Zhaojin International: AI model call costs decrease, optimistic about internet and software sector and infrastructure sub-sectors.

Zhao Yin International released a research report stating that the bank has assessed the current application scenarios of generative AI in different industries in China and globally, along with the corresponding potential market size. It believes that areas such as online advertising, search, CRM, office collaboration, and development programming may have relatively higher AI contribution rates and faster AI revenue contribution growth rates. Industries such as e-commerce retail, online advertising, tourism, gaming, healthcare, finance, education, and industrial manufacturing are expected to have larger potential monetization scales compared to other industries. From a stock recommendation perspective, the bank is bullish on three types of companies in the Internet and software sectors: AI cloud companies, companies with mature AI application products, and companies that can better empower their main businesses with AI. At the same time, the bank still sees investment opportunities in the infrastructure sector, particularly in the areas of computing power supply chain, edge-side supply chain, and wafer foundries. Zhao Yin International's main viewpoints are as follows: Efficiency optimization of AI algorithms drives further cost reduction in using large models The DeepSeek case proves that by optimizing model architecture and training algorithms, the consumption of computational resources in the training and inference process can be further reduced, and the cost of using large models is expected to decrease further. Looking at recent industry trends, after the success of DeepSeek, 1) mainstream cloud service providers both domestically and internationally announced the launch of the latest models under DeepSeek for API calls, which may benefit the revenue growth of AI-related cloud services for cloud service providers; 2) Leading large model manufacturers gradually lower the prices of their existing multimodal large models to compete and maintain market share. The bank expects the future development of the industry to follow two main paths: 1) before true general artificial intelligence (AGI) emerges, continue to increase investment in computing power for the exploration of new generation large models; 2) continuously optimize existing large models to reduce usage costs, which will help promote the gradual prosperity of AI applications. In addition, further exploration in the area of edge-side AI may help boost the enhancement of "AI + hardware" experience. AI infrastructure investment opportunities: The emergence of DeepSeek will not reduce the demand for computing power; on the contrary, it will bring a significant increase in usage volume due to lower resource usage costs (Jarvis paradox). The bank is optimistic about: 1) the computing power supply chain, which includes the design and manufacture of computing power chips, storage, network connectivity, etc. Beneficiaries include Zhongji Innolight, Shengyi Technology, and NAURA Technology Group. 2) Edge-side supply chain, such as the upgrading of consumer electronics leading to increased demand for chips and shorter replacement cycles. Beneficiaries include ZTE Corporation, Will Semiconductor, Maxscend Microelectronics, Shengyi Technology, and Shennan Circuits. 3) Wafer foundries, such as Semiconductor Manufacturing International Corporation and HUA HONG SEMI. 4) The development of autonomous driving and artificial intelligence will accelerate due to AI, benefiting autonomous driving chips, CIS image sensors, etc. Beneficiaries include Will Semiconductor. Review and potential scale estimation of AI application market 1) The main logic of enabling generative AI and large models in industries focuses on creating incremental demand and cost substitution. The incremental demand creation comes from a more precise understanding and intelligent matching of user needs, as well as providing more effective data insights based on outstanding data processing capabilities. 2) From the perspective of the long-term AI contribution rate, online advertising, CRM, and some application software and office collaboration areas may have relatively higher AI contribution rates and faster AI revenue contribution growth rates. The main reason is that the effect of AI enabling is clearer, quantifiable, and easier to promote the decision-making process of users to pay. 3) In terms of AI application scale, industries such as e-commerce retail, online advertising, tourism, and healthcare are expected to have larger potential monetization scales compared to other industries. The main reason is that these industries have a larger scale and have more widespread demand for cost substitution. Focus on AI application investment opportunities 2025 is expected to be a key year for the development and commercialization of AI applications, thanks to: 1) continuous decrease in model training and API call costs; 2) further narrowing of the gap between open-source and closed-source models, reducing the threshold for model deployment and invocation; 3) Since 2023, software and Internet companies have gradually matured and achieved a certain user base level in the research and optimization of AI product applications. In the Internet and software sectors, the bank is particularly bullish on three types of companies: AI cloud companies (such as Microsoft, Amazon, Alibaba, benefiting from the increased demand for computing power brought by the development of AI applications), companies with mature AI application products (such as Salesforce, ServiceNow, expected to speed up the commercialization of AI products by 2025), and companies that can effectively empower their main businesses with AI (such as Meta, Google, Tencent, where AI applications drive business cost reduction, efficiency improvement, or demand growth).