CITIC Securities: Large-scale implementation of AI applications officially starts, optimistic about the prospects of AI applications going global.

CITIC SEC released a research report stating that the next generation of large models (Gemini 3.0/GPT5) is substantially favorable for complex reasoning scenarios. The AI industry is truly moving towards the stage of large-scale implementation, with business models gradually evolving from AI project delivery, AI empowerment of existing applications, to AI functional application evolution. Compared to the domestic market, the overseas market is ahead of the domestic market in terms of technological progress, payment environment, business models, and market space. Overseas AI application ARR data continues to grow, accelerating its implementation. The agency is optimistic about domestic companies' prospects for AI applications going global based on their product and engineering capabilities, innovation, and speed of iteration. The main points of CITIC SEC are as follows: Trends in the AI application industry: Moving from a technological singularity to a commercial singularity, the phase of large-scale implementation has officially begun. Since the release of the next generation of large models (Gemini 3.0/GPT5), there has been a significant acceleration in the implementation of AI applications, favoring complex reasoning scenarios and improving quality, efficiency, and cost reduction. Category one: AI project delivery (2023-) takes the lead. Since the significant changes in the capabilities of large models in early 2023, there has been a high demand for AI project delivery, with the industry mainly focusing on pioneering and exploratory non-standard projects. Category two: Accelerated upgrading of AI-enabled existing applications (2025-). Since the release of DeepSeek in early 2025, the cost-effectiveness of AI applications on the ground has accelerated, with many application companies mainly adding AI modules for free trial or pilot testing. Category three: AI functional applications (2026-) are expected to flourish. The agency expects that starting in 2026, with further improvement in the capabilities of large models, more independent products will emerge in AI applications, with the business model mainly focusing on independent charging (excluding large model API charges). According to the data from aiapps100, the current overseas AI application ARR data continues to grow, with the ARR scale of the top 100 overseas AI applications reaching nearly $40 billion in July. In the primary market, the annual revenue scale of agent-type applications represented by Manus has reached $90 million. In the secondary market, whether it is the AI project delivery-type Palantir or the AI-enabled existing application Applovin, the rapid growth of AI applications continues to drive the improvement of company performance. Overseas market: Nurturing a huge demand wave, with potential market space far greater than domestic markets for both C-end and B-end. 1) Technological advancements: In terms of models, the global evolution speed of large models continues to exceed expectations, with the daily token usage of companies like Byte and Google experiencing significant growth. In terms of computing power, with the exponential growth in data processing volume, the agency predicts that the investment scale of the four major CSPs in the AI server domain by 2025 will be $162.4 billion (YoY +94%). 2) Payment environment: Referring to the pricing of Kingsoft WPS and Microsoft Office, the domestic market differs significantly from overseas markets in terms of software payment habits due to factors such as user habits, market ecosystem, pricing strategies, and intellectual property protection. 3) Business models: Since the second half of 2024, overseas SaaS companies have gradually evolved into Agent forms on the B-end application side, which helps users quickly adopt AI application products and also improves the profitability of high-value scenario applications. 4) Market space: C-end: Overseas user base is larger, with data from Similarweb showing that the overall monthly visit volume of AI applications in the U.S. is approximately 2.5 billion times, with countries like India, Brazil, and the UK each exceeding 300 million times, while China is about 1.1 billion times per month; B-end: Overseas companies have a large number of enterprises, with the average profit of the Fortune 500 companies in the U.S. being more than twice that of Chinese companies, indicating stronger payment capabilities. Domestic AI applications going global: Best solution and inevitable choice to seize a share in the global AI application market. 1) Product and engineering capabilities: According to China Economic Net, in 2025, the total number of STEM graduates in China exceeded 5 million, with STEM doctoral graduates expected to reach over 77,000, more than twice that of the United States, indicating a large talent base and solid engineering capabilities. At the same time, based on the product launches by companies like Wanxing Bo Expol and Seekee, domestic engineering teams have demonstrated rapid learning, agile development, and continuous iteration capabilities. 2) Success case validation first: Beijing Kingsoft Office Software, Inc.(WPS) succeeded in achieving a "curve overtaking" on the mobile side in competition with Microsoft Office through keen judgment on the mobile internet wave, decisive strategic transformation, and in-depth localized product innovation. During the current AI wave, domestic AI application companies are expected to achieve output reduction based on mature internet product capabilities, engineering capabilities, and localized operation strategies. 3) Overseas revenue situation of domestic software companies: The agency selected software companies from the CITIC Computer Index, and based on the criteria of a market value of 7 billion yuan and overseas revenue share of more than 5% in 2024, found that the highest proportion of overseas business revenue for TOP AI application companies in 2024 can reach 90%, with revenue growth reaching more than tenfold. Risk factors: Risks of AI core technology development falling short of expectations, risks of enterprise data security, information security risks, risks of intensified industry competition, and geopolitical risks.