Technology stocks experienced a significant correction in July, with sectors like optical modules, computing chips, and memory seeing declines of over 20% from their peaks. Market sentiment shifted sharply from optimism to anxiety. This volatility was driven by a convergence of pressures, including negative overseas narratives (such as Meta selling idle computing power, sparking fears of peak capital expenditure), high trading concentration, profit-taking ahead of the semi-annual report window, and forced deleveraging of margin positions, creating a negative feedback loop. The sector has clearly overshot as these forces combined.

After the extreme sentiment has been released, a calm assessment is needed. On one hand, A-share technology stocks account for only about 30% of total market capitalization, compared to over 40% for the US tech sector. This suggests room for valuation expansion. On the other hand, the unwinding of crowded trades is progressing rapidly, making it premature to conclude the rally has peaked or is at an absolute high. Fundamentally, the medium-to-long-term logic for AI remains intact. The dynamic of "carbon-based deflation, silicon-based inflation" continues to deepen, with traditional labor and consumption-related industries facing headwinds, while the supply-demand gap for AI computing power, the silicon-based track, is widening, supported by strong earnings delivery.

From a global supply chain perspective, capital is prioritizing hardware segments with the strongest supply rigidity and revenue that can be recognised upon delivery. Memory and computing chips are in short supply, with earnings rising. Internet giants, as hardware buyers, require years to see returns on their massive capital expenditure, limiting their short-term earnings flexibility. This year, the Philadelphia Semiconductor Index has significantly outperformed the "Magnificent Seven" tech stocks and the software sector, which is a direct reflection of this logic. The combined capital expenditure of global leading companies this year is over $700 billion, essentially a redistribution of value along the chain towards hardware. The root cause of the imbalance is hardware scarcity: computing power supply roughly doubles every seven months, while demand, measured by token consumption, can grow tenfold over a year. This growth gap sustains a tight equilibrium in the supply chain.

From a demand perspective, the annualized recurring revenue of major large-model companies remains steeply upward, even if short-term growth is constrained by computing power supply. The AI hardware investment thesis is far from a turning point. Looking ahead, AI sector investment will focus on three main directions. First, tracking the earnings reports of major overseas tech companies to capture the opportunity for a rebound in overseas computing hardware expectations. The recent pullback in overseas computing power stems from concerns about memory price increases squeezing budgets and the contraction of the computing power leasing track. These concerns may be overblown. As leading North American companies report results, their capital expenditure guidance for next year is expected to be revised up from the current estimate of $1 trillion to $1.2-1.3 trillion. After the pessimism is absorbed, the valuation of related supply chain stocks is likely to recover.

Second, deeply exploring the semiconductor cycle theme by investing in high-quality targets that benefit from both industry growth and domestic substitution. The semiconductor industry is currently experiencing a strong cycle, with global and domestic sales growing rapidly year-on-year, supported by solid fundamentals. While the pace of memory price increases slowed in the second quarter, the trend remains intact, and prices are expected to stay high in the third quarter. Leading memory companies are releasing significant profits, which may raise the peak capital expenditure targets for local wafer fabs, benefiting the equipment and materials sectors. Combined with strong demand for domestic computing chips amid tight supply, the process of self-sufficiency is accelerating, providing robust support for the sector's fundamentals.

Third, tracking the iteration and commercialisation of large models, using the underlying model's capabilities and the slope of annualised revenue growth as yardsticks to identify leaders with long-term potential. As a core driver of the new productivity revolution, the trillion-dollar AI industry space has not yet been fully unlocked. We will continue to follow the industry's rhythm, select high-quality companies based on fundamentals, and seize long-term investment opportunities. Note: Market conditions and other factors may change, and the above views are subject to change without notice. Funds carry risks, and investment should be made with caution.