Zheshang: Increased capital expenditure by major internet companies expected to drive the establishment of AIDC, creating a "arms race".

This report by Zheshang predicts that Bytedance's capital expenditure will reach 80 billion yuan in 2024, close to the total of Baidu, Alibaba, and Tencent (approximately 100 billion yuan). Bytedance's capital expenditure is expected to reach 160 billion yuan in 2025. Bytedance's future token consumption is estimated to be high, and an increasing amount of computing power is expected to be invested. In September 2024, Wu Di, the head of intelligent algorithms for Bytedance's Volcano Engine, stated that by 2027, the daily token consumption for Douyin is expected to exceed 100 trillion, more than 100 times the current amount. Meanwhile, internet companies such as Baidu, Alibaba, and Tencent are also participating in large-scale model competition, aiming to expand model parameters, increase user numbers, and usage rates. The development of large models depends on massive computing power investment, and it is expected that a new wave of IDC construction will emerge in the internet industry in the future. IDC is at the bottom of the cycle and may face a shortfall in the future After the expansion of the Chinese IDC market in 2020, the industry entered a downward cycle. According to data from Kezhi Consulting, the growth rate of the traditional IDC business market slowed from 22.1% in 2020 to 6.4% in 2023. The sales price of data center cabinets has also declined, with the price of custom cabinets without electrical components falling by nearly 20% since 2020, and in some regions, prices have approached the cost line, with a slight decrease in the prices of wholesale retail project cabinets. With supply shrinking and demand explosively growing, as IDC supply is inflexible and construction periods typically range from 1.5-2 years, there may be a shortfall in the IDC market, leading to an increase in IDC construction and service prices. AIDC places higher demands on energy consumption, electricity prices, and heat dissipation Existing IDC centers were mainly built before the era of large models, and may struggle to meet the demands of intelligent algorithms. The current incremental demand for IDC comes mainly from the growth in AI large model training and inference requirements. Traditional IDC (Internet Data Centers) mainly provide data services for internal and external enterprise users, while AIDC (Intelligent Computing Centers) provide computing power, storage, and related services for AI training and inference. To meet the increasing demand for computing power with large models, computing clusters are evolving towards the level of thousands of cards, leading to increasing demands on energy consumption, electricity, and heat dissipation in AIDC. AIDC primarily utilizes GPU and other chips, with a single cabinet power reaching tens of kilowatts, requiring a large amount of electricity to support its high-performance computing and data processing tasks. Therefore, AIDC needs to acquire energy consumption metrics and access relatively inexpensive electricity. In this context, green energy and energy storage become better choices. In terms of heat dissipation, driven by the trend of high heat density in computing devices and data centers, as well as the high-efficiency heat dissipation requirements, AIDC needs to be equipped with efficient cooling systems, such as liquid cooling systems. Related targets IDC: Beijing Sinnet Technology (300383.SZ), Range Intelligent Computing Technology Group (300442.SZ), Shanghai AtHub (603881.SH), INESA Intelligent Tech Inc. (600602.SH), Kehua Data Co., Ltd. (002335.SZ), Guangdong Aofei Data Technology (300738.SZ), RunJian Co., Ltd. (002929.SZ). Liquid cooling: Guangzhou Goaland Energy Conservation Tech (300499.SZ), Shenzhen Envicool Technology (002837.SZ), Guangdong Shenling Environmental Systems (301018.SZ). Switches: ZTE Corporation (000063.SZ), Unisplendour Corporation (000938.SZ), Ruijie Networks (301165.SZ), Shengke Communications (688702.SH), Shenzhen Phoenix Telecom Technology (301191.SZ). Risk warning Unexpected GPU supply, lower-than-expected computing power demand, delays in building computing centers, and market competition risks.