【Shanzheng Electronics】Annual Strategy for the Electronics Industry: AI Sparks a New Round of Hardware Inflation, Domestic Computing Power Accelerates Breakthrough

Log in to the Sina Finance app, search 【information disclosure】 to view more evaluation grade details

Investment Highlights

Storage enters a super-cycle. On the AI training side, trillion-parameter-scale models require large-capacity, high-bandwidth storage support; on the inference side, multimodal input and KV cache consumption require a great deal of storage space, so HBM and high-speed eSSD server configurations have become standard. Due to the 2022 semiconductor downturn affecting the supply side, storage vendors keep capital expenditures conservative. Giants such as Samsung and Micron exited niche production capacity and shifted to high-margin areas. HBM supply elasticity is constrained by barriers in 3D stacking process technology, so the supply-demand ratio in 2026 will further slide to 7%. DRAM continues to experience shortages as the impact of HBM crowding out propagates over time, while NAND faces capital and process ramp-up pressures when migrating to higher stacking layers. Driven by the combined effect of three factors—cautious capacity ramp-up, yield constraints, and business-structure adjustments—the storage market shortage situation will persist, and the industry will enter a super-cycle with both volume and prices rising.

Semiconductor industry conditions improve. The explosion of AI compute and accelerating penetration of new-energy vehicles drive a surge in demand for power, analog chips, and passive components. Combined with structural supply shrinkage and capacity constraints on the supply side, this forms a full-chain price-increase cycle. Power devices benefit from the penetration of HVDC technology, increasing the value per unit for SiC/GaN devices. The analog chip inventory cycle reverses; high-end products crowd out capacity, and cost pressures transmit downward, driving a volume-and-price rise. For passive components, higher prices are driven by the increase in precious metals and contract manufacturing/packaging and testing costs, sending the entire industry into an upward price cycle from top to bottom, covering all categories.

Domestic compute drives the accelerated rollout of substitution. Domestic AI large-model inference and AIDC construction release strong compute demand. By 2029, the market size of China’s accelerated server sector will exceed $140 billion. Combined with overseas export controls, this forces breakthroughs in domestically produced chips and manufacturing links. In advanced process technology, the self-sufficiency rate for high-end chips is relatively low. Local wafer fabs increase investment in advanced process capacity layouts, and Semiconductor Manufacturing International Corporation’s (SMIC) 7nm and below advanced process capacity ramps up faster. Advanced packaging has become the key to boosting compute performance. Chiplet and HBM technologies drive upgrades in packaging and testing demand, enabling companies like SPIL and Tongfu Microelectronics (通富等国产厂商) to capture spillover orders. By 2030, the global advanced packaging market size will exceed $79.4 billion. Semiconductor equipment will experience a “demand expansion + domestic substitution” resonance and will fully benefit from capacity expansion by wafer fabs such as SMIC and HuaHong, as well as storage vendors such as ChangXin and Yangtze Memory.

AI innovation drives dual upgrades in PCB materials and architecture. AI servers and 800G/1.6T switches create demand for high-density interconnects and low-loss transmission, pushing PCB technology to leap forward in both architecture and materials. Architecturally, traditional multilayer boards evolve toward higher-layer-count HDI. For example, Nvidia’s GB300 Switch tray is upgraded from a 5+12+5 HDI structure to a higher-layer 6+14+6 HDI structure to support denser interconnects. Connections on the mid-layer of the PCB replace copper cables to achieve high-density interconnects. On the materials side, the shift moves toward low Dk and low Df; M9-grade CCL becomes mainstream for high-end demand. Notable gaps exist for advanced raw materials such as HVLP ultra-low-profile copper foils, Q-copper, and carbon-hydrogen resins. Upgrades in technology and supply-demand gaps jointly support continued increases in value-added across the PCB industry chain, with upstream and downstream companies benefiting fully…

Optics and AI are expected to drive AR glasses into a new interactive terminal. AI technology and optical innovation upgrade AR glasses from auxiliary tools into a “physiology-level” interactive terminal. In 2026, global sales of AI and AR glasses are expected to reach 1.8 million and 0.95 million units respectively, a substantial increase compared with 2024. In optical systems, the waveguide solution is accelerating replacement of Birdbath; diffractive waveguides, with their lightweight advantages, are becoming the mainstream. On the display side, LCoS, Micro-OLED, and Micro-LED form a three-way standoff. Through optimized combinations of optics and display, the industry’s “weight, performance, and battery life” contradictions can be addressed. Domestic suppliers in the AR field have achieved a 142.3% year-on-year growth in shipment volumes, and deterministic innovation dividends are emerging in areas such as optical lenses and display panels.

Investment recommendation: 1）For storage chips, consider: ChangXin Technology,兆易创新等. For semiconductor chips, consider: Cambricon, Yangjie Technology, JieJie Microelectronics, JieHuaTe等. For wafer manufacturing and packaging/testing, consider: SMIC, HuaHong, JCET, and HUIcheng Shares等. For semiconductor equipment, consider: Onto? (芯碁微装), Micro-Guide Nano (微导纳米), Jingce Electronic (精测电子), JingzhiDa (精智达), and Xinyuan Micro (芯源微)等. For electronic components, consider: Copper Crown Copper Foil (铜冠铜箔), Feilihua, Sanhuan Group (三环集团), Fenghua HiTech (风华高科), and Sunlks? (顺络电子)等. For consumer electronics, consider: Zhongrun Optics (中润光学), Bluetek? (蓝特光学), Tianyue Advanced (天岳先进)等。

Risk warnings: Risks that industry conditions fail to meet expectations; risks that technology R&D and mass production fail to meet expectations; supply-chain reconstruction risks; risks from international trade frictions; risks that the pace of domestic substitution fails to meet expectations; risks that the intensity of industrial policy support fails to meet expectations, etc.

