Guosen: Optimistic about the continuous implementation of AI applications over the next 25 years, with inference power ready to be unleashed.

Guosen released a research report stating that global large models are continuously iterating, with model capabilities constantly improving, empowering the development of AI applications. The price of large model API calls continues to decline, benefiting application vendors by reducing costs and driving up the penetration rate of AI applications. Optimistic about the continuous implementation of AI applications over the next 25 years, it is recommended to pay attention to companies such as Beijing Kingsoft Office Software, Inc. Additionally, the vigorous development of AI applications is driving an increase in demand for inference-side computing power, and there is continued optimism for the development of domestic computing power. Furthermore, ByteDance's API calls for large models have rapidly increased, indicating a huge potential demand for computing power, with related industry chains expected to benefit. It is recommended to pay attention to Hygon Information Technology (688041.SH), Inspur Electronic Information Industry (000977.SZ), among others. Guosen's main points are as follows: AI applications: Model iterations solidify application foundations, leading to multi-domain applications for AI In 2024, OpenAI released the GPT-O series of models, utilizing chain technology to enhance reasoning capabilities. Google's Gemini large model continues to iterate, with the release of Gemini2.0Flash in December, supporting multimodal output. The capabilities of these models are constantly improving, solidifying the foundation for AI applications. Meanwhile, the price of large model API calls is decreasing, benefiting AI application vendors and accelerating the penetration of AI applications in various scenarios. Currently, AI applications have been implemented in fields such as healthcare, finance, education, office work, and CRM, and there is optimism for the vigorous development of AI applications in the next 25 years, potentially leading to a breakthrough. AI computing power: Cloud providers continue to increase capital expenditure as AI applications drive demand for inference processing Since 2023, global artificial intelligence has been rapidly developing, with cloud providers heavily investing in AI infrastructure. In Q3 2024, Microsoft, Google, Meta, and Amazon saw their capital expenditures increase by +78.6%, +62.2%, +36.1%, and +88.3% respectively compared to the previous year, driving a new wave of capital expenditure. Simultaneously, the deployment of B+C applications is driving an increase in demand for inference computing power. According to the "2023-2024 Chinese Artificial Intelligence Computing Power Development Assessment Report (IDC & Wave)", it is projected that China's share of inference computing power will be 67.7% in 2024, a 26.4 percentage point increase from the previous year. It is expected that the share of inference computing power will continue to rise in the future. Additionally, the rapid increase in ByteDance's API calls for large models suggests that its demand for computing power will grow rapidly, benefiting its related industry chains. AIAgent: On the eve of market explosion, potentially disrupting traditional working models The AIAgent market is still in its early stages. According to Roots Analysis, the global AIAgent market is estimated to be worth $5.29 billion in 2024 and is expected to reach $216.8 billion by 2035, with a CAGR of 40.15% from 2024 to 2035. As AIAgent automation levels increase, there is potential for a shift towards goal-oriented working paradigms. According to data disclosed by Y Combinator, in the Winter 2024 cohort, AI Agent projects accounted for nearly 80% of the projects, indicating that they are on the cusp of large-scale applications. Quantum computing: Google releases Willow quantum chip, market size rapidly growing Quantum computing utilizes quantum properties such as superposition, entanglement, and quantum interference for data storage and computation, providing exponential acceleration and parallel computing capabilities. In December 2024, Google released the new quantum chip Willow, achieving quantum error correction below the threshold for the first time, signaling the initial development of the industry. According to IDC, the global quantum computing and related market was worth $1.605 billion in 2023 and is expected to grow to $8.948 billion by 2028, with a CAGR of 41.01% from 2023 to 2028, indicating rapid market growth. Risk warning: Slow progress in large model development, lower-than-expected capital expenditure by cloud providers, and insufficient iteration and supply of domestic computing power.