Behind the sharp decline of the computing power industry chain, why did Microsoft cut capital expenditures first?

This article is sourced from: BayesCrest

Recently, there has been an increasing amount of news regarding the oversupply of data center construction!

Whether it's Alibaba's Joseph Tsai warning about excessive construction of data centers in Europe and the United States, or TD Cowen revealing that Microsoft has abandoned any new data center projects in the U.S. and European markets with a designed power consumption of 2 GW, it indicates that adjustments regarding AI capital expenditures (capex) are currently underway! These messages have also triggered a series of adjustments in the computing power industry chain.

Why is Microsoft the first to cut capex?

As the largest supporting cloud CSP for OpenAI, Microsoft's capital expenditure (CapEx) growth rate is among the highest in the second half of 2024 to early 2025:

Microsoft Azure has recently undertaken large-scale AI computing power demands (including internal and partner AI training and inference), necessitating the rapid construction of high-density data centers to support the rapid growth of AI-related businesses. Microsoft has repeatedly mentioned in its financial reports that it will continue to increase investments in data centers, servers, and network infrastructure to promote the development of cloud and AI services.

Since 2023, Microsoft has been closely collaborating with OpenAI, and the training and deployment of related AI models require massive GPU/TPU clusters and supporting data centers. Although the report does not detail the specific computing power requests from OpenAI to Microsoft, from other public information and statements from Microsoft's management, it can be seen that the infrastructure construction to support large-scale model training and inference has become one of the important driving forces behind Microsoft's recent CapEx increase.

The report indicates that the CapEx listed by Microsoft in its financial report includes a portion of investments in the form of "financing leases"; this accounting treatment will significantly inflate the reported capital expenditure scale in the short term.

From the report and industry news, it appears that after making significant investments, Microsoft has begun to conduct more refined evaluations and adjustments of leases and new data centers over the past few quarters, including the cancellation or postponement of some third-party data center leases in the North American and European markets. The reasons for this phenomenon may include the following main points:

① Phase mismatch between supply and demand

After rapidly expanding to support AI training demands, Microsoft has found that there is a certain risk of "overcapacity" or "idle" capacity in some regions. Therefore, when demand enters a relatively stable or transitional period, it will reassess external leasing contracts that have not yet been finalized or fully signed. This does not mean that Microsoft’s long-term demand judgment for cloud and AI has fundamentally changed, but rather leans towards "activating existing resources and controlling the pace of short-term investments."

② Changes in internal projects and collaboration needs

If Microsoft itself or OpenAI experiences a slowdown in training demands at certain stages, or switches to other more efficient hardware/computing platforms, it may also lead to short-term adjustments in data center capacity needs. The report mentions that Microsoft is currently focusing more on the deployment rhythm for the medium and long term, and some ultra-large-scale construction projects (especially those relying on leasing third-party data centers) have been canceled or postponed before signing, which can also be seen as an adjustment of the company's internal project/collaboration strategy The ROIC (Return on Invested Capital) of major manufacturers showed a significant rebound in the fourth quarter of 2024; for Microsoft, maintaining a relatively healthy ROIC level is equally important. With a surge in capital investment, if marginal returns and ROIC face temporary pressure, management will prudently assess new capacity expansion plans to ensure overall capital utilization efficiency.

Google and Meta have "replenished" or "received" some of the space released by Microsoft in international markets and the U.S. market; OpenAI may also build more in-house or collaborate with multiple third parties. Besides Microsoft, Google and Meta are also actively expanding their data centers. The current industry landscape is changing: in this context, Microsoft's short-term slowdown in data center expansion can also be seen as a normal behavior of resource integration and strategic reassessment.

Comparison with Other Cloud Providers: The Industry Significance of Microsoft's CapEx Adjustments

While the overall high investment trend of cloud providers remains unchanged, the capital expenditures of giants like Amazon, Google, and Meta are significantly increasing in Q4 2024 and subsequent years, indicating that the cloud computing and AI markets remain prosperous in the medium to long term. Microsoft's "contraction" is more of a temporary strategic adjustment rather than a signal of the industry entering a downturn or turning point.

Different companies have varying rhythms and strategies; Microsoft's CapEx surged significantly in the early stages, and the current pullback is more pronounced; whereas Meta and Google experienced a slowdown in investment growth over the past year or two, but have rebounded quickly now. Each company is facing a misalignment in the rhythm of "expansion—adjustment—re-expansion" in response to AI and cloud business demands, but the overall trend is that total investment continues to increase.

Comparing Microsoft, Meta, Google, and Amazon: Insights into CapEx Rhythm for Q4 2024 and 2025

The table below provides a brief comparison of the capital expenditure situations of the four major manufacturers based on the report's data for the fourth quarter of 2024 (calendar year basis, same below) and partial guidance for 2025. Note: Microsoft's "Q4 2024" corresponds to its FY25 Q2 fiscal quarter.

From the table above, we can see:

Microsoft: The year-on-year growth rate of capital expenditure in Q4 2024 is still very high (+96.52%), but recently, there has been some "initial" contraction or lease adjustments in the deployment rhythm of data centers to pursue better resource utilization and ROIC.

