TF Securities: Policy driving domestic computing power welcomes development opportunities. Focus on domestic computing power chain investment opportunities.

Zhitong App has learned that TF SECURITIES has released a research report stating that under the AI wave, there is a broad demand for computing power in China. With limited overseas computing power supply, domestic supply is gradually filling the gap. The firm believes that in the long run, domestic substitution of computing power may become a trend. To accelerate domestic substitution, both the government and market entities are making concerted efforts to catch up with the computing power gap. With the continuous focus on domestic computing power infrastructure and domestic computing power, industry-related manufacturers are seeing investment opportunities. Key points from TF SECURITIES: 1. The AI revolution is driving a hot demand for computing power, presenting development opportunities for domestic computing power. Under the AI wave, there is a broad demand for computing power in China. In October 2023, the Ministry of Industry and Information Technology and five other departments jointly issued the "Action Plan for High-Quality Development of Computing Power Infrastructure" with development goals. From 2023 to 2025, the compound annual growth rate of China's computing power scale is expected to be 18.5%, with the goal of exceeding 300 EFLOPS in computing power scale by 2025, and the proportion of intelligent computing power reaching 35%. At the same time, with the accelerated landing of ToB/C large-scale applications in 2024, the intensive release of overseas multimodal large-scale models such as Sora and Claude3 once again demonstrates the fast iteration speed and wide application scenarios of AI models in the market. Technological innovation will continue to drive the demand for underlying computing power. Limited overseas computing power supply, domestic supply is gradually filling the gap. In 2023, the United States imposed strict restrictions on the sale of AI chips to China, including A800/H800 being restricted from sales in the Chinese market, affecting the supply side of domestic computing power demand. It was reported that NVIDIA's "special edition" AI chip H20 for China will be available for pre-order after the NVIDIA GTC2024 conference this year. However, in terms of performance, it is about one-fourth of the H100. Domestic computing power is expected to continue to break through, as Liu Qingfeng, the founder of iFlytek, stated in August last year that Huawei's GPU capabilities can already match NVIDIA's A100. NVIDIA also mentioned that Huawei is its current competitor in four out of five areas, including AI-related graphics processors. The firm believes that in the long run, domestic substitution of computing power may become a trend. To accelerate domestic substitution, both the government and market entities are making concerted efforts to catch up with the computing power gap. 2. Policy support and top-level design help build the foundation for computing power. At the central level, on February 19, the State-owned Assets Supervision and Administration Commission of the State Council held a special promotion meeting on "AI Empowering Industry Renewal" for central SOEs, emphasizing that central SOEs should integrate the development of artificial intelligence into overall work planning, accelerate the construction of a batch of intelligent computing power centers, and carry out special AI+ actions. At the meeting, 10 central SOEs signed an initiative to actively open up AI application scenarios to society. On the local government side, on February 29, Guizhou issued the first batch of "computing power vouchers" to 72 companies from within and outside the province. The "Guizhou computing power voucher" mechanism is used to purchase eligible Guizhou computing services or Guizhou data trading products. At the beginning of 2024, many places have already issued computing power subsidy policies or opened applications for computing power subsidies to strengthen computing infrastructure, enhance computing power supply, reduce the burden of enterprise computing power usage, and promote the development of industries such as artificial intelligence. This includes cities like Shenzhen, Shanghai, Hangzhou, Chengdu, and others. 3. State-owned entities are accelerating the implementation of AI infrastructure According to the IT Times, in the construction of intelligent computing centers in the past six months, intelligent computing centers with backgrounds in central enterprises, state-owned enterprises, or local governments have entered a high period of investment. There have been numerous bidding projects ranging from hundreds of millions to tens of billions, with the total bidding amount exceeding one hundred billion yuan. In terms of operators, data from the Ministry of Industry and Information Technology shows that by the end of 2023, the three major operators provided the public with 970,000 internet data center racks, with a net increase of 152,000 for the whole year. Currently, the three major operators can provide external public basic computing power exceeding 26Eflops. China Telecom's AI computing server procurement project (2023-2024) has a total amount exceeding 8.4 billion yuan and will continue to build and operate large-scale intelligent computing centers in cities like Shanghai, Wuhan, and Tianjin, expected to provide 15EFLOPS of computing power. China Mobile (2024-2025) will purchase 6,000 AI training servers for specific scenarios, with an estimated investment of over one hundred billion yuan according to the IT Times. There is also a demand for centralized procurement of liquid-cooled AI training servers, air-cooled AI training servers, data center switches, distributed file storage, and other supporting equipment. 4. Focus on investment opportunities in domestic computing power With the continuous focus on domestic computing power infrastructure and domestic computing power, related industry manufacturers are seeing investment opportunities: Optical modules/optical devices: Companies like Zhongjixuchuang, Tianfutongxin, Xinyisheng, Bochuang Technology, Yuanjie Technology (covering the electronic group), Shijiaguangzi, etc. Domestic computing power Huawei line: Recommended to focus on companies like Gaoxin Development, Fiberhome Communication, Shenzhou Digital (covering the computer group), Feiling Kesi, Huafeng Technology, etc. Intelligent computing centers: Runze Technology (covering the mechanical group), Runjian Shares, Aofei Data (covering the computer group), etc. Server switches: Companies like ZTE, Ziguang Shares (covering the computer group), Hudian Shares (covering the electronic group), and recommended to focus on Shengketongxin, etc. Risk Warning: AI development falls short of expectations, and computing power advancement is slower than expected.