CITIC Construction Investment: AI computing power investment and commercialization are linked and continue to be optimistic about cybersecurity, domestic GPUs, and SaaS

The Zhitong Finance App learned that CITIC Construction Investment released a research report saying that in 26Q3, the cash capital expenditure of the four major US manufacturers increased 108.0% year on year, and operating cash flow increased 10.4%; based on operating cash flow reduced cash capital expenses, the total free cash flow fell to -25.04 billion US dollars. Microsoft's operating cash flow still covers current investment, and the remaining three companies have negative free cash flow. The combined revenue of AWS, Google Cloud, and Azure increased 55.7% year over year, Copilot usage, model calls, and agent deployment increased, and SAP and ServiceNow promoted AI commercialization. Computing power infrastructure investment is linked to enterprise AI applications. Servers and storage networks, domestic chips and supernodes, and enterprise data governance and professional workflow applications form an important undertaking direction for industrial investment. The bank continues to be optimistic about cybersecurity, domestic GPUs, and SaaS companies that have been the first to release results.

CITIC Construction Investment's main views are as follows:

Major North American manufacturers increase investment in computing power, and cloud revenue and application consumption are growing at the same time

The growth rate of cloud revenue continues to rise, and demand for AI is driving computing power consumption. In 26Q3, the revenue of AWS, Google Cloud and Azure was US$459.8, 307.1, and 33.17 billion US dollars, respectively, up 39.3%, 102.6% and 48.2% year over year; the total revenue of the three companies was US$109.86 billion, up 55.7% year on year and 13.9% month on month. The total year-on-year growth rate increased further from 47.9% in Q2. Enterprise AI infrastructure, model inference services, and core cloud businesses all contributed to the growth, and Google Cloud also benefited from TPU system sales. Model calls and increased agent deployment in production environments drive demand for GPUs, CPUs, and data processing, and delivery of additional capacity drives the transformation of demand into revenue.

Cash capital expenditure grew rapidly, and infrastructure investment remained high. In 26Q3, the total cash capital expenditure of Amazon (AMZN.US), Google, Microsoft and Meta was US$20.54 billion, up 108.0% year on year and 22.3% month on month. The year-on-year growth rate continued to increase from 87.5% in Q2. Among them, the four companies' cash capital expenses were 758.8, 513.6, 393.7, and 33.94 billion US dollars, respectively, up 121.7%, 114.4%, 103.0% and 80.3%, respectively. The investment covers CPU, GPU, storage networks and data centers, which not only supports cutting-edge model training, but also meets model inference, agent execution, and enterprise data processing requirements. Cloud business growth and long-term contract reserves provide a demand base for capacity expansion, and demand for server, storage, network equipment and data center construction continues.

Capital expenditure is faster than operating cash flow, and each manufacturer's free cash flow is divided. In 26Q3, the total operating cash flow of the four companies was US$175.50 billion, up 10.4% year on year, lower than the growth rate of cash capital expenditure during the same period. Based on a unified estimate of operating cash flow minus cash capital expenditure, the total free cash flow of the four companies fell from US$7.84 billion in Q2 to US$25.04 billion in Q3. Among them, Microsoft's free cash flow was US$18.63 billion, and operating cash flow can still cover current cash capital expenses; Amazon, Google, and Meta were -334.6 billion US dollars, -90.7, and -1.16 billion US dollars, respectively. Infrastructure procurement and construction expenses are ahead, and revenue and cash returns are gradually realized as production capacity is invested, customer transfers, and contract execution. High-intensity investment puts pressure on current cash balances.

Meta (META.US): The sharp rise in advertising volume and price continues, and the adoption rate of generative AI tools has increased. On October 21, Meta disclosed 26Q3 results, achieving revenue of US$64.60 billion, up 26% year on year, slightly down from the 28% growth rate in the previous quarter; advertising revenue was US$63.06 billion, up about 26% year on year, and advertising impressions and average unit price increased by 10% and 14% respectively. At the product level, the number of advertisers using generative AI advertising creative tools increased from over 9 million in the previous quarter to close to 10 million, and the number of advertisers using video generation tools increased 25% over the previous quarter. The company estimates that advertisers who have adopted the Meta AI Business Assistant's suggestions have nearly 5% higher conversion volume and 9% higher advertising investment than similar non-adopters when the cost of a single conversion is the same. AI improves advertisers' returns through creative production, ad matching, and delivery optimization, and drives core business growth.

