I. Market Views and Outlook
1. Increasing Market Divergence: Repricing Long-Term Value and Tail Risk
Market divergence intensified in September. Higher interest rates and weaker expectations of policy support pushed up risk premia. At the same time, AI is reshaping competitive dynamics, business models and value distribution in some industries, making their long-term cash flows and terminal values harder to assess.
The S&P 500 fell 0.5% and the Nasdaq rose 1.9%, while the S&P 500 Equal Weight ETF (RSP) and the Russell 2000 declined 5.2% and 5.4%, respectively. Financials, consumer discretionary and real estate were broadly under pressure. Meanwhile, the semiconductor ETF (SOXX) gained 11.3%, the technology sector ETF (XLK) rose 5.0%, and the software ETF (IGV) fell 3.2%. Returns were clearly concentrated in a small number of technology and AI-related areas.
Performance within AI also diverged markedly. Infrastructure areas such as CPUs, custom chips and optical communications led the way, while memory also posted substantial gains. Software was generally weak, although AI security was relatively strong. Returns were concentrated in selected AI infrastructure and AI security areas rather than spread across the entire AI sector.
The uncertainty created by AI goes beyond changes in how users access services: it may also challenge the foundations of some traditional competitive moats. Brand recognition, the stickiness of software workflows and platform distribution advantages depend to a considerable extent on the limits of human attention and decision-making capacity. If AI agents increasingly screen, compare, decide and even transact on users' behalf, interactions between users and brands, software and platforms could be restructured, along with the distribution of traffic and value. This does not mean traditional moats will inevitably disappear, but it increases uncertainty about their durability, long-term cash flows and fair valuations.
Weakness in traditional sectors cannot be attributed entirely to AI; the macroeconomic environment also created pressure. During the month, the U.S. 10-year Treasury yield rose 54 basis points to 5.3%, the dollar index gained approximately 2.1%, and high-yield credit spreads widened by about 49 basis points. The Federal Reserve raised rates by 25 basis points, taking the policy range to 3.75%–4.00%. Markets had to absorb both higher discount rates and higher tail-risk premia.
2. A Clearer Path to Personal Superintelligent Assistants
The prospect of personal superintelligent assistants has long attracted interest. We have consistently viewed AI's evolution from a question-and-answer tool into an intelligent assistant that understands individual needs, coordinates different services and completes tasks—even transactions—on users' behalf as a likely long-term direction. The main uncertainties have been when it would emerge, what form the product would take and which company would ultimately lead.
The recent rapid spread of Muse offers a more concrete product path for this long-term direction. It is too early to judge whether Muse will become the ultimate form or the eventual winner. Nevertheless, it moves personal superintelligent assistants from an abstract concept toward a product direction that can be observed and tested, increasing the visibility of this trend.
Morgan Stanley estimates that applications of this kind could affect approximately US$30 trillion in annual consumer spending across retail, travel, mobility, delivery, advertising and logistics. This figure represents the underlying transaction volume that could be affected; it is not equivalent to the revenue or addressable market that agents themselves could capture.
If this path develops, its implications will extend to both AI infrastructure and applications. Personal intelligent assistants need to understand needs continuously, call models and complete tasks, potentially creating broader and more frequent inference demand. They could also become a new gateway to applications and a layer that aggregates value, giving rise to new product formats and business models. The way consumers discover, compare, decide and transact could be restructured, changing the allocation of traffic and profits among brands, internet platforms, service providers and agents.
3. AI Demand Remains Debated: Focusing on Structural Opportunities
The development of consumer-facing agents expands the potential for AI applications and inference demand, and has partly eased concerns about slowing AI demand. In the near term, however, their task intensity, depth of commercialization and compute consumption have yet to demonstrate an infrastructure impact on the scale of enterprise agents.
The pace of AI demand growth, the sustainability of today's heavy capital spending and the ability to turn that investment into revenue and returns remain unresolved questions for the market. We therefore do not make a simple choice between AI and non-AI. We focus more closely on specific opportunities, assessing demand visibility, competitive advantages, commercial execution, returns on capital and valuation. Taking fundamental delivery and current pricing together, memory and connectivity remain the areas where we see the strongest structural opportunities.
Memory remains a key area of research focus. We retain strong confidence in long-term AI demand, and continued expansion of AI applications requires infrastructure, of which memory is an essential part. Larger models, longer context windows and higher inference concurrency all increase demand for memory capacity and bandwidth. Given our view of long-term AI demand, we would not become overly pessimistic about memory demand and earnings prospects solely on the basis of past industry cycles.
Supply expansion is constrained by manufacturing processes, yields, equipment and customer qualification, making new capacity difficult to bring online quickly. We believe the market may be overestimating how fast additional supply will arrive and underestimating the resilience of memory prices and earnings. Current valuations already reflect considerable downside expectations, while dividends and share buybacks provide further support for expected returns. Overall, the risk-reward profile remains attractive.
Connectivity is another key area of focus. In connectivity, equipment and materials, we place greater weight on customer validation, technological barriers and returns on capital after capacity expansion. Benefiting from AI spending does not mean every company will grow at the same pace. There are time lags between customer budgets, orders and revenue, and the distribution of profits along the value chain may change as supply bottlenecks ease. Demand realization, competitive dynamics and returns on capital are more decisive for a company's long-term value than broad judgments about industry conditions.
II. Other Areas of Focus and What We Are Watching
We will use quarterly operating data to reassess our assumptions about growth, margins and capital expenditure. For companies whose share prices have risen sharply, we need to distinguish the contributions of improved earnings prospects and higher valuations. For those whose prices have fallen substantially, we will reassess whether fundamentals and long-term value have changed. We will not extrapolate high growth far into the future simply because an area has recently performed well.
Beyond AI, fintech, emerging markets and selected consumer companies remain within our research scope. We seek opportunities with independent sources of growth and lower correlation with our existing research areas. Any new area must still meet our requirements for competitive advantage, conviction and expected returns. Our investment focus remains on identifying, at reasonable prices, companies capable of generating sustained cash returns.