CITIC SEC: Large-scale models and edge AI exerting effort, focus on investment opportunities in Internet and AI applications.

CITIC SEC released a research report stating that when observing global technology market investments in 2025, from a market perspective, Chinese technology assets are believed to have a better investment value compared to American assets. Within the Chinese technology sector, the preferred choice is the Chinese internet sector, with a focus on the short-term macroeconomic recovery and policy stimulus bringing about a turning point in performance for the sector, as well as the continuous prosperity of the medium to long-term AI ecosystem providing opportunities for valuation reshaping. The report also highlights investment opportunities in the Chinese domestic AI industry chain. Key points from CITIC SEC's report include: Market Review: Since 2024, the US technology sector has been leading the global market, with A-shares and Hong Kong stocks outperforming the market, but the Hengqiang Index valuation has returned to its level before 924. US Market: In 2024, US technology stocks continued to lead the global market, with the Nasdaq rising by 34% during the year. Breaking it down by sector, in 2024 Q3, the S&P North American Software Index rose by 19%, while the Philadelphia Semiconductor Index rose by 5%, with software outperforming significantly. Looking at a longer period, from early 2023 to the end of 2024, the S&P North American Software Index rose by 109.9%, aligning closely with the rise in the Philadelphia Semiconductor Index (110.2%) during the same period. The report indicates that recent trends in the US technology sector's market performance reflect a shift in market interest towards the positive application of AI. In November, the AI service provider Palantir rose by 61%, while the AI advertising service platform Applovin rose by 99%, indicating that companies with significant potential for AI applications and fast-paced deployment are entering a track of expanding performance and valuation. Additionally, in terms of structure, the US technology sector's market performance is also spreading to small and medium-sized companies; in September 2024, the market capitalization growth of the top ten technology companies in the US accounted for 77% of the Nasdaq's year-to-date market capitalization increase, but by December 2024, this figure had decreased to 71%. Hong Kong Market: The Hang Seng Technology Index rose by 17% during 2024, with the majority of the growth driven by the market uptrend in September. In terms of structure, POP MART, Xiaomi, and Meituan were the top performers, with growth rates of 371%, 95%, and 91% respectively. After October, the Hengqiang Index experienced a pullback, with the valuation level as of December 20 returning to its level before 924, marking the market percentile at around the 10th percentile since 2021. Domestic Market: In 2024, A-share technology stocks outperformed the market, with the Shanghai Composite Index rising by 14% during the year. The indices for Electronics/Communications/Computers/Media/Automobile rose by 17%/28%/17%/19%/19% respectively. During the market uptrend since September, the technology sector as a whole rebounded more strongly than the overall market, with the computer sector, which had lower performance growth expectations in the previous period, leading in this round of market movement. In September, the Shanghai Composite Index rose by 17%, while the indices for Electronics/Communications/Computers/Media/Automobile rose by 19%/24%/35%/26%/23% respectively. Looking at the year as a whole, the communications industry benefited from overseas AI development, propelling sub-sectors such as optical modules and AIDC, leading in overall growth. In terms of themes, AI and domestic self-sufficiency are two major themes, with companies in the overseas AI industry chain and companies providing accompanying support to domestic AI self-sufficiency, such as computing chips and data platforms, seeing significant growth. Based on the active allocation proportion of public funds, as of 2024 Q3, the overall allocation proportion for the electronics sector has consistently remained at a historical high, while the allocation proportion for the communications industry has been continuously expanding throughout the year. Progress in the Artificial Intelligence Industry: Reinforcement learning becomes a new path for model performance improvement, accelerating the deployment of software and edge AI, and the Chinese AI industry is expected to move towards independent market trends. Large Models: Clear competition landscape; the growth rate of model parameters has slowed down, and reinforcement learning has become a new path. In terms of model parameter size, due to restrictions from cluster scale and high-quality data supply, the release speed of large parameter models lagged behind market expectations since the release of Claude 3.5 in mid-June. By December 2024, with the release of the o3 model by OpenAI, the reinforcement learning paradigm has become a new method to continue improving model performance, enhancing model mathematical and logical capabilities. By combining prolonged reasoning time and a thinking chain application, OpenAI has continued to improve the accuracy of model responses and its ability to answer complex mathematical questions. Domestic manufacturers have also accelerated the penetration of large models in B and C ends by reducing model costs; the cost of large models for ByteDance has continued to decrease, with a rapid increase in C-end DAU growth at over 15% compound monthly rate and B-end tokens calls exceeding market expectations, rapidly rising from 120 billion times in May to 40 trillion times in November. AI Software Applications: At the application level, US technology companies continue to advance the accelerated deployment of large model technology. 