
The Zhitong Finance App learned that CITIC Construction Investment released a research report saying that overseas computing power investment continued to expand, focusing on demand continuity and supply delivery for 27 years. It is estimated that the capital expenses of the five major overseas CSPs in '27 reached 1.12 trillion yuan and 1.38 trillion US dollars respectively under conservative and optimistic scenarios, corresponding year-on-year increases of 38% and 70%, respectively; leading cloud companies have sufficient stock resources, and the pressure on delivery and financing for emerging cloud factories is relatively greater. Revenue from leading CSPs remains high, but the growth rate of Google (GOOGL.US) and Microsoft (MSFT.US) orders is slowing down, and it is necessary to track new contracts and actual usage. Google, Amazon (AMZN.US), and Meta (META.US) have a total procurement commitment of 1.63 trillion US dollars, and the intention to build in the medium to long term is still strong. AI investment continues to drag down cash flow. The free cash flow of Google, Amazon, and Meta has all turned negative. The continued expansion of AI investment depends more on the traditional main business hematopoiesis; Microsoft's office and Meta social services are relatively stable. Follow-up attention will be paid to outsourced sales of self-developed chips, changes in MaaS volume price and cost, and monetization of AI search and agent entry value.
CITIC Construction Investment's main views are as follows:
Overseas computing power investment is still expanding, and the focus has been on demand continuity and supply delivery for 27 years.
The total capital expenditure of the five major CSPs (Microsoft, Amazon, Google, Meta, and Oracle) in '25 was about US$412.3 billion, up 75% year on year; the total capital expenditure of the top five CSP's in the first three quarters of '26 was about US$560 billion, up 98% year on year. According to estimates, under a 27-year conservative and optimistic scenario, the total investment of the five companies was about 1.12 trillion yuan and 1.38 trillion US dollars, respectively, corresponding to a year-on-year increase of about 38% and 70%.
Ongoing orders from leading CSPs are still supported, but the growth slope has become the core observation point for demand; in terms of expenditure, procurement commitments continue to expand, and the desire to build medium- to long-term computing power is still strong.
In 26Q3, Google Cloud, Azure, and AWS revenue increased 103%, 48%, and 39% year-on-year respectively. The launch of new computing power and improved efficiency of existing resources supported the growth. On the profit side, the pace of increasing profit margins of Google Cloud and AWS tends to moderate, and new depreciation and operating costs will gradually be reflected. However, the year-on-year growth rate of existing orders from Google and Microsoft has slowed down, and large contracts are more concentrated on leading model manufacturers, so we need to pay attention to new contracts and actual customer usage.
In terms of future expenses, major manufacturers are still locking in equipment, energy, and related services ahead of schedule. Google, Amazon and Meta procurement commitments reached $924.4 billion, $254.4 billion, and US$454.5 billion respectively in 26Q3, up 887%, 249%, and 460% year-on-year respectively, for a total of about US$1.63 trillion.
Free cash flow has weakened overall, and AI investment pressure has increased the requirements for traditional main businesses.
As infrastructure investment expanded, Amazon, Google, and Meta's free cash flow profit margins turned negative. As of 26Q3, Microsoft alone remained positive. Mature businesses are an important source of capital for large companies to continue to invest in AI, and each major business is affected differently.
Google and Amazon are more directly exposed to user search and shopping decisions: Google needs to pay attention to the dilution of search ad revenue due to the lower level of AI mode, and Amazon needs to observe whether advertising growth can resume after the impact of promotions subsides across seasons.
Microsoft M365 commercial cloud revenue increased by 18% and seats increased by 5%. High-end packages and AI commercialization support units increased customer revenue, and rigid office scenarios showed resilience; Meta social relationships and content consumption were difficult to be directly replaced by generic models, and AI recommendations and marketing optimization also helped to monetize ads.
Subsequent industry variables: Focus on external sales volume of CSP self-developed chips, MaaS volume and price changes, and long-term import value.
Track TPU and Trainium external customer expansion and Maia and MTIA internal computing power cost reduction;
Can AI search, Copilot, and Muse rely on entry advantages to generate new revenue; focus on demand and commercialization space brought by agents; and observe whether model price reductions can be digested by increased usage and reduced costs.
Risk Alerts
Risk of demand for AI computing power falling short of expectations, risk of concentration of leading customers and order fulfillment, risk of limited chip and server supply, risk of delayed power access and data center delivery, risk of falling utilization of new computing power than expected, risk of increasing competition for cloud services and falling prices, risk of equipment depreciation and rising operating costs, risk of slowing advertising demand and shifting user traffic to third-party AI platforms, risk of falling short of expectations in commercialization of AI products, risk of falling short of expectations on return on capital expenditure, risk of customer repayment delays and centralized payment of procurement and lease payments, capital gaps exceeding expectations and the risk of rising financing costs.