China Securities Co., Ltd. Wu Chaoze: In 2025, we still have high hopes for domestic computing power and AI applications.

On February 25, Wu Chao, Director of the Securities Research Institute and Chief Analyst of the TMT industry at China Securities Co., Ltd., shared his views on artificial intelligence at the China Securities Co., Ltd.'s 2025 "AI+" investment strategy meeting. Wu Chao stated that in 2025, he still firmly believes in two directions - the maturity of domestic computing power and the implementation of AI applications. The investment direction with the most certainty this year is domestic computing power. It is crucial to observe whether the capital expenditure of Chinese operators will end the downward cycle earlier due to the changes brought by AI in the next two months. Specifically, Wu Chao pointed out that although there were many new changes during the Spring Festival, his forward-looking perspective for 2025 remains unchanged, as he still sees great potential in two directions. Firstly, domestic computing power is one of the most promising directions in 2025. The equalization of large models and the localization deployment of models make the computing power sector, especially domestic computing power, a highly certain direction for upstream investment. The discussion on this will follow. The other part is the "implementation of AI applications." After the Spring Festival, major companies such as Tencent and Alibaba quickly absorbed and aligned with the capabilities of DeepSeek, indicating the changes happening in the industry. Open source not only promotes free and low-cost solutions but also drives localization deployment. In the future, model innovation will not stop and may even surpass the average human level in predictive and emotional abilities. By then, the application of AI in serious scenarios such as scientific research, medical innovation, and industrial manufacturing will become more widespread, potentially arriving sooner than we imagine. It is evident that the iteration of models in the past two years is only the beginning. In the next ten or even twenty years, models may undergo more version updates, and their capabilities and impact on application scenarios need to be dynamically assessed. Regarding large models, Wu Chao's view can be divided into two aspects. On one hand, global big companies will quickly integrate services similar to Alibaba Cloud. This will also influence OpenAI's own strategy. For example, the o3-mini launched by OpenAI during the Spring Festival, although slightly more expensive than DeepSeek, has significantly reduced costs. Similar model products will continue to emerge in the future. On the other hand, the debate between open source and closed source will continue. In the era of mobile internet, Apple's iOS is a successful case of closed source, while Google's Android represents the success of open source. Looking back at the models themselves, open source and closed source are likely to coexist in the future. Wu Chao mentioned that investments can be roughly divided into four parts: the impact on computing power itself, the impact on applications, the impact on the end, and the impact on data. Additionally, the concept of intelligent agents was emphasized frequently by them last year, and they still hold onto this direction in 2025. Overall, the beta of computing power still exists, indicating that the demand for computing power has not ended, and we are not yet in a bubble. However, the proportion of demand for inference will be greater structurally. Furthermore, in the long term, there is still significant growth potential in the capital expenditure of Chinese cloud companies. Data from Q1 2023 indicates that the overall capital expenditure of Chinese internet companies is expected to be around 300-400 billion RMB. Although it has nearly doubled compared to the past domestically, it still falls short compared to North America due to exchange rates. Reflecting on cloud computing data in 2015 during the era of mobile internet, the capital expenditure of Chinese cloud companies was about 70% of North America's. Now, China's capital expenditure is only about one-seventh of North America's. Whether the capital expenditure of Chinese operators will end the downward cycle earlier due to the changes brought by AI is crucial to observe in the next two months. If the capital expenditure of operators ends early, then the capital expenditure of cloud companies and operators will become an important source of capital expenditure for computing power upstream. This is also a source of certainty for future investment opportunities. Wu Chao said that this year is actually more like a true 2013, where hardware has been laying the foundation for a long time, penetration rates are rising, and reliable operating systems are in place, leading to the emergence of large models today. However, the exploration of applications is the future and the main focus of the entire industry. When the industry can finally be called mainstream and truly become a productive force, the value of applications will undoubtedly surpass infrastructure investments.