CITIC Construction Investment: AI agent expansion scenarios to accelerate commercial application implementation are expected to drive computing power demand

The Zhitong Finance App learned that CITIC Construction Investment released a research report saying that Meta Muse and the Yunqi Conference are jointly strengthening the logic of implementing AI applications to drive computing power demand. Meta launched Muse, a personal AI assistant, which supports email, travel reservations and transaction execution. The number of downloads reached 2.8 million times in 12 days, and promoted integration with the Shopify payment ecosystem to promote commercialization of consumer-grade smart devices. At the Yunqi conference, Ali revealed that Qwen 4 is being trained, released the Zhenwu V900 chip, proposed the goal of global data center capacity exceeding 20 GW in 2032, and promoted the implementation of enterprise intelligence through Qianwen Office. The bank believes that expanding usage scenarios for consumer-side agents and deepening the participation of enterprise-side agents in the process is expected to increase the demand for multiple rounds of reasoning, tool calls and continuous operation, drive consumption of computing, storage and network resources, and focus on high-speed optical modules, switches, servers, liquid cooling and power supplies.

CITIC Construction Investment's main views are as follows:

Meta Muse promotes the upgrading of AI from dialogue and interaction to task execution, and commercialization of consumer-grade smart devices ushered in new progress. On September 8, Meta launched Muse, a personal AI assistant that can help users send emails, book trips, and complete transactions. It uses a free basic version and a monthly paid subscription model of $20 and $100. According to Reuters quoting Apptopia data, the cumulative number of downloads reached 2.8 million in the 12 days before the product was launched, indicating strong consumer demand for task-execution AI. On the ecological side, Shopify announced that it will promote the integration of Muse and Shop Pay checkout capabilities to open up smart shopping and merchant transactions; Amazon has restricted Muse's access to its platform, reflecting the further increase in the importance of traffic entry, transaction control, and data authorization during the development of smart devices. The bank believes that the industrial significance of Muse is to expand the value delivery methods of AI: as intelligent agents participate in product screening, service reservation, and transaction execution, the business model is expected to be further extended from subscription payments to transaction services, and application ecology and execution capabilities will become the key to competition.

The Yunqi Conference focused on model, chip and cloud infrastructure collaboration to promote the deepening of domestic AI applications to large-scale applications. At the 2026 Yunqi Conference, which opened on September 22, Alibaba revealed that Qwen 4 is being trained. Subsequent Qwen 4.5 and Qwen 5 plans will evolve to a parameter scale of 5 trillion to 10 trillion yuan, and continue to promote model self-iteration and multi-modal capacity building. In terms of computing power, the company released a new-generation Zhenwu V900 chip, revealing that its performance has reached three times that of the previous generation M890. The cluster plan based on this chip can support up to 500,000 cards, and proposed a goal of global data center capacity exceeding 20GW by 2032. The above indicators are the product capabilities and construction plans disclosed by the company. On the application side, Qianwen Office launched enterprise context products, connected to Salesforce's customer management capabilities, and promoted the integration of intelligent entities into the actual business processes of the enterprise.

The bank believes that the two developments are jointly strengthening the industrial logic of application implementation driving demand for computing power. Consumer-side agents expand usage scenarios, and enterprise-side agents deepen process participation, which is expected to increase the demand for multiple rounds of reasoning, tool calls, and continuous operation, and drive the consumption of computing, storage and network resources. Follow-up focus is on tracking user retention and payment conversion, enterprise deployment results, and cloud vendor construction and implementation, and focusing on demand transmission in infrastructure such as high-speed optical modules, switches, servers, liquid cooling and power supplies.

Risk Alerts

Changes in the international environment have had an impact on supply chain security and stability; the impact of tariffs exceeded expectations; the development of the artificial intelligence industry fell short of expectations; market competition intensified; exchange rate fluctuations affected the exchange earnings and gross profit margin of export-oriented enterprises; the cloud computing business development of telecom operators fell short of expectations; the capital expenditure of cloud vendors fell short of expectations; and demand in the communication module and intelligent controller industry fell short of expectations.