According to a report on March 25 by the Financial Association (Editor: Feng Yi), due to market concerns about the excessive computing power, the Hong Kong stocks related to the AI Infrastructure Concept experienced a significant drop today, even dragging down the overall performance of Technology stocks.

As of the market close,$KINGSOFT CLOUD (03896.HK)$down nearly 6%,$GDS-SW (09698.HK)$down about 5%;$CHINA UNICOM (00762.HK)$、$BABA-W (09988.HK)$Several stocks also fell more than 3%.

In terms of news, Goldman Sachs has drastically lowered the expectations for Global AI Server shipments and supply chain stock prices.$Alibaba (BABA.US)$Chairman Cai Chongxin's remarks about the bubble in AI Datacenter construction quickly drew market attention.

On March 24, Goldman Sachs' analyst team lowered the forecast for rack-level AI Server sales. Goldman Sachs believes that while the second quarter of 2025 will be a strong quarter for ODM/cooling supply chains, it holds a more conservative outlook for the total annual shipments, and it expects the product transformation period may affect shipments again in the third quarter of 2025.

As a result, Goldman Sachs correspondingly lowered the Target Prices for supply chain-related companies (Quanta, Foxconn, FII, Wistron, Aiven, and Hostpro), with declines ranging from 7% to 21%.

Furthermore, Alibaba Group's Chairman Daniel Zhang stated at the HSBC Global Investment Summit held in Hong Kong that signs of a bubble are starting to appear in the construction of AI Datacenters. Many investment announcements for Datacenters in the USA are 'repetitive' or overlap with each other.

On the other hand, driven by the large-scale application of DeepSeek. The market had previously had high expectations. $BABA-W (09988.HK)$ 、 $TENCENT (00700.HK)$ 、$China United Network Communications (600050.SH)$Regarding the capital expenditure plans of various technology companies and telecommunication operators. However, based on the latest guidance disclosed by major companies, there is no significant deviation from market expectations.

According to a report by Guosen, after the release of the 2024 Earnings Reports, Tencent's management stated that in 2025, Tencent plans to further increase capital expenditure, which is expected to account for a low double-digit percentage of revenue, with projections for capital expenditure in 2025 being 80-100 billion yuan.

In reality, for the fourth quarter of 2024, Tencent's capital expenditure related to AI amounted to 39 billion yuan, making the growth guidance for 2025 appear quite limited in comparison.

Earlier, China United Network Communications, which disclosed its Earnings Reports, anticipated that its fixed asset investment would be around 55 billion yuan this year, with computing power investment expected to grow by 28% year-on-year.

However, on a quarter-on-quarter basis, China United Network Communications' total capital expenditure for 2024 is projected to be 61.37 billion yuan, a year-on-year decrease of 17%. According to the guidance, capital investment by China United Network Communications may decline overall this year.

Notably, on March 19, both Tencent and China United Network Communications saw a significant decline of over 3% the day after releasing their results. As of the time of this report, the stock prices of both companies remain below the closing price on the day of the Earnings Reports.

Comprehensive analysis indicates that the strength of AI investments is below expectations, increased industry competition, and policy changes are also potential risk points in the computing power industry frequently mentioned by many Institutions recently.

However, there still exists considerable disagreement in the market regarding whether there is actually a surplus or shortage of computing power. Relatively speaking, while the stock market has its ups and downs, the industry perspective appears to be more rational.

On one hand, breakthroughs and innovations in large model technology are indeed continuously optimizing the efficiency of computing power usage. On the other hand, some viewpoints suggest that as efficiency improves and costs decrease, the expansion of AI applications will also drive greater demand for computing power.

On March 17, the international investment bank Jefferies released a deep research report on China's AI capital expenditure and cloud growth, predicting that by 2025, China's AI capital expenditure will exceed 500 billion yuan; by 2030, China's AI capital expenditure will reach 1 trillion yuan, at which point computing power will achieve a fivefold increase and cloud revenue will triple.

Foxconn Chairman Liu Yangwei also stated at this week's 2025 GTC conference that AI will become the most important operation driver for Foxconn this year. When asked whether Foxconn's server revenue is expected to surpass the iPhone business, he replied, "It will definitely happen within two years," and it might even be achieved as early as this year. This also indicates that the current demand for computing power infrastructure remains strong.

$CHINA MOBILE (00941.HK)$At the 2024 annual performance briefing, it was stated that in 2025, the company plans to invest 37.3 billion yuan in the field of computing power, increasing its proportion of capital expenditures to 25%, and investments in inference resources will be made based on market demand, with no upper limit set.

Tianfeng Securities pointed out in a previous report that global datacenter investment will reach 57 billion USD in 2024, with AI demand as the main driving force, and computing power may be limited by the balance of energy and technological innovation. The current demand for computing power remains strong, and the entire industry is facing a vacancy rate dropping below 3%.

Tianfeng Securities believes that the current major contradictions in the industry are reflected in the mismatch between computing power supply and demand, such as the AI industry's demand for chips and intelligent computing centers, which differs somewhat from the construction direction of previous years, requiring adjustments and transformations.

Editor/Rocky