【Semiconductors: AI boom helps lift industry conditions, and domestic substitution enters deep water】

Storage: Inference and computing-driven storage enters a super-cycle

Large models head toward a data competition; global CSP capital expenditures accelerate. The development of large models is shifting from parameter competition in the early stage to data competition. Generative AI represented by LLMs is gradually evolving from the base model stage toward more complex forms such as AI agents and physical AI. Overall compute demand enters explosive growth. According to Huawei’s forecast, by 2035, the total compute power for the entire society will grow by 100kx versus 2025 to reach 10²⁷ FLOPS. To meet AI demand, global CSPs increase spending to build infrastructure such as AIDC and compute clusters. According to Trend Force, in 2026 capital expenditures for the world’s top eight CSPs increase to more than $710 billion. Judging from the North America CSP earnings-call guidance for the 2026 fiscal year, AI capital expenditures accelerate further. Taking Meta as an example, Meta raises its capital expenditure guidance to $115 billion–$135 billion, up from 2025’s $70 billion? (70B元) by 50–80%.

Figure 1: In 2026, the top eight CSPs are expected to have Capex exceeding $700 billion

Source: Trend Force, Shanxi Securities Research Institute

Figure 2: AI compute power is expected to grow by 1,000x from 2025 to 2030

Source: Architect Technical Alliance, Shanxi Securities Research Institute

Large models and AIDC construction drive a surge in DRAM and NAND demand. On the training side, trillion-parameter-scale models increase the demand for memory capacity. The faster improvement in GPU highly parallel computation further heightens “memory wall” pressure, raising the need for high-capacity, high-bandwidth storage. On the inference side, multimodal input/output and long-context inference generate large amounts of KV cache, and combined with demand for retaining user-generated content, it further consumes storage space. Driven by both training and inference, compute clusters for AIDC under construction must be paired with high-bandwidth memory HBM and high-capacity, high-speed eSSD to enable low-latency data supply and compute operations coordination, thereby driving an overall increase in storage market prices. According to Counterpoint statistics, in 2026 Q1, memory prices’ quarter-on-quarter increase reaches 80%–90%. Trend Force forecasts that the global memory market size in 2026 will reach $551.6 billion, and the year-on-year growth rate in 2027 is expected to be 53%.

Figure 3: PC and server memory prices surge quarter-on-quarter (USD)

Source: Counterpoint, Shanxi Securities Research Institute

Figure 4: In 2026, the expected memory market value exceeds $500 billion

Source: TrendForce, RuiXinWen, Shanxi Securities Research Institute

HBM demand is driven by AI; supply and demand remain tight in the industry. AI computation requires massive high-bandwidth read/write parameters and KV caches. HBM can break through the bandwidth bottleneck of traditional memory through 3D stacking and TSV, meeting needs for extremely high bandwidth, low latency, and low power consumption. Currently, the market is dominated by SK hynix, Micron, and Samsung. Due to constraints from slow expansion of advanced packaging supporting capacity, high-yield barriers in 3D stacking process technology, and the wafer fab capacity expansion cycle, near-term HBM supply elasticity is insufficient. According to Trend Force, in 2026, HBM capacity supply growth is expected to be 32%. On the demand side, there is a highly concentrated pattern: top AI compute power vendors such as Nvidia, AMD, Google, and AWS together account for 90% of total HBM demand. In the AI up-cycle, to maintain fast-growing compute capacity and inventory safety, CSPs purchase more HBM than actual consumption, further accelerating the supply-demand gap. According to SEMI’s forecast, the HBM shortage is expected to expand from about 5% in 2025 to about 6% in 2026, and further expand to about 9% in 2027.

Figure 5: The HBM supply-demand gap will persist through 2027

Source: SemiAnalysis, The Semiconductor Home of “杰尼龟”, Shanxi Securities Research Institute

Figure 6: In 2026, expected HBM storage supply increases by 32%

Source: TrendForce, Shanxi Securities Research Institute

HBM crowds out general-purpose storage capacity; DRAM shortages continue to evolve. With AI development, explosive growth in HBM directly compresses the market space of general-purpose DRAM. Samsung, Micron, and Hail? (海力士) gradually withdraw from producing niche storage products, shifting capacity to higher-margin areas such as data centers and AI servers. DDR4/DDR5 and niche DRAM capacity shortages emerge for PCs, mobile devices, and traditional servers. According to Trend Force, the share of non-HBM wafer capacity for the world’s top five DRAM vendors will shrink from 81% to 76%, and revenue from non-HBM will reduce from 67% to 59%. On the demand side, new end devices such as AR glasses and foldable-screen phones support consumer electronics demand. OEMs, to secure market share, tend to sign long-term agreements to lock in capacity, further compressing the circulation of storage products in the market. According to Omdia, global server DRAM demand in 2026 is expected to grow 27% to 18,843 MGB, while DRAM demand in mobile devices, PCs, smart cars, and other areas increases 11% to 22,265 MGB.

