‘Zheng Zheng Jing Jing’: A-shares will resume trading tomorrow — what stocks should you buy? During the A-share market holiday, overseas markets have shifted from buying the AI application story to focusing on the scarcity of AI hardware, thus four sectors are worth watching.
Star Financial Research Institute points out:
1. Optical Communications
Optical interconnects are the core channel for data exchange between GPUs in AI clusters. As cluster scale expands and port speeds upgrade, the cost share of optical modules continues to rise. Industry research shows that Ethernet optical module sales doubled in 2024, with nearly 90% growth expected in 2025, and 120% growth projected in 2026.
*Optical communications maintains high growth*
Overseas investment banks estimate that the global optical module market size could reach $68 billion, $131 billion, and $148 billion in 2026, 2027, and 2028 respectively. On the industrial front, 800G remains highly prosperous, 1.6T is entering mass production, and the industry is advancing toward 3.2T and CPO/NPO. Order backlogs for some high-speed optical chip companies have already extended to 2028. Upstream segments such as optical chips and optical components, due to greater difficulty in capacity expansion, may exhibit stronger pricing elasticity.
2. PCB and CCL
The outperformance of Hong Kong’s PCB (printed circuit board) sector during the holiday was no accident. Since the second half of 2025, rapid growth in demand for AI servers and high-speed networks has led to widespread supply tightness in PCBs, IC substrates, and core materials. Shortages began with advanced materials such as low-CTE and low-Dk2 glass fiber cloth and HVLP4 copper foil, then spread to CCL (copper-clad laminates), high-end PCBs, ABF (a high-performance insulating film) substrates, and manufacturing equipment.
*PCB market remains strong*
Data from overseas investment banks show that demand for AI server CCLs will surge from 42 million sheets in 2026 to 131 million sheets in 2028, representing a compound annual growth rate of about 77%; average prices will climb from $165 to $362, with a compound annual growth rate of about 48%. Kingboard Laminate Holdings (01888) has issued multiple rounds of price increase notices this year, and Nan Ya Plastics raised its product prices by another 20% to 25% starting in September, with price hikes fully transmitting downstream. The PCB sector’s rally essentially reflects the realization of a chain: ‘material shortage → capacity strain → price rise → earnings upgrade’.
*Domestic memory catching up rapidly*
3. Memory
Demand for HBM (high-bandwidth memory) and enterprise SSDs (solid-state drives) in AI servers continues to rise, with annual bit demand growth for enterprise SSDs expected to exceed 80% in 2026. TrendForce forecasts that in Q4 2026, contract prices for standard DRAM (dynamic random-access memory) will increase by 10% to 15% quarter-on-quarter, while NAND Flash contract prices will rise by 15% to 20% quarter-on-quarter.
Domestic memory manufacturers are also accelerating their catch-up. CXMT Technology (SHA: 688825) achieved an EBIT margin as high as 82% in Q2 2026, ranking first among six major memory companies compared; YMTC’s NAND shipment share has reached 14%. The memory sector’s logic combines cyclical price increases and domestic substitution, offering strong earnings visibility within the AI computing chain.
4. Power and Grid
Electricity is becoming an increasingly critical constraint in AI computing expansion. Overseas investment banks estimate that even including on-site self-generation and fuel cells, U.S. data center developers will still face a net power deficit of about 34% by 2028, equivalent to 32 gigawatts. Several global asset management firms have listed power and grid infrastructure as key ‘bottleneck assets’ in their Q4 outlooks.
This assessment also applies to A-share sectors such as power equipment, grid construction, and liquid cooling thermal management. Computing expansion ultimately hinges on ‘electricity’, and the scarcity of power infrastructure may be re-priced by the market.
On the risk side, three variables need attention.
First, the transmission from oil prices and U.S. Treasury yields. International oil prices surged during the holiday, with Brent crude nearing $100 per barrel. If Brent crude climbs further from around $100 to $125 per barrel while U.S. Treasury yields continue rising, it would negatively impact the U.S. macroeconomy, potentially forcing the Fed to intervene and suppressing global risk appetite.
Second, the pace of recovery in liquidity. Market turnover shrank significantly before the holiday; whether turnover can effectively rebound afterward is a key indicator for judging the sustainability of the market rebound.
Third, geopolitical and overseas policy uncertainties may still cause short-term disruptions to the tech sector.
Considering the signals from overseas markets during the holiday, global asset managers’ strategic shifts, and the real supply constraints in the supply chain, the post-holiday A-share AI chain investment logic points to relatively high certainty in optical communications, upstream PCB materials, memory chips, and power grid infrastructure.
*Focus on corporate earnings realization*
However, investors must also clearly recognize that after a significant run-up in the first half of the year, AI computing hardware valuations are no longer low. The market’s next phase will likely shift from ‘valuation expansion’ to ‘earnings realization’.
Amid the interplay of long-term industry trends and short-term volatility, focusing on segments supported by real orders, possessing capacity bottlenecks, and with verifiable earnings may be a more prudent strategy. AI remains the main theme, but the post-holiday rally won’t be broad-based — it will be a victory for hard bottlenecks. Senior Investor, Shek King-chuen
(Investing involves risk; each investor’s risk tolerance varies, so independent judgment is essential. The author may trade securities based on market conditions.)
