CITIC Securities: Continuing to focus on opportunities to revalue China's technology assets brought about by the implementation of AI applications

The Zhitong Finance App learned that CITIC Securities released a research report saying that looking ahead to the second half of 2025, the main investment line in the technology industry will continue to revolve around AI. With the continuous advancement of big models and the exploration of ecology and scenario implementation by Chinese technology companies, AI is moving from a single point of technological progress to a broader stage of productivity implementation. Improvements in product forms and production efficiency brought about by AI have changed specific product forms or industry patterns in the fields of advertising, games, management software, smart cars, etc., creating new value for the industry.

Over the past six months, China's AI industry has received continuous attention from the whole society, driven by DeepSeek's circle-breaking effect. With the expected release of DeepSeek R2 and GPT-5 in the second half of the year, CITIC Securities determined that the process of revaluing China's technology industry is still not over. Investors are advised to continue to pay attention to the investment opportunities brought by the expansion of the AI ecosystem and application implementation to the industry segment.

CITIC Securities's main views are as follows:

Review and outlook: We continue to pay attention to opportunities to revalue Chinese technology assets brought about by the implementation of AI applications.

Looking back at the global technology sector trends in the first half of 2025, Chinese technology stocks benefited from the wave of value revaluation brought by DeepSeek. Hong Kong stock technology rose more clearly, and experienced a correction after experiencing rapid valuation correction; US stocks were limited by uncertainty brought about by the tariff war and overvaluation in the early stages, and the overall increase was weak. As of May 27, 2025, the Hang Seng Technology Index/Golden Dragon China Index/Nasdaq Index/Philadelphia Semiconductor Index were +15%/+6%/-3%/-6% respectively; the A-share technology sector was highly fragmented, and the CITIC Automobile/Media/Computer/Electronics/Communications Index were +12%/+6%/+1%/-3%/-4%, respectively. Looking ahead to the second half of the year, although China's AI industry faces constraints such as insufficient computing power and external environmental disturbances, large model capabilities are closely following the global frontier, and the domestic computing power industry is also progressing rapidly. The AI ecosystem construction of leading technology companies continues to advance, and technological progress continues to evolve into implemented productivity. The subsequent gradual implementation of AI in software applications and terminal hardware will open up more market space for the technology industry. From a valuation perspective, the Hang Seng Technology Index's PE-TTM is 20.6 times, at the 21.6% fraction since 2023, and the NASDAQ index is 39.2 times in the 44.9% fraction over the past four years. We believe that China's technology assets still have a comparative advantage, and the revaluation process is still not over.

Model and computing power: The model continues to advance, tools are gradually being enriched, and computing power remains high.

1) Looking at the base model, in the first half of 2025, the cost performance ratio of the large model improved rapidly. On the one hand, it is reflected in rapid progress in the model's endogenous capabilities, including reasoning and tool calling capabilities; on the other hand, it is reflected in the overall reduction in the model's pricing, which has benefited from the open source of the model and improved hardware efficiency. Among them, Chinese manufacturers represented by DeepSeek, Qwen, and Doubao are rapidly narrowing the capability gap with leading US model makers, and have greatly promoted model equality;

2) Judging from the landing support, with the rapid improvement in the cost performance ratio of large models, the prospects for agent landing are becoming more and more promising. On the one hand, Manus is showing two major possible routes: model+tool and agent collaboration; on the other hand, the industry has gradually determined the overall concept of using communication protocols such as MCP and A2A to achieve standardized tool call and agent collaboration, while continuously improving the model's endogenous capabilities.

3) Looking at computing power infrastructure, both North America and China are at a high level of computing power investment. The combined capital expenditure of the four major CSP in 2025Q1 was +60% year-on-year, and the full-year guidance was better than expected. Under a neutral scenario, we expect the total capital expenditure of the four largest CSPs in North America to be US$345 billion in 2025, +37% compared to 2024. Domestically, in the 2025Q1 quarter, Alibaba/Tencent invested 246/27.5 billion yuan in capital expenditure respectively, an increase of 141%/91% year-on-year. For the cloud business, current market demand is booming, and AI is deterministic about optimizing existing businesses, especially advertising and recommendation systems. We expect Chinese internet giants to maintain their strategic strength and continue to invest in cloud and AI.

4) Looking at the underlying technology, domestic computing power chip manufacturers are facing an opportunity to accelerate deployment due to the US continuing to increase the control of high-end computing power chips. At the same time, it is becoming more and more difficult for chip companies in mainland China to obtain advanced external manufacturing processes. In this context, the only option is to develop independent advanced process capabilities. The scarcity of the only available advanced process fabs will increase significantly.