Computing power investment increased month-on-month, and the lower limit of annual capital expenditure was raised. In 26Q3, Meta's capital expenditure including financial lease principal payments was US$34.91 billion, up 12.3% from US$31.08 billion in Q2; the full-year guidance narrowed from US$130 billion to US$145 billion to US$145 billion. Operating cash flow for the same period was US$32.79 billion, and free cash flow fell from US$780 million in Q2 to US$2.12 billion. The company uses computing power resources for recommended advertising, model training, and new product operation. The core advertising business provides current revenue support, and new applications such as Muse increase subsequent inference requirements.

Microsoft (MSFT.US): Azure growth rate increased, Copilot usage and pay-as-you-go expansion. On October 28, Microsoft disclosed FY27Q1 results, corresponding to the natural year 26Q3, achieving revenue of US$92.28 billion, an increase of 19% over the previous year; operating profit of US$45.92 billion, an increase of 21% over the previous year. Microsoft Cloud revenue was $63.9 billion, up 30% year over year, up 3 percentage points from the previous quarter; Azure's revenue growth rate increased from 42% to 48%. On the application side, Copilot's daily active users increased by nearly 50% month-on-month, and the number of automated tasks nearly doubled compared to July; the number of Copilot customers using pay-as-you-go billing increased by more than 75% month-on-month. Corporate procurement extends from seat subscriptions to actual task execution, and increased usage intensity increases the source of ARPU growth.

Contract reserves remain high, and the pace of computing power delivery affects current investment. The commercial RPO at the end of the quarter was $677 billion, up 73% year over year, basically the same as the previous quarter's $678 billion, of which about 30% is expected to confirm revenue over the next 12 months. Capital expenditure, including confirmed financial leasing assets, was 45.2 billion US dollars in the current quarter, compared to 41 billion US dollars in the previous quarter. About three-quarters were invested in short-term assets such as CPUs and GPUs. Affected by delays in the delivery of computing power from some third parties, the company adjusted the 2026 natural year capital expenditure forecast from about US$175 billion to about US$170 billion, and moved related deliveries to the second half of the fiscal year. At the same time, Azure's fixed exchange rate revenue is expected to increase 49%-50% in the next quarter, and customer demand is still higher than available supply.

Google (GOOGL.US): The growth rate of cloud revenue exceeded 100%, and the number of model calls increased month-on-month. On October 28, Alphabet disclosed 26Q3 results, achieving revenue of US$127.58 billion, an increase of 25% over the previous year. Google Cloud's revenue was US$30.71 billion, up 103% year over year, compared to 82% in the previous quarter; operating profit was US$11.12 billion, and operating margin increased to 36.2% from 35.6% in the previous quarter. The growth comes from sales of enterprise AI infrastructure, AI solutions, core GCP services, and TPU systems. In terms of usage, the first-party model API processed nearly 32 billion tokens per minute, an increase of 40% over the previous year, reaching more than four times that of the same period last year. Model capability updates and increased enterprise calls together drive cloud service consumption.

Cloud order reserves continue to grow, and infrastructure construction and third-party procurement go hand in hand. Google Cloud Backlog increased from $513.9 billion to $561 billion in the previous quarter, an increase of about 9.2% over the previous quarter. The company expects to confirm slightly more than 55% of this revenue over the next 24 months. The quarterly capital expenditure was $51.36 billion, up 14.3% from month to month. Technology infrastructure investment was about 65% for servers and network equipment, and 35% for data centers. The company maintains the annual capital expenditure guideline of US$1950 billion to US$205 billion and is expected to grow significantly in 2027. In order to ease supply constraints, Google began using third-party transition computing power at the end of Q3. Q4 will expand usage, and short-term cloud profit margins will bear corresponding cost pressure.

Amazon: AWS growth continues to accelerate, and enterprise model calls and agent deployments have increased. On October 29, Amazon disclosed 26Q3 results. AWS revenue was 45.98 billion US dollars, up 39.3% year on year, up 2.3 percentage points from 37% in Q2, and accelerated for the sixth consecutive quarter; AWS operating profit was 17.8 billion US dollars, compared to 11.4 billion US dollars for the same period last year. At the product level, Bedrock's token processing volume increased nearly 17 times over the same period last year, and the spending of Fortune 500 customers reached nearly seven times that of the same period last year; AgentCore has already been used by tens of thousands of customers, and the number of agents deployed in Q3 is double that of the previous three quarters combined. Enterprise requirements extend from model call to production environment agent operation, increasing infrastructure consumption for inference, tool execution, and operation management.

