Guangda Overseas: Microsoft pauses AI data center construction. How to view the impact of tariffs on US cloud vendors?

EB SECURITIES' overseas research team released a research report stating that in the short term, Trump's tariff policies may put pressure on the performance of US technology companies, mainly reflected in the increase in data center construction costs, potential risks of retaliatory tariffs on digital services, more cautious IT spending and cloud budgeting by downstream customers. It is expected that this pessimistic outlook will lead to a conservative guidance for US tech companies' 25Q2 performance and put pressure on valuations. EB SECURITIES' overseas research team stated that in the long term, Trump's potential tax cuts for US companies may benefit corporate profits, with strong certainty of increasing demand for AI inference endpoint loads. Currently, US tech companies such as Alphabet Inc. Class C, Amazon.com, Inc., and Meta have not shown intentions to reduce capital expenditures and slow down AI investments. Microsoft Corporation's temporary halt to early data center projects is a strategic adjustment made against the backdrop of aggressive investments in 24Q, suggesting a focus on the recovery of earnings and valuation potential in the US tech sector in the long term. NVIDIA Corporation (NVDA.US), Meta (META.US), and Taiwan Semiconductor Manufacturing Co., Ltd. Sponsored ADR (TSM.US) are recommended to be watched closely. Microsoft Corporation's decision to slow down or halt some data center construction projects is a strategic adjustment made against the backdrop of aggressive investments in 24Q. In April 2025, Microsoft Corporation suspended a $1 billion data center project in Ohio. On April 10, the President of Microsoft Corporation's cloud computing business stated that the investment strategy for AI infrastructure requires "agility and precision." Microsoft Corporation is strategically adjusting its investment pace by slowing down or halting some early projects. During the 24Q3 earnings call, Microsoft Corporation mentioned that capital expenditures in 25Q are expected to increase sequentially, with a 13% increase in capital expenditures in 24Q4, exceeding market expectations. Microsoft Corporation's cash capital expenditures as a percentage of operating cash flow in 24Q1-24Q4 were 34.3%, 37.3%, 43.7%, and 70.9%, showing a rapid increase quarter by quarter. While there may be greater uncertainty in the global macro environment and AI algorithms and hardware iterations in 2025, the certainty of increasing demand for AI inference loads is strong, and it is expected that Microsoft Corporation will continue its trend of expanding capital expenditures. Additionally, Microsoft Corporation mentioned a potential shift of focus back to traditional cloud services from AI and non-AI business areas. In the 24Q4 earnings call, Microsoft Corporation stated that due to the need for sales teams to balance between AI and non-AI businesses, Azure AI revenues exceeded expectations, but Azure non-AI business revenues were slightly lower than expected. Furthermore, due to continuing increases in capital expenditures to support the expansion of AI infrastructure, Microsoft Corporation's cloud business gross margin in 24Q4 was 70%, a 2 percentage point decline year on year, and the company expects the cloud business gross margin in 25Q1 to continue to decline to around 69%. Considering that AI cloud businesses have relatively high costs and low profit margins, Microsoft Corporation may refocus its strategic priorities on traditional cloud services to address the pressures of the macro environment. Alphabet Inc. Class C reaffirmed its $75 billion capital expenditure plan for 2025 in April 25, highlighting the huge opportunities in the AI field. On April 9, 2025, the CEO of Alphabet Inc. Class C emphasized at the Google Cloud Next event that the company will invest $75 billion in 2025 to improve chips and servers needed for core products such as search and advertising, while also supporting the development of Gemini models and AI product services, consistent with the guidance provided in the 24Q4 earnings call (estimated capital expenditures of $75 billion in 2025, a 43% year-on-year increase, significantly higher than the consensus expectation of 26.9%). Alphabet Inc. Class C's cash capital expenditures as a percentage of operating cash flow in 24Q1-24Q4 were 41.6%, 49.5%, 42.5%, and 36.5%, with capital expenditures putting relatively small pressure on free cash flow. Amazon.com, Inc. also emphasized the importance of R&D spending and capital expenditures in the AI field in April 25, without specifying specific amounts. According to Reuters, on April 10, 2025, the CEO of Amazon.com, Inc. emphasized continued commitment to AI investments to support cloud services, AI features for e-commerce, and the Alexa AI voice assistant. In the 24Q4 earnings call, Amazon.com, Inc. mentioned that capital expenditures in 2025 are expected to reach $105.2 billion, a 34.5% year-on-year increase. Amazon.com, Inc.'s cash capital expenditures as a percentage of operating cash flow in 24Q1-24Q4 were 78.6%, 69.7%, 87.1%, and 57.1%, with a more aggressive capital expenditure strategy compared to Alphabet Inc. Class C. Meta has given positive guidance on capital expenditures in 