Guosheng Securities: The new logic of the IDC industry under AI development will start from energy consumption reduction and destocking.

Guosheng Securities released a research report stating that the rapid development of AI will drive the demand for AI data centers after absorbing existing resources, leading to improvements in prices and profitability. Unlike the previous cycle, this round will see an increase in demand for electricity over IDC space due to the enhancement of computing power density. Therefore, the new logic of the IDC industry under AI development will start from energy consumption and inventory reduction. Perhaps the electricity consumption that previously required three floors can now be accommodated in half a floor. With the overall utilization rate of IDC calculated by power increasing, and against the background of increased demand for high-end services, room renovation will lead to gradually full loading of shelves. Resources in core cities and remote cities will each take what they need for training and reasoning, gradually starting a new cycle in the industry. Guosheng Securities' main points include: From IDC to AIDC: Expected to shorten the investment return cycle Traditional IDC caters to "broad coverage" general computing needs, while AIDC targets "deep computing" needs for AI, big data, and high-performance computing. The main differences between traditional IDC and AIDC are manifested in end customers, technical architecture, operating models, hardware resource requirements, among others. The core difference is that AIDC, with higher power density per rack, is expected to have a shorter investment return cycle compared to traditional IDC: Hardware facilities: IDC primarily uses general servers with lower power density (usually 4-8kW per rack), while AIDC has higher power density (20-100kW per rack). Cooling requirements: IDC mainly uses air cooling combined with room environmental control; AIDC widely uses liquid cooling. Investment return cycle: IDC has a longer construction cycle with long-term hosting and leasing as the main revenue model; AIDC has higher initial construction costs (hardware, cooling system), but with higher power density, the return period is relatively shorter. Service model: IDC mainly focuses on general hosting services; AIDC is expected to provide more vertical and professional services, such as customized cabinet deployment, specialized cooling solutions, etc. Pull of AI on data center demand: Reduce inventory first, then raise prices The rapid development of AI will drive the demand for AI data centers after absorbing existing resources, leading to improvements in prices and profitability. Unlike the previous cycle, this round will see an increase in demand for electricity over IDC space due to the enhancement of computing power density. Therefore, the new logic of the IDC industry under AI development will start from energy consumption and inventory reduction. Perhaps the electricity consumption that previously required three floors can now be accommodated in half a floor. With the overall utilization rate of IDC calculated by power increasing, and against the background of increased demand for high-end services, room renovation will lead to gradually full loading of shelves. Resources in core cities and remote cities will each take what they need for training and reasoning, gradually starting a new cycle in the industry: Inventory pressures and inventory reduction: The previous wave of cloud computing brought about a peak in IDC construction, leaving a large amount of empty or low-rent inventory, especially in core cities and remote markets. The utilization rate of these existing data centers needs to gradually increase with the increase in AI demand, and reducing inventory is the primary task of the market in the short term. AI drives revitalization of existing resources: High-density computing tasks of AI reconfigure existing data center resources, for example, through increased deployment of GPU servers, renovation of data center power supply systems, some empty traditional IDCs can quickly transform into infrastructure to meet AI demand, thereby increasing resource utilization. Outlook - Logic of price increase: After effectively utilizing existing resources, the demand for new AI data centers will drive the market supply-demand balance towards tightness, and rentals for high-power density AI specialized data centers are expected to rise. How to choose third-party hosting companies in the AIDC era In the AIDC era, the core competitiveness of third-party IDC companies will revolve around technological capabilities, customer resources, and operational efficiency. Firstly, technological capabilities are key, companies with the ability to support high-density computing (such as GPU server deployment), advanced cooling solutions (liquid cooling), and efficient network architectures will have long-term competitive advantages; secondly, customer resources determine growth potential, companies with premium customer resources such as internet giants (ByteDance, BAT, etc.), AI companies, or research institutions will have more stable business expansion potential; in addition, operational efficiency enhances profitability, companies that achieve energy efficiency optimization through low PUE values, and reduce expansion costs through modular design are expected to have relatively good profitability. Long-term optimistic about the computing power sector, the current focus is on the progress of domestic computing power and related industry chains. It is particularly recommended to pay attention to leading companies in optical communication such as Zhongji Innolight (300308.SZ), Eoptolink Technology Inc., (300502.SZ), Suzhou TFC Optical Communication (300394.SZ), as well as major Chinese companies such as ZTE Corporation (000063.SZ), Ruijie Networks (301165.SZ), Cambricon Technologies (688256.SH), Accelink Technologies (002281.SZ), MeiG Smart Technology (002881.SZ), Wuxi Taclink Optoelectronics Technology (688205.SH), T&S Communications (300570.SZ), Hubei DOTI Micro Technology (301183.SZ). It is also recommended to pay attention to the follow-up development of AIDC and the pace of inventory clearance, and to focus on high-quality data center manufacturers such as Range Intelligent Computing Technology Group (300442.SZ), Beijing Sinnet Technology (300383.SZ), Guangdong Aofei Data Technology (300738.SZ).Shanghai AtHub (688288.SH) and others.Risk warning: AI development falls short of expectations, computing power demand falls short of expectations, market competition risk.