Over the past month, AI industry fundamentals continued to show validation, while AI-related assets experienced a significant valuation correction. Many AI companies trade with long-duration valuation characteristics, where a substantial portion of intrinsic value is derived from future cash flows. As a result, even modest changes in expectations regarding long-term demand growth, capital return profiles, or competitive dynamics can lead to meaningful valuation adjustments.

The recent correction was driven by both technical factors, including crowded positioning and deleveraging, as well as a rapid reassessment of AI-related expectations. In our view, the pace of price adjustment has exceeded the pace of change in fundamental evidence, resulting in a degree of short-term overshooting. Therefore, the recent volatility should be viewed more as a rapid repricing of expectations and valuations, rather than a fundamental reversal of the AI investment cycle.

Market attention has gradually shifted from the previous narrative of “large AI opportunity and insufficient infrastructure supply” toward questions around whether AI demand growth may slow too quickly and whether the significant capital investment required can generate sustainable returns.

The most important variable currently remains the trajectory of AI demand growth. Based on public information and market estimates, the combined annualized revenue of several leading AI labs may increase from approximately US$40 billion at the end of last year to more than US$200 billion by the end of this year. A moderation in such an exceptionally rapid growth rate would be natural; the key question is whether the slowdown remains within a reasonable range.

Given that AI infrastructure investment has already reached a trillion-dollar scale, a sharp deceleration in revenue growth could impact the economics of continued investment, industry supply-demand dynamics, and infrastructure valuations.

AI Adoption: From Capability Exploration to Enterprise ROI Validation

Coding and AI agents have entered enterprise workflows for less than one year. During the initial stage, users primarily focused on experimentation and exploring capability boundaries. As investment expands, enterprises are increasingly evaluating measurable productivity gains and reassessing the economic value proposition.

Currently, heavy AI users remain concentrated in technology companies. Some enterprises have begun to adopt stricter budget discipline, while traditional industries represent a significant long-term opportunity but require further validation regarding adoption speed. Factors such as product usability, workflow redesign requirements, and implementation complexity may influence the pace of penetration.

Our long-term view on AI remains unchanged. The cost of intelligence continues to decline rapidly, model capabilities have not yet demonstrated a clear ceiling, and AI adoption across the real economy remains at an early stage.

Executives of leading frontier AI companies, including Sam Altman, continue to express strong confidence in further model capability improvements over the coming years. The underlying drivers are not simply larger models, but improvements in inference-time compute, search and verification mechanisms, the transition of agents from “answering questions” to “completing tasks,” and the increasing use of AI in model development and scientific research.

As AI systems become capable of independently completing longer and more complex workflows, their value proposition may expand from individual responses toward higher-value applications including scientific research, software engineering, industrial design, and professional services.

These developments support our view that AI capability curves have not yet reached a plateau.

However, management commentary should only be viewed as directional evidence; ultimately, actual model performance and commercial adoption data will determine the outcome.

At the same time, improving AI capability does not automatically translate into industry revenue growth. “Intelligence deflation” has two sides: lower costs can expand usage and unlock new applications, but if users simply accomplish existing tasks at lower cost, total spending may not increase proportionally.

The key question is whether capability improvements and newly created economic value can grow faster than the decline in unit intelligence costs, enabling AI to evolve from a productivity-enhancement tool into a driver of new revenue creation.

Therefore, beyond model rankings, we believe the more important indicators are:

• The reliability of AI agents in completing complex tasks;

• The speed of enterprise workflow adoption;

• Corporate AI budgets and willingness to pay;

• The economic value generated per unit of compute.

AI Infrastructure: Our Base Case

Our base case remains that AI demand growth will gradually normalize from the exceptionally rapid pace seen previously, but is likely to remain strong enough to support continued infrastructure investment.

Compared with scenarios of either continued acceleration far beyond expectations or a sharp demand collapse, we believe a moderated but sustainable growth trajectory remains the most likely outcome. Under this scenario, the fundamental demand drivers and pricing foundation of AI infrastructure remain intact.

However, current market valuations continue to embed high expectations for future growth. As a result, tolerance for near-term earnings disappointment remains limited. Even modest deviations in quarterly revenue, orders, or guidance may be interpreted as evidence of changes in long-term growth assumptions, leading to significant valuation adjustments.

The higher the proportion of long-term value embedded in a company’s valuation, the more sensitive the stock price tends to be to short-term deviations. Recent market reactions following earnings announcements from companies such as Corning and Nokia illustrate this dynamic.

Within the AI value chain, recent financial results have been relatively more supportive for cloud service providers. Cloud revenue growth has accelerated, profitability has improved, and while capital expenditure remains elevated, investment levels currently appear broadly aligned with revenue realization.

Meanwhile, as open-source AI ecosystems gain adoption, some value that previously accrued primarily to closed-model providers may increasingly remain within cloud platforms, potentially strengthening cloud providers’ position in the AI value chain.

However, this potential shift in value capture ultimately depends on continued expansion of overall AI demand and enterprise AI spending.

Overall, we remain balanced in our assessment of AI. We do not believe short-term market volatility should lead to excessive pessimism toward the long-term AI opportunity, but neither should structural optimism justify ignoring growth normalization, capital efficiency, or valuation risks.

Our focus will remain on monitoring fundamental evidence, including AI usage trends, enterprise spending behavior, adoption across non-technology sectors, returns on AI infrastructure investment, and incremental supply additions. We will continuously update our assessment and adjust portfolio positioning as these variables evolve.