On September 23, the A-share market provided investors with a vivid lesson in "structural market" dynamics, as major indices closed in the red with over 3,500 individual stocks declining. Yet, hard-tech segments such as PCB, semiconductor equipment, and scientific instruments bucked the trend and rallied, with names like Youlide and Zhongyi Technology hitting their daily limit ups. On one hand, there was pre-holiday profit-taking driving index pullbacks; on the other, localized gains in the AI computing hardware chain. Today's market action perfectly underscored the importance of "choosing the right track."

Pre-Holiday Risk Aversion Intensifies

Today, A-share indices opened under pressure and gradually trended lower through the session. By the close, the Shanghai Composite fell 0.39%, the Shenzhen Component Index dropped 0.64%, the ChiNext Index slipped 0.6%, and the STAR 50 Index declined 0.25%. At the individual stock level, 1,890 stocks rose while 3,563 fell, with decliners accounting for over 60% of the market. Meanwhile, capital was clearly "voting with its feet," as total turnover reached 1.78 trillion yuan, down 370.51 billion yuan from the previous trading day, reflecting a typical pre-holiday risk-off pattern.

Notably, the external environment was actually a tailwind today. Overnight, the Nasdaq Composite rose for a fourth consecutive session to hit another record high, while the Philadelphia Semiconductor Index extended its winning streak to six days. The fact that favorable external conditions failed to lift A-shares suggests the weakness was primarily driven by domestic factors. After three consecutive up days, why did A-shares fall today? Three factors are likely to blame.

First, the approaching dual holidays are suppressing risk appetite. With the Mid-Autumn Festival and National Day on the horizon, capital naturally seeks to lock in gains and reduce exposure. The pre-holiday window typically sees reduced volumes and profit-taking, hitting previously high-flying stocks the hardest.

Second, the previously hot AI application segment is cooling off. This is today's clearest "seesaw" effect. AI-related applications including AI animation, AI marketing, AI film/TV, media, and gaming all declined sharply, with Longban Media, Zhongshi Media, and Huamei Holdings hitting limit downs. On one hand, some targets had been overheated in earlier speculative rallies despite minimal AI-related revenue; on the other, Anthropic and OpenAI simultaneously launched low-priced large models, raising concerns about profit margins in downstream applications. Capital rotated out of applications and back into the hardware chain where industrial implementation is more solid.

Third, regulatory signals and sentiment divergence played a role. The Shanghai Stock Exchange issued a regulatory inquiry letter to Inner Mongolia Xinhua, a stock that had risen five consecutive limit-ups under the "Hua" theme, curbing hot money's high-level接力 sentiment. However, purely name-based speculation on "Hua" themed stocks continues unabated, with names like Xinhua Winshare, Aojiahua, and Shenhua Holdings hitting batch limit-ups-a classic case of speculative hot money betting with elevated risk. Meanwhile, sectors including digital currency, internet finance, power, coal chemicals, agriculture/aquaculture, defense, and commercial aerospace saw broad pullbacks, amplifying losses across the market.

Fourth Quarter Could Bring a 'Feast Market' Rally

Despite today's pullback, mainstream institutions remain relatively constructive on the outlook, with the core view being that a "feast market" rally could emerge in the fourth quarter, but the main investment line must anchor on sectors with solid fundamentals and catalysts. Huaan Securities noted that as overseas disturbances gradually ease, A-shares are likely to break out of their "trading range," and a "sentiment recovery" opportunity remains on the table for year-end. The fourth quarter could see a sizable "feast market" rally, with AI computing hardware and its supporting segments remaining the top choice, while sectors with strong fundamentals or catalysts also deserve attention.

Data from SEMI (International Semiconductor Equipment and Materials Association) provides long-term support for the hardware chain. In its mid-year total semiconductor equipment market forecast report, SEMI raised its 2026 global semiconductor equipment sales forecast to $165.9 billion, up 23.2% year-over-year and reaching a record high. It also projects growth to continue through 2028, with total equipment sales potentially reaching $229.5 billion, representing five consecutive years of expansion. The significance of this data goes beyond a single upward revision of forecast numbers-it signals that AI-related investment is simultaneously driving expansion across three production lines: advanced logic, memory, and advanced packaging. Equipment demand has shifted from single-segment cyclical strength to synchronized multi-segment expansion.

Structural Opportunities Within the AI Value Chain

Some might look at today's market and say "AI is leading again," but that claim requires qualification. The reality is that the leaders are "AI computing hardware plus upstream supply chain," while the AI application segment actually pulled back. On the same day, sharp divergence emerged within the AI value chain-downstream applications cooled while upstream hardware took over the baton. The structural rotation within the AI chain reflects a clear logic on the part of capital: rather than chasing application names with high valuation elasticity and sharp volatility, it is better to embrace hardware companies that "actually consume production capacity." Whether it is cloud-side GPUs, HBM, or upstream copper-clad laminates, copper foil, and test instruments, everything ultimately comes down to real capacity and orders.

Which sectors and stocks have further upside potential? Combining institutional logic with industry trends, the following lines are most favored, each backed by its own "hard support."

Semiconductor equipment and materials (especially upstream components and metrology/inspection) offer the highest certainty. The logic is WFE (Wafer Fab Equipment) spending upgrades, accelerated domestic substitution, and surging orders, with upstream component localization typically materializing ahead of complete machine delivery. Key names to watch include Huafeng Test & Control, Piotech, and Fuchan Precision (components).

The PCB price-increase chain (copper foil, copper-clad laminates, glass substrates) has a tangible pricing logic. With 800G/1.6T ramp-ups continuing, upstream material supply-demand remains tight. Watch Zhongyi Technology and Wogo Optoelectronics, but be wary of pullbacks from hot money speculation and slower-than-expected price sustainability.

Storage, HBM, CPU, and computing chips benefit from domestic substitution plus Agent revaluation. Meta's Muse agent topped the iOS free chart, prompting institutions to reassess "Agent restructuring of computing architecture." Founder Securities projects the AI and Agent-driven server CPU market could reach $220 billion by 2030.

Scientific instruments are supported by dual drivers of independent control and overseas expansion. Youlide, RIGOL Technologies, CYNK Technology, and LabTech benefit from domestic substitution and high-end instrument breakthroughs. However, the sector's overall market cap is relatively small, with clear short-term speculative characteristics, and investors should be wary of "one-day wonders."

Industry insiders believe that on September 23, A-share indices closed in the red due to pre-holiday risk aversion, but the counter-trend rally in AI computing hardware and upstream supply chain revealed capital's medium-term preference: a rotation from "storytelling applications" back to "capacity-driven hard tech." If the fourth quarter delivers the sentiment recovery institutions expect, this hardware chain-with its solid fundamentals, catalysts, and domestic substitution tailwinds-will likely remain capital's preferred battleground.