Abstract:As the underlying infrastructure of the AI industry, computing power plays an important role in the progress of the industry, and the continued popularity of AI large models is expected to drive the demand for AI computing power to continue to increase.
1.The explosion of artificial intelligence applications will drive a blowout of computing power demand
Generative artificial intelligence represented by ChatGPT is popular all over the world, and artificial intelligence has become a new generation of productivity. Since World War II, the Bombe machine has cracked**, laying the foundation for machine learning;From 1974 to 1993, due to the limitations of early computer processing capabilities, artificial intelligence continued to decline. Since the beginning of the 21st century, artificial intelligence has advanced by leaps and bounds, and in 2017, Google AlphaGo defeated Go world champion Ke Jie, but at this time, artificial intelligence is only suitable for limited rules and cannot be widely used in the field of production and life. At the end of 2022, with the explosion of ChatGPT, artificial intelligence has advanced by leaps and bounds, showing amazing productivity in more and more fields.
Artificial intelligence is widely used, and the implementation of related fields is accelerating. Typical scenarios of artificial intelligence include, but are not limited to, (the application will be expanded in detail in the following chapters): 1) Autonomous driving: assisted driving and autonomous driving above L2 in high-speed, complex urban roads, and closed parks;2) Intelligent manufacturing: Empower the manufacturing industry to improve the production efficiency of each link by improving the yield rate and reducing the consumption of raw materials3) Biopharmaceutical: improve the quality of drug candidates, optimize clinical trial design, and shorten the time required for each step4) Film and television culture: Use AI to reduce animation production, animation modeling, etc.
Data is finally produced into algorithms or applications through computing power, and computing power is the first direction to benefit from the development of artificial intelligence. The artificial intelligence industry chain is divided into four parts, namely computing power, data, algorithms and applications, to make an inappropriate analogy, if artificial intelligence is compared to "stir-frying", computing power is "gas", data is "materials", algorithms are "recipes", and applications are the last "dishes". 1) Computing power: The earliest direction to benefit is "water, electricity, and coal" in the era of artificial intelligence, and the cost of a single training of large models is as high as 2 million to 12 million US dollars2) Data: The basic "nourishment" for training large models, GPT-3 has parameters as high as 175 billion, and the larger the parameters, the more intelligent the AI can be improved, and higher requirements for computing power are also put forward3) Algorithm: A model that adapts to a series of tasks after data training, GPT-4 is currently the most advanced large model, with the characteristics of multi-modality, multi-language, memory and personalization, etc., and will be the stage of large manufacturers in the future4) Applications: It is the broadest market for AI, and all formats will be transformed by artificial intelligence, just like the transformation of the Internet in the past.
2.In the era of digital economy, the computing power industry has a broad space
Computing power is the cornerstone and premise of the operation and development of the digital economy. IDC's "2021-2022 Global Computing Power Index Report" pointed out that computing power has a long-term and multiplying effect on economic growth: for every 1 point increase in the computing power index, the digital economy and GDP will increase by 35 and 18. When a country's computing power index reaches 40 points and 60 points, for every 1 point increase in the computing power index, its driving force for GDP growth will increase by 15 times and 30x;
At present, artificial intelligence technology is accelerating its integration into thousands of industries, and the demand for computing power for ultra-large-scale AI models and massive data continues to rise. According to IDC, China's intelligent computing power will continue to grow rapidly, and it is expected that China's intelligent computing power will reach 1271 eflops by 2026, with a compound growth rate of 52 in the next five years3%, and the compound growth rate of general computing power scale in the same period was only 185%。
3.The United States has increased the control measures for AI chips, and the time for the localization of AI computing power has come in advance
Since Biden took office, the situation of multilateral control of China's semiconductor industry through the alliance has gradually become clear, and the "Advanced Computing Chip Rule (ACS IFR)", the Interim Final Rule for Expanding Export Controls on Semiconductor Manufacturing Projects (SME IFR) and the addition of the BIS "Entity List" have been added to provide information on chip design, foundry, production equipment, chips**, We believe that abundant computing power is the prerequisite for the successful commercialization of AI large models, and the localization of AI computing power has come in advance.
4.The investment logic and combing of the computing power industry
As the underlying infrastructure of the AI industry, computing power plays an important role in the progress of the industry. Considering that the Bureau of Industry and Security (BIS) of the U.S. Department of Commerce upgraded the restrictions on domestic high-performance chips in October 2023, the domestic computing power gap may widen in the future. In the short and medium term, computing power leasing is expected to alleviate the shortage of computing power in a more cost-effective way through the efficient use of existing resourcesHowever, in the long run, the localization of computing power is an inevitable development direction, and it needs to be continuously iteratively upgraded from the dimensions of GPU performance, networking capabilities, and software ecology to achieve accelerated catch-up. Related**: Zhongbei Communication, Winner Technology, Li Tong Electronics, Haiguang Information, Cambrian, Jingjiawei, iSoftStone, Sugon, Tuowei Information, Inspur Information, Unisplendour, Industrial Fortune Union.
Reference Material**:
1.2023-12-6Oriental Fortune** - the era of "lithium mine" of computing power, the melody of "Internet" of application.
2.2023-11-20Huatai**—Pay attention to the opportunity of computing power leasing & computing power localization.
The content shared is intended to sort out the investment direction and reference learning for you, and does not constitute investment advice, not as a basis for trading, you should refer to it based on the principle of prudence, and operate at your own risk!）