From answering questions to taking care of tasks: AI applications are becoming integrated into daily life
Artificial intelligence (AI) is moving from simply answering questions to task execution. Agentic AI can plan steps, call tools according to the user's goals, and complete the work within the scope of authorization. According to a recent report from Morgan Stanley, Muse, a subsidiary of Meta, is expanding its cooperation with companies such as Walmart, Best Buy and Expedia to integrate product search, comparison and travel booking services. In the future, users are expected to use AI to complete processes that originally required switching between multiple websites. The degree of service integration and task execution capabilities will also become important factors in whether an electronic platform can attract and retain users.
Consumption services accelerate integration, computing needs extend to infrastructure
According to a research report for Muse, the potential applications of agent-based AI cover approximately US$30 trillion in consumer spending. If consumers use AI to compare products, select hotels and complete reservations, retailers and tourism operators may need to connect platforms to incorporate products and accommodation into the recommendation and booking process, and part of the marketing budget may also be transferred accordingly. However, product supply, delivery and accommodation are still provided by cooperative operators. Therefore, even if consumers switch to placing orders through AI, the price competitiveness, supply capacity and service quality of the operators will still be the key to obtaining orders.
As retail and travel services are deeply connected, AI's multi-tool invocation and end-to-end task execution will reshape the key to platform customer acquisition and user retention. (Reuters)
Since this type of service requires AI to repeatedly reason, call tools, and retain the content of previous and previous tasks, the computing load is higher than that of a single question and answer, and it also increases the demand for processors and memory. The "Mapping AI’s Rate of Change: From Buildout to Broadening" model released by Morgan Stanley estimates that the proportion of memory in the total cost of AI cabinets is less than 10% in the previous generation of Blackwell cabinets and about 25% in the new generation of Vera Rubin cabinets. Based on the above computing requirements and cost considerations, memory capacity, access efficiency and cost control are all important factors in system design. In addition, the expansion of computing demand has also prompted cloud operators to expand data centers and increase the demand for power supply and related facilities. The same report predicts that large cloud and emerging AI cloud players will continue to expand their computing power supply, and their facility capacity, measured in terms of power capacity, will increase from approximately 36GW in 2025 to approximately 145GW in 2028, approximately four times the original amount. This is an estimate of the expansion of the overall AI infrastructure. The actual progress still depends on factors such as power supply, computer room construction, and equipment delivery.
From industry trends to asset allocation, understand the sources of returns and risks
Agent-based AI creates cross-industry demand, but it does not necessarily bring the same level of investment returns to each industry. The platform must convert usage into revenue, equipment vendors are affected by the procurement progress, and infrastructure operators must bear construction and financing costs. With the rapid technological evolution of AI-related industries, existing competitive advantages may also be changed, so that demand growth may not be reflected simultaneously in corporate profits and stock price performance.
Therefore, asset allocation must consider growth potential, sources of income and risk tolerance at the same time. Dispersion into different industries and assets can reduce dependence on a single enterprise or theme, but the effect of dispersion still depends on the degree of linkage between assets. As AI commercialization progresses, corporate evaluations and the interest rate environment change, investors can regularly review and timely adjust the proportion of stocks and bonds and industrial allocation, which will help diversify risks and retain the flexibility to participate in industry development.
(The author is a professional assistant at the International Financial Business Department of Yuanta Securities)