Meta: The year-on-year increase in CapEx in Q4 2024 is the highest (+116.8%), and it has clearly stated that it will continue to invest heavily in 2025 to support generative AI and core social/advertising business needs. The significant expansion of CapEx basically started to lag behind Microsoft by about four quarters from Q4 2024

Google: maintains a relatively stable expansion pace (year-on-year +28.98%), but the absolute amount remains large, and further increases investment in AI (Google DeepMind) and cloud platform (GCP) in 2025.

Considering the actual market situation, Microsoft's significant increase in CapEx is a common practice to seize the AI opportunity, while the cancellation or postponement of some idle or not yet officially launched data center leases is also a relatively rational investment adjustment.

From the perspective of capital expenditure (CapEx) rhythm, Microsoft and Amazon significantly increased their investments in data centers, servers, and other infrastructure in mid-2023 (or earlier), and continued to rise in subsequent quarters; while Google and Meta also entered an "acceleration period" starting in 2024, their overall startup time was about four quarters later than Microsoft and Amazon.

Microsoft:

In Q3 2023 (natural quarter), CapEx began to show a significant year-on-year increase, remaining in a high growth range in the following quarters; in Q4 2024 (corresponding to Microsoft's FY25Q2), it still achieved a year-on-year increase of +96.52%. In January 2025, the company's management clearly stated that Microsoft is expected to invest $80 billion in FY25 to build AI data centers, with the total scale of CapEx remaining high, but due to the high base in the previous period, the year-on-year growth rate in Q1 2025 may slow down.

Microsoft has already initiated large-scale investments related to "AI + cloud" in 2023, and there are currently some "micro-adjustments" in leases or construction in certain regions, but the absolute expenditure amount remains large.

Amazon

In Q2 2023, CapEx fell to a relative low of about $10.4 billion for a single quarter, and then rapidly climbed in Q3-Q4 2023; reaching $26.3 billion in Q4 2024 (year-on-year +97.74%), setting a recent high. Due to the simultaneous expansion of AWS cloud business and AI demand, Amazon has increased its layout of technological infrastructure in both North America and international markets; at the same time, starting from January 2025, the company has shortened the lifespan of some servers from 6 years to 5 years, which also reflects adjustments in hardware update rhythm and financial depreciation.

Similar to Microsoft, Amazon began to rapidly increase investments in the second half of 2023, with both quarter-on-quarter and year-on-year growth rates remaining high in Q4 2024, and it is expected to maintain a high level in 2025.

Google

In Q4 2023, Google's CapEx significantly increased both year-on-year and quarter-on-quarter, maintaining a relatively high level in subsequent quarters; however, by Q3 2024, its year-on-year growth rate showed a certain decline; the latest investment in Q4 2024 was $14.276 billion (year-on-year +28.98%) From the rhythm, Google's investment priority in servers (especially AI training/inference clusters) and data centers is slightly lagging behind Microsoft and Amazon, but with the support of "Google Cloud + Google DeepMind," it has strong momentum.

Capital expenditures in 2025 may reach $75 billion, with the first quarter expected to be between $16 billion and $18 billion.

Google's investment began to accelerate from the end of 2023 to early 2024, lagging behind Microsoft and Amazon by one to several quarters, but the incremental portion is mainly focused on AI infrastructure (DeepMind, GCP, etc.), and the scale remains substantial.

Meta

During the period from 2021 to 2022, Meta's CapEx fluctuated significantly, peaking at $9.52 billion in Q3 2022; it then declined. Entering 2024, the company is experiencing a rapid increase in CapEx again due to the demand for large models (generative AI) and core social business; in Q4 2024, it invested $14.84 billion (up 116.8% year-on-year), reaching a new high in nearly three to four years. The forecast for 2025 is between $60 billion and $65 billion, a significant increase compared to $39.23 billion in 2024.

Meta is entering a new round of large expenditures in 2024, slightly later than Microsoft and Amazon, similar to Google's "catch-up" rhythm in the AI field.

Therefore, Microsoft's cancellation or postponement of third-party data center leases is a phased phenomenon based on multiple considerations and does not indicate that demand in the cloud computing or AI industry has peaked. In the long-term trend, AI-driven cloud computing demand continues to drive investment in data center upgrades and construction, with Microsoft, Google, Meta, Amazon, and OpenAI all continuously investing in ultra-large-scale computing infrastructure.

Thus, Microsoft's initial "contraction" does not mean an overall industry downturn; on the contrary, Meta and Google are still maintaining rapid expansion at this stage; the overall industry's high capital expenditure trend driven by AI and cloud demand has not changed.

The phased "contraction" is behind supply and demand and financial considerations; Microsoft accelerated its expansion earlier during the AI training boom; when short-term demand rhythms or hardware selections change, it needs to adjust the speed of external leasing or new construction to maintain healthy ROIC.

In the medium to long term, high investment will continue, with the 2025 CapEx guidance (or external forecasts) for the three giants remaining high. As AI continues to drive cloud computing revenue, investments in high-density data centers and servers remain a core strategic direction for all companies