction methods with users, while applications developed based on multimodal models can interact with users through various ways such as voice and video in addition to text conversations. More diverse interaction methods can enrich the use scenarios of AI applications and further promote the volume of applications. Taking OpenAI as an example, according to the AI Product List data, after the launch of the mixed multimodal model GPT4o with voice and video interaction methods, ChatGPT's monthly visit increased by 39.60% year-on-year, breaking through the bottleneck of single-digit growth in previous months. We believe that applications developed based on multimodal models have higher commercial value and can provide more diverse interaction methods, enhancing the integration of applications and scenarios. Continue to pay attention to the progress of multimodal technology in the future.

Theme 3: Independent controllability of the model, OpenAI disconnecting service promotes the attention of domestic large-scale models

Starting from July 9th, OpenAI will stop supporting API usage in countries and regions other than the 161 countries and regions on the list providing API services, China and Hong Kong are not on the list. This OpenAI disconnecting service will bring about the migration of domestic application development to domestic large-scale models, accelerate the iteration speed of domestic large-scale models, and promote the attention of domestic large-scale models. Bullish on companies with universal and vertical large-scale models.

Core points:

We believe that in the coming year, it is necessary to focus on three main themes of AI: 1) Focus on the development of AI applications for B-end users and end-side deployment: To cope with the high cost of model calls and achieve commercialization of applications, applications targeting higher-value B-end users will be implemented first. At the same time, end-side deployment is an important path for AI application to achieve application outbreak, and end-side deployment can reduce the demand for cloud computing power and inference costs. 2) Focus on progress in multimodal technology: The commercial value of the content form generated by multimodal models corresponding to application scenes is greater. In addition, applications developed based on multimodal models have more diverse interaction methods, which can further increase the volume of applications. 3) Focus on the demand for large-scale models with independent controllability: OpenAI disconnecting service is expected to increase the attention of domestic large-scale models, which is bullish for companies with universal and vertical large-scale models. We recommend focusing on: 1) General large models: Kunlun Tech; 2) Vertical large models: Guangdong Insight Brand Marketing Group, COL Group Co.,Ltd., etc.

Risk warning:

Risks of not meeting expectations for AI technology development, policy and regulatory risks, and increased competition in the industry.