Figure 7: DRAM market demand maintains high growth

Source: Omdia, ChangXin Technology IPO prospectus, Shanxi Securities Research Institute

Figure 8: HBM share of DRAM capacity, revenue, and bit output ratio

Source: Trend Force, RuiXinWen, Shanxi Securities Research Institute

NAND capacity is squeezed, and SSD supply-demand imbalance worsens. AI inference demand is reshaping the storage architecture. Large-model inference highly depends on massive KV caches and low-latency retrieval of vector data, making SSDs with ultra-high IOPS and microsecond-level access latency the main media for hot-data and warm-data layers. At the 2026 CES, Huang Renxun proposed Rubin’s context memory storage (ICMS) architecture. By establishing local, rack-level storage, it addresses KV cache constraints in AI inference, further boosting SSD demand. According to TrendForce, from 2025 to 2028, AI will significantly drive eSSD growth. However, on the supply side, SSD production is constrained by cautious capacity expansion by storage vendors, a shift in business focus toward higher value-added products, and capital and process ramp-up pressures as 3D NAND migrates to higher stacking layers. The pace of capacity release will lag relatively. TrendForce expects NAND industry capital expenditures in 2026 to grow only 5%. We believe the pace of capacity release lags far behind the demand of Rubin’s ICMS architecture, and the NAND shortage situation will persist.

Figure 9: AI significantly increases eSSD demand

Source: TrendForce, storage notes (存储随笔), Shanxi Securities Research Institute

Figure 10: In 2026, expected sharp reduction in MLC NAND capacity

Source: Trend Force, Shanxi Securities Research Institute

Chips: Dual-engine drivers—AI demand and domestic substitution—push chip cycle improvement

Domestic compute demand releases; domestically produced accelerated compute chips rise fast. As domestic AI large-model inference applications surge, AIDC scales up, and the penetration rate of domestically produced AI chips rises quickly, demand for China’s accelerated compute servers releases rapidly. According to IDC statistics, in 2025H1, China’s accelerated server market size reached $16 billion, up more than 100% year-on-year. It is expected that by 2029, China’s accelerated server market size will exceed $140 billion. Strong compute demand combined with the United States’ export control on high-end chips to China is forcing China to mass-produce accelerated AI chips and achieve domestic substitution of advanced process technology. For example, in the 3 sets of 10,000-card superclusters deployed by Sugon? (中科曙光), they are put into operation at key nodes of the national supercomputing internet and support hybrid deployment and unified scheduling of domestically produced AI cards across multiple brands and technology routes, ushering domestically produced accelerated compute chips into a period of rapid development.

Figure 11: China accelerated computing server market forecast

Source: IDC, Shanxi Securities Research Institute

Figure 12: scaleX 10,000-card superclusters

Source: Sugon (中科曙光), Shanxi Securities Research Institute

High-voltage DC (HVDC) applications and rising costs drive power devices’ price and volume upward together. Driven by improvements in power density, peak power, and equipment reliability, HVDC technology penetrates faster in scenarios such as data centers, new-energy vehicles, and electric vehicles, significantly boosting power device demand. In particular, as data centers evolve from single-cabinet power of 1 MW, the number of high-voltage MOS/IGBT devices used in cabinet AC/DC, PSUs, and external components such as SST, SSCB, DC/DC, and BBU increases by multiples. On the cost side, raw material price increases and cost transmission from higher prices in foundry and packaging/testing stages jointly drive up the value per unit of devices. Starting in 2025, both domestic and overseas manufacturers gradually raised prices. For example, Infineon increases prices for power switches and related chips, while domestically, Sail? (士兰微) raises prices for MOS-class chips and signal diodes and transistors by 10%.

Figure 13: Forecast of global power device market size (million USD)

Source: Yole, Power device products and usage sharing, Shanxi Securities Research Institute

Figure 14: Global power device manufacturers begin raising prices

Source: TrendForce, Semiconductor Practitioner (半导体行者), International Electronic News (国际电子商情), Huarun Micro announcement, Sina Finance, Shanxi Securities Research Institute

Analog chip demand rebounds; domestic substitution continues to deepen. Traditional industrial sectors gradually recover, and high-enthusiasm tracks such as AI compute, intelligent driving, and new-energy vehicles rapidly penetrate, driving structural improvements in demand for core categories such as signal modulation, high-precision sensing, and power management. According to data from Sullivan (沙利文), China’s analog chip market size in 2026 is expected to reach $245.1 billion, up 25% from 2024. China’s development in analog chips began later; domestic substitution rates in high-end sectors like industrial and automotive are low. In 2024, the revenue share of China’s top five analog manufacturers was only 6.9%. But in the long term, as domestic suppliers make technological breakthroughs in analog chips and accelerate customer onboarding, along with continued policy support and demand for supply chain self-reliance and controllability, their market share will keep rising, further supporting sustained improvement in analog market conditions. According to Sullivan forecasts, by 2029, China’s analog chip localization rate is expected to rise from 23.2% in 2024 to 30.8%.

