CITIC SEC: Microsoft proposes investing 80 billion US dollars in AI infrastructure, a new round of domestic and international computing power "arms race" is about to begin

CITIC SEC released a research report stating that Microsoft proposed an $80 billion investment in AI infrastructure by early 2025, which is expected to become an important event catalyst, triggering a new round of computational power "arms race" among top cloud service providers (CSPs) in North America. At the same time, the AI industry has strong network effects and economies of scale. When leading AI applications gain a user advantage, their competitive advantages in terms of model accuracy, marginal costs, and user stickiness will gradually strengthen, expanding their advantages globally. It is recommended to continue to focus on two investment themes: (1) domestic AIDC industry chain combination, and (2) combination of top CSPs' AI application ecosystem. On January 3, 2025, Microsoft posted a blog on its official website, where Brad Smith (Vice Chair & President) noted that a new generation of AI enterprises is emerging and each company is benefiting from rapid progress in AI large models and chips. Infrastructure is transitioning from GPUs to ASICs with tensor computations. To continue promoting AI innovation and applications, Microsoft plans to invest approximately $80 billion in the 2025 fiscal year to build AIDC (with over 50% of the investment in the US) to train AI large models globally and deploy AI or cloud-based applications. The main points of CITIC SEC are as follows: Reviewing the first round of North American "computational arms race" in 2023, Microsoft fired the first shot in Q1 23. The bank found that Microsoft increased capital spending in Q1 23 ahead of other top CSPs in North America following suit gradually. According to Refinitiv data, Microsoft has continuously increased its capital spending since Q1 23, with a particularly astonishing growth rate of 35% in Q2 23 compared to Q1. As of Q3 24, the investment has maintained a quarterly increasing trend, but the average growth rate has significantly slowed down to below 20%; other North American top CSPs like Amazon, Google, and Meta have followed Microsoft in increasing their capital spending, and sooner or later, set records of a 30% increase in a single quarter. After OpenAI, backed by Microsoft, rolled out popular AI large models, other companies similar in business scale or type to Microsoft began increasing capital investment and R&D efforts to compete fiercely in the AI large model field with Microsoft and OpenAI, initiating an AI computational arms race starting in 2023. This event reflects that the "computational arms race" is still ongoing, potentially dispelling market concerns over the marginal slowdown in AI capital spending. Considering that OpenAI is still a leader in the AI wave, AI inference optimization is gradually becoming commercially viable. Microsoft will focus more on using its leading AI capabilities to help Azure expand its differentiation competitive advantage, thereby increasing the market share and profitability of its cloud business. Therefore, Microsoft maintains a relatively aggressive stance on increasing investment in AI infrastructure construction. According to Omida forecast data, in 2024 Microsoft spent a total of $31 billion on AI servers, with the bank estimating AI servers accounting for about 70% of AIDC costs. If Microsoft's AI infrastructure construction investment in 2025 is increased to $80 billion, it is expected to grow by over 75% year-on-year. Referring to the 2023 North American computational arms race, if Microsoft significantly increases its capital spending, it could lead to giant companies like Google, Amazon, Meta, among others, making defensive AI infrastructure investment expansions to maintain their AI service competitiveness. The bank is optimistic that there is still room for a 50% or more year-on-year increase in AI computational spending in 2025, which may help dispel market concerns over the marginal slowdown in capital spending. Future theme catalysis one: Sino-US competition accelerates global AI infrastructure construction. In 2024, the gap between the penetration rates of domestic AI large models and overseas markets has noticeably narrowed. According to AI Product Rankings data, Duobao's monthly active users reached approximately 60 million, ranking second globally in AI applications and first domestically, with a growth rate 5.5 percentage points higher than ChatGPT. At the same time, Microsoft mentioned in the blog that the rapid development of the Chinese AI industry has intensified Sino-US competition, especially set to play out on a global scale over the next four years. Given the potential network effects, competition between China and the US in the AI industry for international influence will largely depend on first-mover advantages. Microsoft therefore suggests that the US government quickly support the American AI industry in seizing market opportunities globally. Therefore, the competition between the Chinese and American AI industries to gain international market influence may further accelerate the pace of AI infrastructure construction globally. Subsequently, it is important to continue monitoring the pace of AI infrastructure construction by Chinese and American top companies in regions like Southeast Asia and Europe. The future plans of Chinese and American top companies in AI computational infrastructure development in these regions could become important event catalysts, bringing investment opportunities in energy, infrastructure, and computational power industry chains. Future theme catalysis two: Duobao going mainstream or initiating a domestic computational "arms race." According to the annual report information from various companies, the capital spending of domestic top internet companies in 2023 was still below 10%. As the Duobao AI large model ecosystem needs to expand, the AI services launched for various consumer and business applications may widen ByteDance's ecological moat. Other top domestic companies may increase their AI investment efforts to counterattack based on their own ecological advantages. Following OpenAI's launch of ChatGPT, North American top CSPs began to follow Microsoft in increasing their capital spending to revenue ratio (Q3 24 North American top CSPs had already increased the capital spending to revenue ratio to above 15%), the domestic computational "arms race" is expected to commence in 2025. Top domestic internet companies and top smartphone manufacturers still have considerable room for AI computational investment. It is optimistically estimated that the combined investment scale in 2025 will be close to $50 billion. Therefore, it is crucial to focus on the disclosure timing of various top companies' performances in 2025, which may reflect on the domestic computational industry chain, thereby paying attention to investment opportunities exceeding expectations in the industry chain. Risk factors: AI technological development does not meet expectations; global AI infrastructure construction pace does not meet expectations; traffic growth does not meet expectations; capital spending by cloud providers and operators does not meet expectations; technological progress of domestic computational industry chains does not meet expectations; geopolitical risks.