Currently, global technology stocks are reaching new highs driven by the AI wave. From the soaring capital expenditures of North American cloud providers to the exponential growth in domestic computing power demand and the super cycle in memory chips, a technology boom cycle fueled by genuine profitability is unfolding. Amid this fervor, which areas still offer certainty?
At the Guotai Asset Management Summer 2026 Strategy Conference, themed "Gentle Breeze Strengthens, All Things Compete to Flourish," fund manager Peng Lingzhi shared his investment outlook for the technology sector.
**Overseas Computing Power: Capital Expenditures Exceed Expectations; Chinese Manufacturing is Irreplaceable**
Regarding overseas computing power, the core logic lies in the sustained gap between the exponential growth in AI demand and the physically constrained supply.
Capital expenditure data shows that the four major North American cloud providers spent approximately $240 billion in 2024, which increased to $410 billion in 2025 and is projected to climb further to $710 billion in 2026. Market expectations suggest it may approach $1 trillion next year. The continuous outperformance of capital expenditures provides the most solid fundamental support for overseas computing power.
Simultaneously, with the rapid growth of numerous AI application companies over the past three years, previous market concerns about a "lack of killer applications" are being disproven. The commercial closed loop for AI has formed, meaning computing power demand is not built on sand.
Investment directions primarily revolve around optical modules and PCBs. Optical modules are transitioning from 800G volume production to 1.6T, with 3.2T on the horizon. Technological innovation continues along the path from optical modules → optical chips → silicon photonics → CPO. PCBs benefit from the explosive demand for AI servers, with technological upgrades from M7 to M8 and M9 further driving increases in both volume and price.
Notably, mainland China accounts for 80% of global PCB production capacity and over 60% for optical modules. Chinese suppliers play a crucial role in the global AI industry chain, serving as the key bridge for translating overseas computing power investments into A-share market opportunities.
**Domestic Computing Power: Inference Demand Explodes; Significant Room for Capital Expenditure Growth**
We are also optimistic about domestic computing power for several reasons:
* **Explosive growth in inference demand.** As large models shift from training to inference, China's advantages in application development are becoming evident. The domestic daily Token consumption surged from 0.1 trillion in early 2024 to over 140 trillion by March 2026. Furthermore, domestic large models have achieved performance close to overseas counterparts at extremely low costs, leading to "computing power inflation"—cost reductions, far from being negative, have instead stimulated accelerated demand, causing a sharp increase in computing power consumption. * **Domestic capital expenditures lag but an inflection point has emerged.** Currently, AI has become a focal point in major-power technological competition. However, the capital expenditures of domestic cloud providers remain less than one-tenth of their overseas counterparts—a gap unlikely to persist long-term. As leading domestic companies reduce share buybacks and revise capital expenditure plans upward, we believe the catch-up in domestic capital spending is just beginning. * **Potential for significant increase in market share of domestic chips.** Following the supply cut-off of overseas high-end chips, the localization rate increased from 19% in 2023 to 41% in 2025, and is projected to reach 60% in 2026 and potentially 93% by 2028. As the competitive focus shifts from single-card performance to system integration capabilities, leading companies with superior node server capabilities may see their advantages become more pronounced.
Notably, Q1 2026 reports show leading GPU companies achieving over 100% growth in prepayments, indicating explosive performance realization is underway. Domestic computing power has entered the phase of earnings delivery.
**Semiconductors: Driven by Memory Super Cycle; Market Has Not Fully Priced It In**
Semiconductor equipment, materials, and components are our most favored areas.
The core driver is the memory super cycle brought by AI. AI servers' demand for memory far exceeds that of traditional sectors. Given the long expansion cycles for memory wafer fabs, large-scale capacity additions are not expected until the second half of 2027, suggesting tight supply and demand conditions will persist until then.
The expansion of domestic memory production is timely, driving demand for equipment and materials. Currently, domestic memory chips account for less than 10% of global self-sufficiency. We project this could rise to 30% by 2030, representing a 3-5 times growth potential. We believe the accelerated expansion of domestic memory production over the next five years will directly boost demand for semiconductor equipment, materials, and components.
From the equipment perspective, memory capacity expansion was only 70,000-80,000 wafers per month in 2025, is projected to reach 120,000-150,000 in 2026, and is expected to hit 200,000 by 2027. The earnings realization for equipment companies may accelerate quarter by quarter. Key investment areas include etching equipment, thin-film deposition, metrology/inspection, and lithography machines—segments with high value and low localization rates. Among these, metrology/inspection equipment, with its software-like attributes, boasts extremely high gross and net profit margins, making leading companies in this space particularly noteworthy.
The logic for materials is similar. Materials possess a "selling shovels" characteristic, being difficult to replace once adopted. Segments with large market shares, such as silicon wafers, photoresists, electronic specialty gases, and targets, offer vast potential for domestic substitution.
**Q: Is the current technology sector overheated?** **A: There are localized bubbles, but the upward trend in high-growth areas has not reversed.**
Compared to the 2000 U.S. stock market bubble when the Nasdaq had a P/E ratio of 150-200x, the Nasdaq 100's forward P/E for next year is around 20-25x, near the median of the past decade. Even in the memory chip sector, which has seen the largest gains recently, P/E ratios are only around 6-9x. Furthermore, during this market cycle, Nvidia has consistently exceeded earnings expectations for multiple quarters, and domestic GPU companies have seen prepayments double, indicating this technology rally is supported by profits, not mere thematic speculation.
We acknowledge that some localized areas may be overextended, but systemic collapse typically requires a disproof of the underlying business model. Currently, the AI commercial closed loop is operational, and capital expenditures are still increasing. We believe demand is far from peaking.
**Risk Disclosure:** Views are for reference only and may be adjusted dynamically with market changes. They do not constitute investment advice or guarantees. The market involves risks; investment requires caution.
Comments