Figure 15: China analog chip market size

Source: Sullivan, Shanxi Securities Research Institute

Figure 16: 2024 China analog IC market share

Source: Naxin? (纳芯微), Shanxi Securities Research Institute

Tighter supply combined with price transmission pushes analog chips into an upward cycle. The analog chip industry has entered a new upward cycle: the supply-demand pattern continues to optimize, the inventory cycle reverses, and cost-side rigid support drives simultaneous increases in both volume and price in the industry. On the supply side, industry capacity experiences moderate recovery; mature-process capacity continues to shrink. For low-end general part numbers, constrained supply elasticity results from earlier capacity adjustments and supply crowding by high-end automotive-grade and AI-related products. The industry inventory cycle shifts from active de-stocking to active replenishment. From the operating indicators of core manufacturers, in 2025, TI and ADI’s inventory turnover rates improve. Shengan? (圣邦股份) and Naxin? (纳芯微) see quarter-on-quarter improvements in gross margin and inventory turnover. On the cost side, upstream metal prices rise, and price increases in foundry and packaging/testing stages transmit cost pressure downward. Analog IC leaders such as TI, ADI, and Infineon adjust prices significantly for their products. Domestically, Biyen? (必易微) and Meixinsheng? (美芯晟) also issue price-adjustment letters. The industry’s price-increase logic transmits smoothly from international leading firms to domestic manufacturers, further solidifying the foundation for the analog chip industry’s upward cycle.

Figure 17: Domestic and overseas leading analog manufacturers’ financial indicators

Source: Wind, Shanxi Securities Research Institute

Foundry and packaging/testing: Domestic compute demand releases; pricing for foundry and packaging/testing rises together

Domestic AI chip demand drives domestic substitution of advanced processes. Global advanced process capacity shows a continuous growth trend. According to SEMI statistics, from 2024 to 2028, advanced process capacity below 7nm increases from 850k wpm by 69% to 1.4 million wpm, while 2nm and below capacity grows from 200k wpm in 2025 to 500k wpm in 2028. Among them, a few top players such as TSMC and Samsung hold the majority share. The share of domestic advanced process capacity remains relatively low, resulting in low self-sufficiency for high-end chips. To meet the huge demand brought by accelerating penetration of domestic AI chips, local wafer fabs increase investment in advanced process capacity deployment, and SMIC’s 7nm and below advanced process capacity ramps up faster.

Figure 18: Rapid expansion of global advanced process capacity (unit: %)

Source: SEMI, Shanxi Securities Research Institute

Figure 19: Technology roadmap of major global foundries

Source: Semi Vision, RuiXinWen, Shanxi Securities Research Institute

Domestic packaging/testing suppliers seize dual opportunities. Moore’s Law is approaching its limit. Advanced packaging achieves high bandwidth, low power consumption, and three-dimensional integration through technologies such as Bump, RDL, TSV, and hybrid bonding, becoming a key for continuing compute performance improvements. According to Yole data, in 2024, the global advanced packaging market size is about $46 billion, expected to exceed $79.4 billion by 2030, with a CAGR of 9.5%. Under high-density integration trends, Chiplet and HBM place higher requirements on testing steps. Foundries and packaging/testing vendors such as TSMC, SPIL, and Tongfu have already introduced high-precision probe testing, distributed online testing, and system-level testing solutions. Currently, the global packaging/testing market is still dominated by TSMC, Samsung, and ASE. Domestic firms such as SPIL? and Tongfu? (长电、通富) and Huatian accelerate deployment and achieve high-end mass production. However, in the AP domain, the localization rate is still low. Under export controls and AI development, domestic packaging/testing firms may benefit from a dual opportunity of order spillover and domestic substitution.

Figure 20: Global advanced packaging market structure (unit: $ billions)

Source: Yole, Shanxi Securities Research Institute

Figure 21: Forecast trend of localization rate improvement in domestic packaging/testing industry

Source: Semiconductor Industry Research, Shanxi Securities Research Institute

Semiconductor equipment: AI improves equipment cycle; domestic demand and substitution resonate

Global semiconductor equipment investment conditions improve. Driven by AI and HPC development, capital expenditures in storage and GPU areas increase significantly. Demand for wafer fab equipment, advanced packaging equipment, and test equipment rebounds, supporting an overall rebound in equipment investment. According to SEMI’s forecast, the global semiconductor equipment market is expected to grow 7.4% in 2025 to reach $125 billion. In 2026, it could further rise to $138 billion, with the year-on-year growth rate increasing to 10%. Under the dual pull from storage and advanced processes, wafer fabrication equipment (WFE) is expected to grow 10% to $122 billion in 2026. Within that, logic foundry equipment is expected to grow 6.6% due to advanced process drivers, while DRAM and NAND benefit from storage capacity expansion, expected to grow 12% and 10% respectively. The packaging/testing segment maintains high growth rates supported by advanced packaging demand driven by AI, smartphones, and HPC. Packaging equipment grows 15% to $6.3 billion, and testing equipment grows 5% to $9.8 billion.

Figure 22: Forecast of WFE market size (unit: $ billions)

Source: RuiXinWen, SEMI, Shanxi Securities Research Institute

Figure 23: Forecast of packaging/testing equipment market size (unit: $ billions)

Source: RuiXinWen, SEMI, Shanxi Securities Research Institute

Domestic equipment demand and substitution resonate. Export controls on semiconductor equipment from the US to China continue to tighten. Combined with the domestic AI compute boom and advanced process and storage expansion demand supported by two memory-listing events, the domestic semiconductor equipment industry faces a “demand expansion + domestic substitution” resonance. On the demand side, the share of expansion in advanced logic and storage production lines increases. Wafer fabs such as SMIC and HuaHong, as well as storage manufacturers such as ChangXin and Yangtze Memory, increase capital expenditures to expand capacity, providing continuous order support that drives purchases of front-end equipment. On the substitution side, capacity shares for mature processes rise. R&D continues to advance in “bottleneck” areas such as lithography, metrology, and inspection. Domestic equipment demand broadens from single points to the whole value chain. Vendors secure supply chain safety by prioritizing purchases of domestic equipment for local wafer fabs, and by product-line integration through M&A. According to CMSP reports, China’s market share for domestic semiconductor equipment rose sharply from 25% in 2024 to 35%, and core processes such as etching and thin-film deposition have already surpassed 40%.