Ecology and application: AI is fully implemented from point to surface.

From a macro perspective, the new quality of productivity represented by AI has essentially become one of the main lines of competition between China and the US. The increase in total factor productivity brought about by the new quality of productivity is also the proper meaning of China's economic transformation. Judging from the degree of application, the AI model is evolving from single-point technological breakthroughs to a larger industrial layout of new quality productivity, from software to end side, from ecological construction to actual productivity implementation, and the application progress of AI by Chinese manufacturers is speeding up:

1) Leading Internet vendors are actively exploring the integration of AI with their existing ecosystem in the medium to long term. We believe that AI will reshape the main business of major Internet companies in various aspects such as advertising and marketing, content creation, and search. Leading Internet vendors are expected to further deepen their business moats through AI opportunities and explore monetization opportunities in various directions. In terms of content production and marketing, AI has also seen actual economic benefits. Improved algorithm capabilities have brought about an upgrade in ad distribution and matching efficiency of leading Internet advertising platforms; while the ability of generative models has also enriched creative choices during content production and reduced the cost and difficulty of content creation.

2) In terms of software, in 2025, agent product forms will continue to be enriched, players will continue to diversify, and the degree of automation will gradually increase. Looking at the C-side, agent products such as overseas OpenAI Operator and Anthropic Computer Use and domestic intelligent spectrum Autoglm continuously verify model complex task disassembly, environmental dynamic sensing, and direct terminal operation capabilities, and are expected to be used as a technical foundation to support the accelerated implementation of high-value AI agents in various fields; from the B-side, domestic and foreign manufacturers such as enterprise management, medical industry, financial industry, etc. will continue to launch next-generation products, and their capabilities will further improve. Combined with each manufacturer's understanding of vertical scenarios and data, it is expected that 2025H2 accelerates product application.

3) Smart cars are expected to become the next Internet portal after PCs and mobile phones. In terms of sales, we expect total automobile sales in China to reach 33.44 million units in 2025 under a neutral assumption, +6.4% year over year. Among them, the passenger car industry's sales volume (including exports) will reach 29.24 million units, +6.4% over the same period last year. Furthermore, the EU's restrictions on Chinese trams or a turning point are expected to become a new theme. We continue to be optimistic about the Chinese automobile industry's overseas opportunities. Overseas sales are expected to reach 6.74 million units in 2025, +15.3% year over year, including 5.7 million passenger cars, +15.0% year on year, and exports of new energy vehicles are expected to reach 1.8 million units, +47% year over year. At the same time, the trend of automobile intelligence and electrification is further accelerated. The L2+ penetration rate is expected to double in 2025, and the sales share of highspeed/urban NOA models is expected to increase to 35%, bringing investment opportunities in segments such as smart driving chips and lidar.

Risk factors:

In the second half of 2025, we urge investors to pay full attention to the risks posed to investment in the technology industry caused by changes in the external geopolitical environment and intensification of trade frictions. Subsequent interpretations of tariffs and trade frictions may prevent domestic exports of technology products such as optical modules, lithium batteries, automotive vehicles, and components to North America or weaken product competitiveness; changes in geopolitics may prevent IC design companies in mainland China from using advanced process production capacity in overseas wafer foundries for streaming. In addition, the application of AI is still in its early stages. The AI applications we have seen have only produced economic benefits in a limited number of fields, while larger applications rely on the integration process of AI models and existing digital infrastructure, and companies' investment in AI may fluctuate cyclically as a result. We urge investors to pay full attention to the risks posed by this investment cycle in the technology industry. We also suggest that macroeconomic recovery is falling short of expectations; the risk that related industrial policies fall short of expectations; progress in enterprise core technology and product development falls short of expectations; AI applications are falling short of expectations; risks such as the capital expenditure of cloud vendors falling short of expectations; the uncertainty of global and domestic epidemics; and the risk that domestic government and enterprise IT spending will fall short of expectations due to weak macroeconomic growth.

Investment Strategy:

We believe that the implementation of new quality productivity brought about by AI will be the main line of technology investment in the second half of 2025. AI is gradually moving from single-point technological progress to the implementation of a wider range of new quality productivity, and it is also more closely integrated with existing digital intelligence infrastructure. In the context of AI application implementation, we recommend leading technology companies with medium- to long-term ecological advantages and companies that benefit from the opportunities brought by AI application implementation and computing power construction.