CITIC SEC: AI training and inference needs contribute to DRIVE's continued optimism about top-tier cloud vendors' investment opportunities in 2026.

CITIC SEC released a research report stating that the strong demand for AI model training and inference is expected to continue to drive the performance of cloud computing vendors. Although the shortage of computing power has to some extent limited the acceleration momentum, with continued investments from major companies and the gradual delivery of data centers, the supply shortage is expected to marginally improve by 2026, leading to a more stable pace of order fulfillment. Overall, considering the pace of order fulfillment and the data center reserves of various cloud vendors, it is expected that the year-on-year growth rate of cloud computing revenue in 2026 will further accelerate compared to 2025. First-tier cloud vendors are expected to have good investment opportunities in 2026. Key points from CITIC SEC: Report origin: Revenue growth of North American cloud vendors continues to rise. Since Q3 2025, the revenue growth trend of leading cloud computing companies in the American stock market has become clear, signaling an acceleration in overall industry revenue growth. Looking ahead to 2026, the bank expects the prosperity of cloud demand and the incremental demand for AI cloud to remain the focus, along with the gradual easing of supply-side constraints such as chips and data centers, leading to an improvement in the pace of order fulfillment. Based on this, the report will provide a reference for investment decisions by analyzing the overall trend of cloud computing in 2026 and the pace of performance realization through a combination of qualitative and quantitative analysis. Growth sources: AI computing power rental as the primary source of growth. AI clouds have become the core incremental engine driving the growth of the US cloud computing market. However, from a business model perspective, the current AI cloud market is still dominated by the basic model of computing power rental. Both major players and emerging players focus on the deployment and sublease of GPU/TPU and other computing resources, while major demand is tied to large customers. Profitability is achieved through chip resource supply and infrastructure services. In contrast to the more mature traditional cloud computing market's differentiations in elastic scalability, general PaaS tool chain, etc., the current core competition focus of the AI cloud market is on the speed of acquiring flagship chips, cluster scale, and deployment efficiency. Factors determining the performance of the AI cloud are more focused on computing power supply and cost optimization. Supply and demand analysis: Head customers dominate, short-term resource constraints evident. The strong demand for computing power from leading AI model companies is the core support of the demand side of the AI cloud market. According to a report by Reuters in November, OpenAI's annualized revenue is expected to exceed $20 billion by the end of 2025, and Anthropic's annual recurring revenue is close to $7 billion. Business expansion has driven a sharp increase in demand for computing power. So far this year, OpenAI has signed over $140 billion in computing power procurement contracts, while Anthropic has also increased its computing power procurement and become an important customer for Amazon Trainium and Google TPU chips. In the context of a shortage of computing power, leading model companies are accelerating the deployment of multicloud strategies to break supply bottlenecks, diversify supply chain risks, and optimize costs. The demand for orders has overflowed from core cloud partners, driving the market towards a "multipolar competition" transformation. At the same time, incremental order opportunities are brought to cloud vendors with self-developed chip capabilities, high computing power density, and cost advantages. The bank expects the total capital expenditure of the North American top four CSPs to reach $406 billion in 2025 (up 61% year-on-year) and to increase to $517 billion in 2026 (up 27% year-on-year). There is a time lag between the release of computing power supply and the investment of capital expenditure. Given that major cloud vendors increased their CAPEX in the second half of 2023 and combined with the pace of landing key data centers, computing power supply is expected to gradually ease by quarters in 2026, and the performance of each cloud vendor is expected to further release. In addition, the introduction of high-performance chips such as Google TPU and AWS Tranium chips will also bring supplements. Performance estimation: Based on the backlog estimation, the overall revenue in 2026 is in an accelerating channel. The bank expects that with the gradual delivery of data centers, the shortage of supply is expected to marginally improve by 2026, leading to a more stable pace of order fulfillment for cloud vendors. It is expected that cloud vendor revenue will maintain a strong trend in Q4 of 2025, and the overall trend in 2026 will enter an acceleration channel. Risk factors: Risks of AI technological progress and commercialization falling short of expectations; risks of overall slowdown in global IT spending; risks of major cloud vendors slowing down in technology research and development; risks of cloud computing infrastructure construction not reaching expected scale effects; risks of industry ecosystem construction not meeting expectations; risks of core talent loss; risks of geopolitical conflicts leading to the inability to use key technologies; risks of stricter industry regulations; risks of cloud vendor performance improvement and capital expenditure control falling short of expectations.