Figure 24: Forecast of China semiconductor equipment industry market size

Source: Yuliang Information, Sullivan, Shanxi Securities Research Institute

Table 1: 2024 semiconductor equipment domestic localization rate

Source: Yuliang Information, Shanxi Securities Research Institute

【Electronic Components: AI innovation drives a full-stack reshaping of the industry chain’s value】

PCB: entering the era of high-layer count, high-speed rates, and new materials

AI compute and high-speed interconnect drive PCB spec upgrades. The large-scale deployment of AI servers, 800G/1.6T switches, and high-speed network infrastructure places higher demands on ultra-high-density interconnects and low-loss transmission, driving simultaneous innovation in PCB architecture and materials. At the architecture level, wiring density and signal/power integrity requirements push PCB connections to replace copper-cable connections and move traditional multilayer boards toward higher layer counts. Taking Nvidia as an example, GB300’s Switch tray upgrades from 5+12+5 HDI to a higher-layer 6+14+6 HDI structure to meet denser interconnect needs. On the materials side, CCL, copper foil, and woven glass cloth upgrade in the direction of ultra-low loss to maintain signal integrity and system reliability.

Figure 25: High-end HDI board structure

Source: Engineer Xiaojie, Shanxi Securities Research Institute

Figure 26: Rubin architecture uses PCB mid-layer connections to replace copper cables

Source: SEMI, XinXiaoHu, Shanxi Securities Research Institute

Table 2: Synchronized innovation in architecture and materials for Nvidia boards

Source: Fourier’s Cat (傅里叶的猫), Shanxi Securities Research Institute

Systemic CCL material upgrades. Copper-clad laminate (CCL) is the core base material for PCBs, with a cost share of 40%. It is mainly formed by copper foil, glass fiber cloth, and resin lamination. Among them, the resin determines base dielectric properties and heat resistance; glass fiber cloth directly affects overall Dk (dielectric constant), Df (dissipation factor), and CTE (coefficient of thermal expansion). The copper foil’s surface roughness (Rz) is key in determining conductor losses at high frequencies. Currently, demand for AI servers and high-speed communications drives PCB evolution toward low Dk and low Df, forcing the CCL material system to make a systematic shift to M6–M9 grades. Under this trend, upstream raw materials see coordinated upgrades: the resin system iterates toward carbon-hydrogen resins and PTFE-modified improvements; glass fiber cloth evolves toward low-dielectric cloth and Q-cloth high-end development; copper foil continues to reduce surface roughness to meet high-frequency, high-speed transmission requirements.

Figure 27: PCB and CCL cost breakdown

Source: SemiVision, Shanxi Securities Research Institute

Figure 28: CCL composition structure

Source: SI simulation workshop, Shanxi Securities Research Institute

Table 3: Panasonic high-speed copper-clad laminate technology evolution roadmap

Source: PCBA design and manufacturing, Shanxi Securities Research Institute

High-end copper foil iterates toward high-frequency, high-speed, and high-density ultra-fine traces. Copper foil is the most important material within CCL, with a cost share of 39%. The copper foil’s surface roughness directly affects loss due to the skin effect of high-frequency signals. With AI server demand, it is upgraded to high-end copper foils such as HVLP and peelable copper to meet needs for high-frequency low loss and high-density fine-line requirements. HVLP ultra-low-profile copper foil, with excellent characteristics of Rz ≤ 0.4μm, can significantly suppress high-frequency losses and has become a key carrier for ultra-low-loss CCL at M9 and above levels. In AI substrate board (载板) applications, traditional copper foil cannot meet process requirements for ultra-thin and ultra-fine traces. Peelable copper as a carrier improves fine-line process yields by providing the process advantage of stable manufacturing of ultra-thin copper layers. According to Global Info Research, global carrier copper foil output value is expected to reach $1.6T by 2031, with a 2025–2031 CAGR of 14.5%.

Figure 29: Forecast of DTH output value

Source: Global Info Research, Shanxi Securities Research Institute

Figure 30: High-frequency, high-speed scenario CCL copper foil roughness

Source: Mitsui Kinzoku, Shanxi Securities Research Institute

Glass fiber cloth iterates toward low dielectric and low thermal expansion. Glass fiber cloth accounts for about 26% of CCL cost. Its core function is to strengthen mechanical strength and control dielectric properties; its performance is moving toward low dielectric constant (Dk), low loss factor (Df), and low coefficient of thermal expansion (CTE). Driven by low Dk/Df requirements, glass fiber cloth has upgraded from conventional E-glass to Low-Dk first/second-generation cloth, and further evolves toward quartz cloth (Q-cloth). Low-CTE glass fiber cloth has extremely low thermal expansion coefficients, ensuring stable PCB dimensions during high-temperature soldering and operation. It meets advanced packaging requirements for ICs in high-thermal environments and has become an essential material for IC substrates. Currently, first/second-generation cloth is the mainstream choice for high-end CCL, mainly used for M7–M8 grade products. Q-cloth, with low Dk (3.4), Df (0.0004), and CTE (0.6), becomes a core material for M9-grade products. Currently, high-end cloth capacity is seriously insufficient. Nippon? (日经纺) expansion is cautious. According to a forecast by FuBang Securities Investment Advisory, in 2026, Low-Dk cloth capacity is about 10 million square meters per month, corresponding to total demand of first to third generation cloth of 18.5 million square meters per month.