1) Companies with large models are accelerating the release of software features, adapting to a variety of ecosystems and application environments. Google introduced Gemini 2.0 Flash version, Anthropic introduced Compute Use; OpenAI released ChatGPT iOS version and Canvas. 2) B-end SaaS companies have rapidly improved their understanding and technical level of Agents, with companies such as Workday and Salesforce accelerating the launch of Agent products and beginning to explore the monetization of AI Agents, leading to a simultaneous increase in company valuation levels. 3) Project-based AI companies and AI advertising platforms lead in deployment speed...The Robotaxi project is set to begin mass production in 2026. In the realm of humanoid robotics, as the Siasun Robot & Automation component industry chain gradually matures and large models are introduced, companies such as Nvidia, Tesla, and Figure continue to iterate on the commercialization of Siasun Robot & Automation, with several potential catalysts expected in 2025. Looking back at the domestic market, many technology companies have released AI glasses, AI headphones, and AIoT products, leading to a continuous fermentation of the market and numerous popular products. Companies like Xiaomi and Ideal are also actively exploring the application of AI in the fields of autonomous driving and intelligent cabins.Investment thesis: Looking ahead to 2025, the bank is optimistic about the performance of the Chinese technology sector driven by AI. Chinese technology: The valuation level of the Hang Seng Technology Index is at a low percentile since 2021, and foreign ownership is also at historically low levels. Share buybacks continue to enhance the EPS of internet companies, providing a strong safety margin. Positive changes in macroeconomic policies will bring a double boost to the sector's performance and valuation, providing significant upside potential. Cyclical sectors such as e-commerce, local services, transportation, freight, online recruitment, real estate services platforms will benefit significantly. Companies with a solid foothold and high earnings certainty will enjoy a valuation premium. Looking forward to the medium to long term, the domestic AI industry chain represented by ByteDance's AI ecosystem will continue to iterate, with multimodal capabilities and accuracy continuously improving. Both B-end and C-end penetration continue, and Chinese internet companies such as Tencent and Alibaba have rich vertical data and application scenarios. By 2025, the landing of more AI software applications or edge products will effectively boost the valuation level of Chinese internet companies. US tech stocks: Looking ahead to 2025, the bank believes that the core focus of global technology sector investment in 2025 will still be on AI, and advises investors to focus on investment opportunities in the software SaaS sector. For the software SaaS sector, the bank judges that many negative factors that previously suppressed the software sector are gradually easing or turning into positive catalysts. Additionally, IT spending by European and American enterprises is gradually recovering after the election, and Agent-driven AI+ software monetization is expected to accelerate in 2025. For the hardware sector, AI data centers are expected to continue to be a driving factor for the fundamental growth of 2025, while the valuation level of related companies is at a relatively high level. Risk factors: Investors should pay full attention to the risks brought by changes in the external geopolitical environment and escalating trade frictions to investments in the technology industry. Potential US tariffs and trade policies may hinder exports of technology products such as optical modules, lithium batteries, complete vehicles, and components to North America or weaken their competitiveness. Geopolitical changes may prevent mainland Chinese IC design companies from using advanced wafer foundry production capacity located overseas. The bank also warns of risks such as slower-than-expected macroeconomic recovery progress, industry policies falling short of expectations, unexpected rate cuts by the Federal Reserve, slower-than-expected progress in core technology and product development, slower-than-expected deployment of AI applications, and lower-than-expected capital expenditures by cloud vendors.