CITIC Securities: It is expected that the attractiveness of leading technology assets in China will gradually emerge, focusing on domestic large models and computing power processes

Zhitong
2025.01.15 00:01
portai
I'm PortAI, I can summarize articles.

CITIC Securities released a research report indicating that after 2025, the attractiveness of AI assets in China is expected to gradually emerge, with leading AI companies likely to achieve stable growth and profitability. As U.S. tech stocks face performance and valuation pressures, the relative attractiveness of Chinese tech assets will increase, with a focus on investment opportunities in domestic large models, computing power processes, and more

According to the Zhitong Finance APP, CITIC Securities released a research report stating that since the launch of ChatGPT, AI assets in the United States have received unprecedented attention from global investors. The firm predicts that after 2025, the attractiveness of Chinese AI assets will gradually become apparent. Leading AI companies are expected to achieve steady revenue growth and profitability through continuous high investment, rapid iteration, and a bold approach, as well as performance-driven alpha investment opportunities. The valuation of leading AI companies has a significant comparative advantage over comparable companies in the United States. Once the U.S. tech stocks face dual pressure on performance and valuation due to a slowdown in CAPEX growth, the attractiveness of Chinese assets will further emerge. Key investment opportunities include domestic large models, domestic computing power processes, intelligent driving, and embodied intelligence companies.

The main points of CITIC Securities are as follows:

Market Review and Outlook: U.S. tech stocks will lead the global market in 2024, and the firm expects the attractiveness of leading Chinese tech assets to gradually become apparent after 2025.

In 2024, the Nasdaq rose by 34%, with a cumulative increase of nearly 100% from the end of 2022 to the end of 2024. NVIDIA led the U.S. tech giants, rising 171% in 2024 and contributing about 19% to the Nasdaq's increase during the same period, with a cumulative increase of 831% from 2023 to date. U.S. tech stocks have seen a dual rise in earnings and valuations, attracting unprecedented attention from global investors. In 2024, the Shanghai Composite Index rose by 14%, while the A-share indices for electronics/communications/computers/media/automobiles rose by 17%/28%/17%/19%/19%, respectively. The Hang Seng Tech Index rose by 17%, all outperforming the indices. The firm predicts that in the next 3-6 months, advancements in AI technology and its application may drive U.S. stocks higher until the CAPEX growth of tech giants slows down. Once the earnings growth of U.S. giants slows and valuations come under pressure, the relative attractiveness of Chinese AI assets will further emerge. By 2025, the continued investment of leading AI companies is expected to gradually reflect in their performance, and their valuations will have a significant comparative advantage over comparable U.S. companies, with the valuation advantage of leading Hong Kong tech companies being even more pronounced.

Domestic AI Models: The performance of AI model companies is expected to continue high growth, with models driving B-end demand as infrastructure and applications driving C-end demand.

The capabilities of ByteDance's Doubao model continue to improve, with the SuperCLUE general model benchmark evaluation for 2024 showing Doubao in the leading tier. According to the aicpb website, the monthly active users of the Doubao APP have reached 71.16 million, ranking first in the country. At the 2024 Volcano Engine Winter Force Power Conference, it was reported that by December 2024, the daily token call volume of the Doubao large model had reached 4 trillion times, a 33-fold increase since May. The firm believes that the usage and penetration rate of Doubao are still in a phase of rapid growth, and the daily token call volume will continue to grow significantly. From a business model perspective, models serve as infrastructure, and as the capabilities of the Doubao model improve, prices have significantly decreased, leading to a continuous increase in call volume from B-end clients such as mobile phones, automobiles, and smart terminals; models also serve as applications, with significant room for growth in MAU and daily token usage for C-end clients. Two related investment opportunities are highlighted: 1) ByteDance's computing power chain IDC company, and 2) leading internet companies with similar products Domestic Computing Power Process: The core competition of AI computing power lies in the capability of advanced process chips, and the status of foundries as core strategic assets is expected to continue to strengthen.

External sanctions have significantly enhanced the scarcity of "available advanced processes," shifting the supply-demand relationship from global matching to limited domestic solutions. The U.S. restrictions on AI and semiconductors for China have tightened, and on January 13, the Biden administration officially announced new global control measures on the export of AI chips, implementing strict controls on the procurement of advanced AI chips from mainland China. Chinese model companies, while procuring NVIDIA chips, also need to strategically consider the supply of domestic AI chips. For domestic AI chip companies, the availability of "international advanced processes" has decreased, while the scarcity of "available advanced processes" has increased. The demand for advanced processes in domestic fields such as mobile phones, computing power, CPUs, and intelligent driving will continue to drive growth. Semiconductor Manufacturing International Corporation (688981.SH), as the leading domestic foundry with advanced process capabilities, is expected to see simultaneous improvements in demand, profitability, and valuation.

Intelligent Driving and Embodied Intelligence: Automotive intelligent driving is the first scenario for AI applications to land and is expected to generalize to embodied intelligence and low-altitude economy fields.

VLM+ end-to-end technology drives automotive intelligent driving from "available" to "usable," from "function" to "experience." The integration of high-speed NOA and urban NOA in L2+ intelligent driving is gradually entering a popularization phase, with the firm expecting L2+ intelligent driving penetration rates of 14% and 30% in 2024 and 2025, respectively, mainly driven by automakers such as BYD (002594.SZ), Xiaomi, and Huawei, which will drive the intelligentization of millions of vehicles. The VLA model, which integrates vision, language, and motion, is seen as a key technology for the next generation of intelligent driving, and VLA technology is also applicable to robotics. Tesla (TSLA.US) leads automotive companies into the field of embodied intelligence, with Xiaomi, XPeng (09868), BYD, and others following suit; Fourier, Galaxy General, Star Motion Era, and Zhiyuan Robotics appeared at NVIDIA's CES 2025 launch event. In addition, automakers are actively participating in the low-altitude economy, such as XPeng Huitian and Geely WoFei.

Risk Factors:

Risks of intensified geopolitical and trade frictions, risks of product export obstacles or weakened product competitiveness, risks of macroeconomic growth not meeting expectations, and risks of AI technology progress and application landing speed not meeting expectations, etc