Figure 31: Use of glass fiber cloth in AI applications

Source: Nitto B? (日东纺) official website, Shanxi Securities Research Institute

Table 4: Performance parameters of high-end glass fiber cloth by specification

Source: SI simulation workshop, Shanxi Securities Research Institute

Resin upgrades in the direction of low loss and high heat resistance. Resin, as the adhesive and insulating material for copper-clad laminates, determines signal transmission quality via dielectric constant and loss factor. Traditional epoxy resins account for about 18% of CCL cost, but their high Dk/Df can no longer meet high-frequency, high-speed transmission needs. Among higher-grade copper-clad laminate materials, mainstream resins gradually shift toward polytetrafluoroethylene (PTFE), carbon-hydrogen resin (PCH), and polyphenylene ether (PPO), which offer low dielectric constants, high thermal stability, and low water absorption rates. For M8-grade CCL, PPO is currently the mainstream base material; epoxy resin is introduced for modification through interpenetrating polymer networks to balance the heat-resistance shortcomings caused by its thermoplastic nature. For M9-grade products, PPO’s single-layer electrical performance is insufficient, so PCH or PTFE with even lower Dk/Df values is further added as a supplement. Although PTFE has the best dielectric performance, due to yield and processing cost constraints, current mass-production solutions are mainly based on PPO mixed with carbon-hydrogen compounds.

Table 5: Electronic-grade resin performance for CCL

Source: NY Capital, Shanxi Securities Research Institute

Passive components: High-end demand and domestic substitution drive an upswing in industry conditions

Top manufacturers drive passive components markets into a price-increase wave. Affected by price increases in upstream precious metals such as silver, tin, and copper, as well as foundry and packaging/testing price hikes, passive components enter a new round of price increases, showing characteristics of broad coverage, fast transmission, and leadership from major players. The categories experiencing price increases include core products such as tantalum capacitors, MLCCs, chip resistors, and inductors. For instance, within the first year, KEMET under Yageo (国巨) raised the price of tantalum capacitors used in AI servers and automotive electronics twice. PUS? (普思) also adjusts pricing for the resistor series. Domestic leading manufacturers follow suit: Fenghua HiTech raises prices across multiple categories including inductive magnetic beads, MLCCs, and resistors. Sun? (顺络电子) announces a price increase for some inductors and magnetic beads. Rapid price adjustment actions by top manufacturers quickly drive the entire industry into an upward price cycle that covers all categories from top to bottom.

Table 6: Price-adjustment situations of major passive component manufacturers

Source: Xinshi? (芯世相), Semiconductor Frontier (半导体前线), New Ceramic (新型陶瓷), Automotive-grade semiconductor hardware (车规半导体硬件), Semiconductor Chain Pulse (半导体链脉), Youxin Electronics? (有芯电子), Shanxi Securities Research Institute

In emerging scenarios, the unit volume of high-end MLCC per device increases dramatically. Among passive components, the capacitor market share in 2024 reaches 65%, with MLCC having the largest volume and the widest application range. With the development of AI servers and edge AI, demand for high-end MLCCs surges. According to a quote in Kechuang? (科创板日报), Murata, a global leader in passive components, stated that AI will consume a large amount of MLCCs. Nvidia’s GB300 needs to be equipped with about 30k MLCCs; a single AI server rack consumes 440k MLCCs. It is estimated that by 2030, MLCC demand for AI servers will increase by 3.3x compared with 2025. In the new-energy vehicle sector, driven by the three-electric system, smart cockpits, and sensors for autonomous driving, the number of passive components per XEV vehicle increases from 3,000 units in traditional fuel vehicles to 18k–30k. According to CnInfo? (智研咨询) forecasts, global MLCC demand in 2028 is expected to grow to 5.95 trillion units, with a market size of 140.8 billion yuan.

Figure 32: Global MLCC market size (in 100k yuan)

Source: CnInfo Research (智研咨询), Shanxi Securities Research Institute

Figure 33: MLCC usage in servers

Source: Shenzhen Electronic Chamber of Commerce (深圳电子商会), Shanxi Securities Research Institute

Tight capacity combined with export controls supports domestic substitution. The global leaders in passive components are mainly manufacturers from Japan, South Korea, and Taiwan. Among them, Japanese firms such as Murata and TDK hold advantages in high-end MLCCs and inductors, and Taiwan players like Yageo have advantages in high-end resistors. Mainland manufacturers have achieved substitution in mid-to-low-end components, but there remains a sizable gap in automotive-grade and aviation-grade domains. Currently, the industry structure faces multiple structural changes: Japanese leaders such as Murata, TDK, and Taiyo Yuden have already started adjusting business structures, reducing capacity in consumer electronics and general-purpose components. China’s comprehensive prohibition on exports of dual-use items to Japan for Japanese military users and purposes will lead Japanese domestic manufacturers to further cut production. At the same time, domestic leading companies such as Huawei and ZTE increase support for local suppliers, which will continue to drive domestic manufacturers to achieve technological breakthroughs and increase market share in high-to-mid-end passive components.