2025, announcing the construction of a new data center cluster and reinforcing its AI strategy. According to Bloomberg's report on April 5, 2025, Meta plans to invest $1 billion in building an AI data center in Wisconsin.Supporting the research and development of the Llama series models and Meta AI inference loads, as well as optimizing advertising recommendation algorithms and short video content sorting algorithms. In the Meta performance meeting in 24Q4, it was mentioned that capital expenditures in 2025 will reach 60-65 billion US dollars, an increase of 52.9%-65.7% year-on-year. The company will continue to optimize capital allocation efficiency, including deploying custom ASIC chips to reduce computing costs, and extending the life of servers and network equipment, etc. In 24Q1-24Q4, the proportion of Meta's cash capital expenditures to operating cash flow were 33.3%, 42.2%, 33.4%, 51.5% respectively. This ratio saw a more pronounced increase in 24Q4.Trump's tariff policy has brought more uncertainty to the return on investment in AI data centers, the global macroeconomic environment, and financing conditions. Since Trump proposed his aggressive tariff policy on April 2nd, the global macroeconomy and technology industry chain face greater uncertainty. The AI data center industry chain covers multiple regions such as the United States, Mexico, mainland China, Taiwan, and Southeast Asia, and its investment return cycle will be widely and complexly affected, potentially becoming a catalyst for strategic adjustment of capital spending by US technology companies. 1) GPU Servers: The Trump administration's tariff exclusion list does not include GPU servers, meaning GPUs exported from Taiwan to the United States are subject to a 32% tariff. However, according to USMCA, digital processing units exported from Canada and Mexico, regardless of their origin, are eligible for tariff exemptions, allowing US customers to bypass the tariff on GPU servers by routing them through Mexico. 2) Data Center Infrastructure: According to SemiAnalysis calculations, Trump's tariff policy may lead to high single-digit growth in data center infrastructure costs, with significant impacts on wafer manufacturing and optical module production. Regarding wafer manufacturing, information disclosed by US construction company Exyte shows that the cost of building wafer fabs in the US is approximately twice that of Taiwan. Taiwan Semiconductor Manufacturing Co., Ltd. founder Zhang Zhongmou indicated that the cost of chip production in the US is 50% higher than in Taiwan, with wafer manufacturing equipment heavily reliant on imports from companies like ASML and TEL. As for optical modules, it is predicted by the China Commerce Industry Research Institute that by 2024, over 80% of global optical module production capacity will be located in mainland China. On the supply side, due to the longer investment return cycle of data centers and the cost pressures brought by tariffs, US technology companies may withdraw some investments and pass on the costs to customers through price increases. According to Microsoft Corporation's 2Q4 earnings conference call, the company plans to monetize investments in AI data centers over a period of 15 years or more. For FY25Q2 (24M12-25M2), Micron's earnings call mentioned that the company expected to pass on the impact of tariffs to customers. Micron began charging additional fees for memory modules and SSDs starting on April 9th, 2025, in response to the cost challenges of tariffs. Cloud providers may also continue to pass on cost pressures downward. On the demand side, downstream customers face broader economic uncertainties, leading to conservative IT spending and budgeting for cloud services in the short term. The increased cost of cloud services due to tariffs may weaken the competitiveness of large cloud providers and prompt tech giants to be more cautious in their capital spending plans. For example, in an interview with DCD in 25M4, data services company StarZData shifted its cloud workloads to cheaper GCP and OVHcloud after noticing rapid price increases on AWS. Since Microsoft Azure primarily serves large customers, its average prices are around 20% higher than AWS. If major cloud providers pass on the cost pressures from tariffs to customers, it could lead to a more fragmented public cloud market, making capital spending plans for tech giants like Microsoft, Amazon, and Alphabet more cautious. Although most areas of digital services are not affected by tariffs, the EU may levy counter-tariffs on digital services or strengthen content regulation, adding pressure to overall revenue for US tech companies. According to the Financial Times on April 14th, the EU Commission President is considering taxing digital advertising for tech giants such as Alphabet and Meta in response to Trump's tariff policy. Constraints in the EU against imported digital services include: 1) Digital Markets Act (DMA): officially implemented in 24M7, the EU's anti-monopoly measures against tech giants aim to prevent predatory practices; 2) Digital Services Act (DSA): effective as of 22M11, emphasizing strengthened platform content moderation; 3) Digital Services Tax (DST): France began taxing digital services in 19M1, causing significant impacts on US internet companies like Alphabet and Facebook.