Debon Securities: DeepSeek accelerates the formation of a closed loop of domestic computing power and the value network is expected to explode.

DB Securities released a research report stating that, against the backdrop of AI-driven development and increasingly strict policy regulations, the supply and demand structure of the domestic IDC industry is expected to marginally improve. The firm believes that companies that possess core regional data center resources and have laid out AIDC infrastructure will be the first to benefit. With the rapid increase in AIDC demand, the firm also believes that the server and switch industries are likely to benefit significantly, and recommends keeping an eye on domestic related companies. Data center leasing can be divided into retail and wholesale types. Retail data centers primarily cater to small and medium-sized enterprise clients, with lower occupancy rates but higher rental income per cabinet. Wholesale data centers, on the other hand, serve large enterprises and offer highly customized large capacity modular spaces, with higher occupancy rates but lower rental income per cabinet. The firm's main points are as follows: Computing power reflects network value, AIDC becomes the pioneer of infrastructure development. The IDC sector utilizes existing internet communication lines and bandwidth resources to establish standardized telecommunication-grade data center environments, providing comprehensive services such as server hosting, leasing, and value-added services for enterprises and governments. The IDC industry in China has seen rapid growth, with the total capacity of data center racks increasing from 3.15 million in 2019 to 8.1 million in 2023. However, this has led to a short-term oversupply issue in the market. The occupancy rates of data centers in first-tier cities remain high, but there is limited demand for computing power in the central and western regions as well as the northeast. In this imbalanced supply and demand scenario, some companies have experienced varying degrees of decline in IDC rental income. Based on the demand driven by AI and the tightening of domestic energy consumption evaluations, the firm believes that the supply and demand structure of the IDC sector is expected to improve. On the demand side: The development of large AI models is expected to drive AIDC demand. Since the popularity of ChatGpt, large models have been continuously emerging in China, with increasing performance levels. For example, Deepseek-R1's performance is on par with OpenAI's o1 official version, but at a lower cost. The API service pricing for Deepseek-R1 is 1 RMB per million input tokens (cache hit) and 4 RMB per million input tokens (cache miss), while output tokens are priced at 16 RMB per million tokens. In comparison, the prices for OpenAI's o1 are 55, 110, and 438 RMB respectively. The firm believes that the increased adoption of low-cost inference models such as Deepseek will lead to a significant reduction in inference costs and prices, potentially bringing prosperity to the application side and driving a substantial increase in total computing power demand, thereby driving AIDC demand. From a capital expenditure perspective, domestic cloud providers, foreign cloud providers, and domestic operators have substantially increased their investments in AI, keeping AIDC demand robust. Data center supply is accelerating clearance, with attention to intelligent center switch, server demand. From a capital expenditure perspective, domestic cloud providers, foreign cloud providers, and domestic operators have substantially increased their investments in AI, keeping AIDC demand robust. Profitability analysis: New data center revenue promising. Data center leasing can be divided into retail and wholesale types. Retail data centers primarily cater to small and medium-sized enterprise clients, with lower occupancy rates but higher rental income per cabinet. Wholesale data centers, on the other hand, serve large enterprises and offer highly customized large capacity modular spaces, with higher occupancy rates but lower rental income per cabinet. Taking Guangdong Aofei Data Technology's new generation cloud computing and artificial intelligence industrial park in Langfang Gu'an B and C buildings as an example, the firm calculated the post-tax investment payback period for the project to be 6.7 years (including the construction period), with a post-tax internal rate of return of 12.3%. In the case of considering loans, the capital IRR is as high as 25.8%. Risk warning: Increased industry competition; computing power capital expenditure falling short of expectations; IDC construction falling short of expectations.