Figure 34: MLCC market structure

Source: China Electronic Components Industry Association, China Business Industry Research Institute, Shanxi Securities Research Institute

Figure 35: Resistor market structure

Source: Huqin Industry Research Institute, Zhen? (贞光科技), Shanxi Securities Research Institute

【Consumer Electronics: Intelligent glasses market accelerates expansion; long-term demand for optical systems with deep snow】

AI and optical technology drive intelligent glasses as the next generation “physiology-level” interactive terminal. Intelligent glasses, benefiting from a “physiology-level” input that is closer to human vision and hearing organs, are evolving from a single auxiliary tool into an AI device interaction center. By embedding large models and combining multimodal sensing, glasses can coordinate devices in real time—such as cars, AIPCs, and smart home devices—breaking through ecological isolation barriers and enabling cross-scenario interconnection and collaboration. In the short term, AI audio glasses are leading volume as a smartphone accessory, while AR glasses with high-definition display and real-world/virtual fusion are seen as the ultimate form-factor for AI hardware carriers. From products released starting in 2025, companies across multiple fields such as consumer electronics, the internet, and new-energy vehicles are leveraging their own technical or ecosystem advantages to lay out cross-border plans for the AI+AR glasses track.

Table 7: Multi-sector companies’ layout of the intelligent glasses track

Data source: MicrroDisplay, iResearch? (亿欧网), Shanxi Securities Research Institute

Global intelligent glasses market expands rapidly, and the AR segment has domestic advantages. According to IDC statistics, in 2025 Q3, global intelligent glasses shipments reached 850k units, up 74.1% year-on-year. Of this, the shipments of audio and audio-capture glasses were 200k units, up 287.5%. AR/VR has experienced slower growth due to sluggish VR demand. In the domestic market, AR growth is faster. In 2025 Q3, China’s intelligent glasses shipments were 500k units, up 62.3% year-on-year. Within that, the shipments of audio and audio-capture glasses were 1.6T units, up 79.2%. The AR/ER category had the fastest growth, up 142.3% year-on-year, reaching a market share of 83.4%. VR & MR continue to face pressure; shipments fell 61.2% year-on-year. In 2026, global intelligent glasses shipments enter a volume-growth phase. According to WeiShen XR’s forecast, in 2026, sales of AI and AR glasses are expected to reach 16 million and 1.65 million units, respectively—up 9.5x and 2.3x versus 2024.

Figure 36: Global shipments by type of intelligent glasses

Source: IDC, Shanxi Securities Research Institute

Figure 37: China shipments by type of intelligent glasses

Source: IDC, Shanxi Securities Research Institute

Optical systems are the core barrier and the key to solving the breakthrough for AR glasses. Optical systems include the optical and display portions. They account for the largest share of AR glasses’ BOM cost (about 40–50%) and have the deepest technology barriers. On the optics side, the waveguide solution is accelerating replacement of Birdbath as it becomes mainstream. Diffractive waveguides are taking the lead due to semiconductor-level process advantages, performing better than arrays in large-scale mass production and ultra-lightweight form factors. On the display side, the technology pathways show a three-way standoff: LCoS relies on a mature industrial chain and a large field of view, favored by giants such as Meta. Micro-OLED is supported by high contrast and maturity, covering the existing consumer market inventory. Micro-LED still faces bottlenecks in full-color implementation and large-volume transfer, but it is considered the ultimate solution for outdoor displays. By effectively combining optics and display, both high brightness and all-weather wearing needs can be met, resolving the industry’s “weight, performance, and battery life” contradictions.

Table 8: Waveguide-type optical solutions for AR glasses

Source: WeiShen XR, Shanxi Securities Research Institute

Table 9: Five mainstream micro-display technologies

Source: Display Home, Shanxi Securities Research Institute

【Investment Advice and Risk Warnings】

Investment Advice

For storage chips, consider: ChangXin Technology,兆易创新等;

For semiconductor chips, consider: Cambricon, Yangjie Technology, JieJie Microelectronics, JieHuaTe;

For wafer manufacturing and packaging/testing, consider: SMIC, HuaHong, JCET, and HUIcheng Shares;

For semiconductor equipment, consider: Onto? (芯碁微装), Micro-Guide Nano (微导纳米), Jingce Electronic (精测电子), JingzhiDa (精智达), and Xinyuan Micro (芯源微);

For electronic components, consider: Copper Crown Copper Foil (铜冠铜箔), Feilihua, Sanhuan Group (三环集团), Fenghua HiTech (风华高科), and Sunlks? (顺络电子);

For consumer electronics, consider: Zhongrun Optics (中润光学), Bluetek? (蓝特光学), Tianyue Advanced (天岳先进).

Table 10: Main companies’ profitability and valuation, as of February 25, 2026

Source: Wind, Shanxi Securities Research Institute

Risk Warnings

Risk that industry conditions are worse than expected. Fluctuations in the global macro economy may lead to downstream demand—such as AI servers, new-energy vehicles, and consumer electronics—coming in below expectations, thereby affecting orders and capacity release in industries including semiconductors and electronic components.

Risk that technology R&D and mass production fall short of expectations. Core technologies such as advanced semiconductor processes, HBM, Chiplet packaging, and AR optical displays are difficult to develop. If domestic suppliers lag in areas such as technological breakthroughs, yield ramp-up, and large-scale mass production, they may miss opportunities for industry development.