The scale of long-term contracts has been expanded, and capital expenses are prioritized to support AWS delivery. The amount of unconfirmed revenue from long-term contracts at the end of the quarter was approximately US$578 billion, up 16.5% from Q2's US$496 billion, with a weighted average remaining period of 6.1 years. The company disclosed Q3 cash capital expenditure of US$75.88 billion, mainly for technical infrastructure and fulfillment networks, maintaining guidance of approximately US$220 billion for the whole year, and is expected to increase significantly in 2027. Continued investment depressed free cash flow. Free cash flow for the past 12 months to the end of September was $42.36 billion. Management expects AWS's revenue growth rate to exceed 40% in 2027 and 2028, and delivery of additional capacity and long-term contract execution form an important support for subsequent growth.

Enterprise software promotes agent deployment, and business data and workflow form the basis for commercialization

SAP: Cloud ERP revenue growth is increasing, and order reserves support the continuation of transformation. On October 21, SAP disclosed 26Q3 results, achieving revenue of 10.270 billion euros, up 13% year on year; cloud revenue of 6.668 billion euros, up 26% year on year, and fixed exchange rate increased 25%, up 4/1 percentage point from Q2 respectively. Cloud ERP Suite's revenue was 5.924 billion euros, and the fixed exchange rate growth rate increased from 27% in Q2 to 28%; the current backlog of cloud orders was 24.144 billion euros, up 5.3% month-on-month, and the fixed exchange rate increased 26% year-on-year. The company adjusted its annual cloud revenue and cloud and software revenue expectations to the upper end of the original guidance range. Cloud ERP migration and enterprise AI construction jointly support business growth.

Joule connects business data with execution processes, and AI applications expand to the core functions of the enterprise. On October 6, SAP showcased the application of Joule Work and Joule Assistants in enterprise business processes, and enhanced cross-system data access and governance through Business Data Cloud and the acquisition of Dremio and Reltio. The company already has more than 110,000 Joule Work Desktop users, and the productivity of core functions has increased by more than 20%; since the beginning of the year, the number of pull requests per developer merging into the codebase has increased by about 70%. SAP combines business data, process authority and agent capabilities. The application covers finance, procurement, sales and development, and provides an AI upgrade portal for existing ERP customers.

ServiceNow: AI contract targets have been raised again, and production environment consumption is growing rapidly. On October 28, ServiceNow disclosed 26q3 results, achieving revenue of US$4.106 billion, up 20.5% year on year; subscription revenue of US$3,988 billion, up 21% year on year; CRPO of US$13.65 billion, up 20.5% year on year, and fixed exchange rate increase of 21%. The company raised the AI ACV target for the end of 2026 from 1 billion US dollars at the beginning of the year to 1.7 billion US dollars. From the beginning of the year to date, Agentic AI consumption in the production environment has increased more than 13 times over the same period last year. While revenue continues to grow, the contract scale and actual use of AI products have increased, and commercialization has penetrated into the business processes of existing customers.

Deployment services promote customer adoption, and platform expansion covers execution and governance. The company added AI deployment support to some large-scale contracts. The AI adoption progress of supported customers in the first five weeks was double that of other customers, and subsequent Assist consumption and product usage were higher. The corresponding deployment service revenue was distributed from subscription contracts to professional services, which had an impact of about 0.25 percentage points on the year-on-year growth rate of Q3 subscription revenue. Q4 is expected to be about 0.5 percentage points. On the product side, Control Tower, AI Passport, and new security capabilities cover agent discovery, authorization verification, continuous monitoring, and exception handling, and promote the expansion of the platform from workflow automation to cross-system AI execution management.

Summary: Overseas cloud revenue, model calls, and agent deployment are growing simultaneously, and enterprise AI demand is spreading to office, development, and core business processes. Copilot usage and pay-as-you-go customers have increased, and Fabric and Bedrock customers have expanded, driving demand for inference, general computing, and enterprise data processing; SAP cloud ERP growth and ServiceNow's increase in AI contract targets reflect the support of existing customers, business data, and workflows for AI commercialization. Capital expenses of the four major manufacturers grew faster than operating cash flow, and the time difference between capacity construction and cash recovery was lower than current cash balances. Computing power infrastructure investment is linked to enterprise AI applications. Servers and storage networks, domestic chips and supernodes, and enterprise data governance and professional workflow applications form an important undertaking direction for industrial investment.

Risk warning: risk of bad debts in accounts receivable; increased industry competition; impact of changes in the international environment.