Investors in the A-share technology sector have had a tangible experience of its performance in the first half of the year. Sectors like AI computing power and memory chips saw substantial gains, with market sentiment reaching a fever pitch at times.
However, a significant shift occurred entering July. A notable market correction has prompted many investors to reassess: is this a technical adjustment or a trend reversal? Does the substantial prior appreciation necessitate a longer digestion period, and could the market focus shift elsewhere in the second half of the year?
To explore this, we have compiled recent insights and public views from several prominent private equity firms, revealing a divergence in their outlooks. Some maintain that the technology theme is not over, viewing the pullback as a window to identify quality stocks. Others caution about valuation pressures and have begun positioning for value opportunities in other sectors.
Dongfang Harbor: "Turbulent Eddies" Within the Unchanged AI Megatrend
On July 6th, Dongfang Harbor published a report titled "The Driving Force and Turbulence of the Era," suggesting that recent volatility in AI computing power and memory is more akin to "turbulent eddies" within the larger AI wave, not a reversal of the trend.
The primary market issue, according to the firm, is the high concentration and appreciation in certain sectors, creating short-term divergences. These include the gap between traditional corporate investment and output efficiency, the disparity between semiconductor sector hype and real purchasing power, and market valuation dislocations. However, reviewing the four factors typically required for a systemic crash—valuation, leverage, interest rates, and a trigger event—the firm notes that only elevated valuations in some AI assets are present currently.
From a longer-term perspective, Dongfang Harbor identifies four enduring themes in the AI era: First, "Recursive Engineering" is replacing traditional prompt engineering, evolving AI into autonomous, self-iterating agents, which dramatically increases computing power consumption. Second, the cost per token continues to decline, with advancements in hardware and software allowing for orders-of-magnitude increases in output for the same investment. Third, a "Token Economy" is gradually taking shape, with business models based on task value beginning to materialize. Fourth, the AI race among tech giants has evolved into a "prisoner's dilemma"-style arms race, where intense competition itself fuels innovation.
In summary, Dongfang Harbor leans towards the view that the major AI wave is not over; volatility is a natural byproduct of an accelerating current. For long-term investors, the opportunity lies in rationally navigating risks while enduring the rapid fluctuations to capture the era's potential rewards.
Zhong Ou Rui Bo: Market Likely to Shift from "Tech Dominance" to "Broad-based Bloom" in Q3
On July 6th, Wu Weizhi, founder and chairman of Zhong Ou Rui Bo, published an article questioning whether the current phase represents a "super growth cycle or an epic bubble." He posits that the market's divergence on AI hardware stems from a misalignment between "industrial perspective" and "investment perspective." From an industry standpoint, this is undoubtedly an unprecedented super-cycle. The combined capital expenditure of the top five US cloud providers for 2025-2026 is projected at $1.2 trillion, with China's top seven firms around 1.2 trillion RMB. The data center construction supply chain is extensive, spanning from GPUs and HBM to optical modules, MLCCs, and transformers, where a shortage in any link can cause supply-demand imbalances and price spikes.
This is a level of industry prosperity unseen in his over-30-year career, far exceeding the scale and impact of the 2008 stimulus plan.
From an investment standpoint, however, the crucial question is: to what extent has the current stock price already reflected this high growth?
Wu cautions against all-in bets based solely on "high growth," using Howard Marks' "second-level thinking" and the example of investors who bought Kweichow Moutai in 2021. While the company was sound, buying at peak valuations led to years of losses. For AI hardware, careful analysis of competitive dynamics and moats in each segment is essential. Sectors with short capacity expansion cycles exhibit stronger cyclicality than growth; a downturn could lead to a severe double whammy on earnings and valuations.
Additionally, Wu observes a noticeable market shift since July. Sectors like communications and electronics, which led gains in H1, have corrected deeply, while sectors like pharmaceuticals and agriculture have rebounded, indicating a move from tech sector dominance towards market equilibrium.
He believes the bull market that began in September 2024 is still intact. The third quarter will likely see a transition from "tech dominance" to a "broad-based bloom," evolving from a tech-led bull market to a comprehensive one. The strategy should maintain a bullish mindset but avoid being overly fixated on daily market noise. Investors should focus on holding sufficiently high-quality and reasonably priced assets for confidence.
Springs Capital: Market Shifts to "Earnings Verification," Advocates Balanced Allocation
Springs Capital summarized the first half with the term "K-shaped divergence."
A vast gap emerged between AI-intensive assets and traditional sector assets. Core AI sectors like electronics and communications surged over 70% in H1, while traditional sectors like retail and agriculture fell over 25%. This divergence is not unique to A-shares but is a global phenomenon observed in US and Korean markets, driven by synchronized industrial logic.