Supply-chain reconstruction risk. Tight capacity in foundry and packaging/testing; fluctuations in upstream raw material prices such as precious metals, glass fiber cloth, and resins; and restrictions on importing equipment and consumables may lead to limited supply across the industrial chain.

International trade friction risk. Changes in trade policies among major economies such as the US, China, and Europe, and upgrades to semiconductor export controls may impact global industrial chain division of labor, technology cooperation, and market demand.

Risk that domestic substitution lags behind expectations. Bottleneck issues in areas such as equipment, materials, EDA, and IP are difficult to resolve in the short term, constraining wafer fab expansion and yield improvements. Downstream customer validation cycles are long, so penetration rates of domestically produced chips and components may rise slower than expected.

Insufficient intensity of industrial policy support than expected. The investment pace of the Third Phase of China’s Big Fund for semiconductors slows down, and subsidies decline, affecting companies’ capacity expansion and R&D spending. Tightened local government招商引资 policies (attracting investment and investment promotion) may also cause projects to land and funding support to be less than expected.

Analyst: Fu Shengsheng

Practitioner registration code: S0760523110003

Analyst: Li Mingyang

Practitioner registration code: S0760525050002

Report release date: March 24, 2026

【Analyst Commitment】

I have been registered with the China Securities Industry Association as a securities analyst. I undertake that, with a diligent professional attitude, I will independently and objectively prepare and issue this report. I am responsible for the content and viewpoints of this securities research report, and I ensure that the information sources comply with applicable laws and regulations. My research methods are professional and prudent, and the analytical conclusions have reasonable basis. This report clearly and accurately reflects my research viewpoints. I have never, and will not in the future, directly or indirectly receive any form of compensation because of the specific recommended opinions or viewpoints contained in this report. I undertake not to seek personal benefits for myself or others by using my identity, position, or information obtained during my professional practice.

【Disclaimer】

This WeChat subscription account is not the publication platform for securities research reports from the Shanxi Securities Research Institute. The content contained herein all comes from securities research reports that the Shanxi Securities Research Institute has formally published. If subscribers use materials from this WeChat subscription account without having full understanding of the complete report, there may be ambiguities in understanding key assumptions, ratings, target prices, and other contents. Subscribers are reminded to refer to the complete securities research reports published by the Shanxi Securities Research Institute, carefully read the various statements attached, information disclosures, and risk warnings, and pay attention to the key assumption conditions under which the related analyses and forecasts can be valid. Also, pay attention to the prediction time cycles for investment ratings and security target prices, and accurately understand the meaning of investment ratings.

Shanxi Securities Co., Ltd. (hereinafter referred to as “the Company”) is qualified to provide securities investment consulting services. The Company will not regard itself as a company’s “obvious” client merely because any institution or individual pays attention to, receives, or subscribes to the push content of this WeChat subscription account. The Company’s securities research reports are based on publicly available information that the Company deems reliable. However, the Company does not guarantee the accuracy or completeness of such information. There are risks in entering the market; investment requires caution. Under any circumstances, the information or opinions expressed in this WeChat subscription account do not constitute investment advice to any person. Under any circumstances, the Company shall not bear any responsibility for any losses caused to any person by the use of any content from this WeChat subscription account. The materials, opinions, and assumptions contained in this WeChat subscription account only reflect the Company Research Institute’s judgments on the date the report is issued. In different periods, the Company may issue reports inconsistent with the materials, opinions, and assumptions in this WeChat subscription account. The Company or its affiliated entities may, where permitted by law, hold or trade the securities or investment targets issued by listed companies mentioned in this WeChat subscription account. They may also provide investment banking or financial advisory services to or seek such services for these companies. Subscribers should consider that the Company may have conflicts of interest that could affect the objectivity of this report. To the extent known, the Company fulfills its disclosure obligations. The copyright of this WeChat subscription account belongs to the Company. The Company reserves all rights to this WeChat subscription account. Without the Company’s prior written authorization, no portion of this WeChat subscription account may be produced, copied, or reproduced in any form, nor redistributed to any other person, nor used in any other way that infringes the Company’s copyright. Otherwise, the Company reserves the right to pursue legal liability at any time.

In accordance with the “Practicing Standards for Issuing Securities Research Reports,” this statement is hereby made to prohibit the Company’s employees from privately providing the Company’s securities research reports to any unauthorized public media or other institutions; to prohibit any public media or other institutions from privately publishing or forwarding the Company’s securities research reports without authorization. Authorization to publish or forward the Company’s securities research reports must be granted by signing an agreement specifying that the authorized institution bears the relevant responsibility for publishing or forwarding.

In accordance with the “Practicing Standards for Issuing Securities Research Reports,” this reminder is hereby made to the Company’s securities research business customers not to forward the Company’s securities research reports to others, and to remind the Company’s securities research business customers and the public investors to use securities research reports published by public media with caution.

In accordance with the “Regulations on Integrity in the Business Operations of Securities and Futures Business Institutions and Their Staff” and the “Implementation Rules for Integrity in the Business Operations of Securities Institutions and Their Staff,” this notice is hereby given to the Company’s securities research business customers to comply with the integrity-in-practice requirements.

A wealth of information and precise interpretation—right here in the Sina Finance app

责任编辑：石秀珍 SF183