*Articles published in Economic Times, whether signed or unsigned, represent the authors’ personal opinions and do not necessarily reflect the position of Economic Times. Economic Times serves as a platform for free expression.
Star Financial Research Institute points out:
1. Optical Communications
Optical interconnects are the core channel for data exchange between GPUs in AI clusters. As cluster scale expands and port speeds upgrade, the cost share of optical modules continues to rise. Industry research shows that Ethernet optical module sales doubled in 2024, with nearly 90% growth expected in 2025, and 120% growth projected in 2026.
*Optical communications maintains high growth*
Overseas investment banks estimate that the global optical module market size could reach $68 billion, $131 billion, and $148 billion in 2026, 2027, and 2028 respectively. On the industrial front, 800G remains highly prosperous, 1.6T is entering mass production, and the industry is advancing toward 3.2T and CPO/NPO. Order backlogs for some high-speed optical chip companies have already extended to 2028. Upstream segments such as optical chips and optical components, due to greater difficulty in capacity expansion, may exhibit stronger pricing elasticity.
2. PCB and CCL
The outperformance of Hong Kong’s PCB (printed circuit board) sector during the holiday was no accident. Since the second half of 2025, rapid growth in demand for AI servers and high-speed networks has led to widespread supply tightness in PCBs, IC substrates, and core materials. Shortages began with advanced materials such as low-CTE and low-Dk2 glass fiber cloth and HVLP4 copper foil, then spread to CCL (copper-clad laminates), high-end PCBs, ABF (a high-performance insulating film) substrates, and manufacturing equipment.
*PCB market remains strong*
Data from overseas investment banks show that demand for AI server CCLs will surge from 42 million sheets in 2026 to 131 million sheets in 2028, representing a compound annual growth rate of about 77%; average prices will climb from $165 to $362, with a compound annual growth rate of about 48%. Kingboard Laminate Holdings (01888) has issued multiple rounds of price increase notices this year, and Nan Ya Plastics raised its product prices by another 20% to 25% starting in September, with price hikes fully transmitting downstream. The PCB sector’s rally essentially reflects the realization of a chain: ‘material shortage → capacity strain → price rise → earnings upgrade’.
*Domestic memory catching up rapidly*
3. Memory
Demand for HBM (high-bandwidth memory) and enterprise SSDs (solid-state drives) in AI servers continues to rise, with annual bit demand growth for enterprise SSDs expected to exceed 80% in 2026. TrendForce forecasts that in Q4 2026, contract prices for standard DRAM (dynamic random-access memory) will increase by 10% to 15% quarter-on-quarter, while NAND Flash contract prices will rise by 15% to 20% quarter-on-quarter.
Domestic memory manufacturers are also accelerating their catch-up. CXMT Technology (SHA: 688825) achieved an EBIT margin as high as 82% in Q2 2026, ranking first among six major memory companies compared; YMTC’s NAND shipment share has reached 14%. The memory sector’s logic combines cyclical price increases and domestic substitution, offering strong earnings visibility within the AI computing chain.
4. Power and Grid
Electricity is becoming an increasingly critical constraint in AI computing expansion. Overseas investment banks estimate that even including on-site self-generation and fuel cells, U.S. data center developers will still face a net power deficit of about 34% by 2028, equivalent to 32 gigawatts. Several global asset management firms have listed power and grid infrastructure as key ‘bottleneck assets’ in their Q4 outlooks.
This assessment also applies to A-share sectors such as power equipment, grid construction, and liquid cooling thermal management. Computing expansion ultimately hinges on ‘electricity’, and the scarcity of power infrastructure may be re-priced by the market.
On the risk side, three variables need attention.
First, the transmission from oil prices and U.S. Treasury yields. International oil prices surged during the holiday, with Brent crude nearing $100 per barrel. If Brent crude climbs further from around $100 to $125 per barrel while U.S. Treasury yields continue rising, it would negatively impact the U.S. macroeconomy, potentially forcing the Fed to intervene and suppressing global risk appetite.
Second, the pace of recovery in liquidity. Market turnover shrank significantly before the holiday; whether turnover can effectively rebound afterward is a key indicator for judging the sustainability of the market rebound.
Third, geopolitical and overseas policy uncertainties may still cause short-term disruptions to the tech sector.
Considering the signals from overseas markets during the holiday, global asset managers’ strategic shifts, and the real supply constraints in the supply chain, the post-holiday A-share AI chain investment logic points to relatively high certainty in optical communications, upstream PCB materials, memory chips, and power grid infrastructure.
*Focus on corporate earnings realization*
However, investors must also clearly recognize that after a significant run-up in the first half of the year, AI computing hardware valuations are no longer low. The market’s next phase will likely shift from ‘valuation expansion’ to ‘earnings realization’.
Amid the interplay of long-term industry trends and short-term volatility, focusing on segments supported by real orders, possessing capacity bottlenecks, and with verifiable earnings may be a more prudent strategy. AI remains the main theme, but the post-holiday rally won’t be broad-based — it will be a victory for hard bottlenecks. Senior Investor, Shek King-chuen
(Investing involves risk; each investor’s risk tolerance varies, so independent judgment is essential. The author may trade securities based on market conditions.)
*Articles published in Economic Times, whether signed or unsigned, represent the authors’ personal opinions and do not necessarily reflect the position of Economic Times. Economic Times serves as a platform for free expression.