In Springs Capital's view, this is not mere speculation but is underpinned by solid industrial fundamentals.
On one hand, North American cloud providers' capital expenditures have grown consistently for years. Large language models are beginning to generate real, rapidly growing commercial revenue, gradually building a positive global cycle from "capex → computing demand → application monetization."
On the other hand, Chinese companies are deeply embedded in the global AI supply chain. Orders in areas like optical modules and MLCCs have seen substantial growth, coupled with accelerated domestic substitution in China. China's integrated circuit exports in May saw a year-on-year growth rate accelerating to 111%.
Looking ahead to H2, Springs Capital believes the underlying positive factors supporting A-shares remain unchanged, and the market is expected to continue its structural trends. As the mid-year earnings reporting season approaches, the market is likely to shift from "sentiment-driven" to "earnings-verification" mode.
Strategically, the firm states it has not blindly followed trends. Instead, it employs a moderately balanced allocation to manage risk and return, focusing subsequently on fundamentals to identify leading companies with core competitive advantages and long-term pricing power.
Chongyang Investment: Cautious on AI Bubble, Seeks Overlooked Gems in the "Middle Ground"
Chongyang Investment recently published an article discussing the path forward for the "extreme divergence" seen since Q2.
Regarding this divergence, Chongyang attributes it largely to fundamental differences. AI benefits from massive capital expenditure and tangible commercialization progress, while traditional sectors face multiple headwinds. Simultaneously, homogeneous information flow, quant-driven momentum trading, and institutional forced selling have exacerbated the split.
On the question of an AI bubble, Chongyang adopts a cautious stance. Its portfolio managers note that historical bubbles often started with genuinely revolutionary technologies, where "having a real industrial logic" is precisely a prerequisite for bubble formation. While this rally has a solid industrial foundation, the gains have been substantial, and current pricing implies highly optimistic assumptions about the realization timeline. Therefore, necessary prudence is warranted in stock selection within the theme.
The preference is to participate at points of higher certainty rather than paying a high premium for overly linear optimistic assumptions.
For the market outlook, Chongyang provides a historical reference: after the 2000 dot-com bubble burst, both US and Japanese stocks exhibited strong momentum reversal. Sectors that fell the most during the bubble period saw robust rebounds over the subsequent 9-15 months. Mapping this to the present, they judge that if the AI trend reverses, the market will see a style rebalancing, benefiting sectors with low valuations and low expectations.
Strategically, the managers are aligned: they will avoid overheated stocks while actively seeking overlooked opportunities. These include: 1) Companies in the vast "middle ground" between AI hardware and traditional blue-chips (e.g., aerospace, bio-manufacturing); 2) Leading companies in traditional sectors that have adjusted for years, which may see valuation repair from a trading perspective if the tech rally corrects; and 3) Innovative drug companies, which have seen significant declines, possess strong industry trends second only to AI, and have re-entered an attractive valuation zone.
Harmony Asset: Capital Expenditure Intensity Nears 2000 Dot-com Levels, Application Layer More Appealing
On July 3rd, Harmony Asset published "How Far Can the AI Theme Go?" stating that the global capital concentration into AI is approaching or even surpassing levels seen during the 2000 internet bubble.
Han Dong, an investment manager at Harmony Asset, noted that the AI infrastructure investments by several North American cloud providers in 2024-2025 are essentially consuming all their free cash flow. Measured by the ratio of capital expenditure to cash flow or revenue, the current investment intensity is near or above 2000 levels. The projected $700 billion investment this year and over $1 trillion next year equate to about half the combined revenue or all the operating cash flow of these major providers, gradually exceeding their internal funding capacity.
A larger issue is the terminal revenue gap. Han Dong pointed out that to sustain the current AI infrastructure investment intensity, future end-user applications need to generate roughly $1 trillion in annual revenue to recoup one year's investment cost within five years. Currently, large language model providers are projected to have an annualized revenue of only $200-300 billion by the end of 2026.
This investment surge is largely driven by corporate FOMO (Fear Of Missing Out), creating a significant gap between high-intensity upfront investment and the actual growth in end-user revenue.
However, Harmony Asset is not entirely bearish.
Liang Shuang, another investment manager at the firm, believes the broad direction of AI investment is certain, and some froth is normal. Han Dong also judges that AI will remain a key investment theme in 2026, but its internal structure may change. The investment focus may gradually shift from hardware (computing infrastructure) towards software (applications and commercialization), with the industry logic potentially transitioning from an "arms race" to "innovation-driven."
Strategically, Harmony Asset prefers seeking structural, more reasonably valued entry points within the broader AI theme. For hardware, it focuses on reasonably valued stocks and those entering major supply chains. The application layer's potential breakout becomes more attractive once infrastructure reaches a certain level.
Comments