The rapid surge in demand for AI servers, high-speed switches, and optical modules is driving a sharp increase in shipments of high-end computing power PCBs, with prices for upstream key materials like copper-clad laminates, copper foil, and various base materials continuing to rise.

Specialty resins, glass fiber, copper foil, high-end PCBs, and packaging substrates form the foundational bedrock of the AI infrastructure buildout. While they may not receive the same level of attention as GPUs, they are indispensable core components for the ongoing iteration of the computing power industry.

Why This Industry Is Seeing Such Divergent Fortunes

The PCB industry is experiencing significant divergence, with leading players in the AI computing power PCB segment seeing their profitability strengthen continuously. Why has a seemingly unremarkable circuit board become one of the most lucrative technology application tracks in the AI era?

The circuit boards used in common household appliances and those that carry NVIDIA GPUs or switch chips are entirely different species in terms of technical sophistication, commercial logic, and valuation. AI computing power PCBs have evolved into a high-tech precision manufacturing industry. Mainland China and Taiwan together dominate the global PCB market, controlling over 80% of production capacity. Within the AI infrastructure boom, while Chinese optical module giants like Zhongji Innolight and XYS are capturing massive profits, leading players in the PCB supply chain are also quietly generating significant earnings within the computing hardware market. However, even within this multi-billion dollar sector, some companies boast gross margins approaching those of the semiconductor industry with soaring profits, while others are mired in heavy-asset challenges and face declining performance.

What drives this divergence? How did China's PCB industry break the monopoly of US and Japanese firms and upgrade into a high-tech sector? Which segments and which companies in the PCB supply chain are critical to the development of global AI computing power? As the boundary between PCB processing and advanced semiconductor processes blurs, who will secure the next ticket in this industrial transformation?

Primary Materials in PCB Structure

Within PCB structure, copper-clad laminate (CCL), prepreg (PP), and electronic copper foil are the three main materials, accounting for approximately 40-70%, 10-20%, and 5% of PCB cost, respectively. The CCL is the core base material of the circuit board, primarily serving to carry circuits and conduct signals, directly determining the quality and performance of the PCB. Deconstructing the CCL reveals it is composed of electronic copper foil, glass fiber cloth, and specialty resin, with copper foil accounting for about 40% of its cost, and glass fiber cloth and specialty resin each about 25%. Prepreg is essentially a semi-finished CCL without the copper foil. Thus, within the overall PCB cost structure, electronic copper foil, glass fiber cloth, and specialty resin account for over 80% of material costs. In essence, PCBs are manufactured through key processes like circuit etching, material lamination, laser drilling, and circuit plating.

Based on process complexity, circuit boards can be categorized into single-layer, double-layer, multi-layer, ultra-high-layer-count, high-density interconnect (HDI), embedded substrate, and IC packaging substrate boards. This complete progression of process sophistication is essentially a microcosm of the evolution of the electronics industry. Remote controls use single or double-layer boards; office computers use multi-layer boards; and high-end smartphones, to pack more components into limited space, already employ HDI or even embedded substrates. However, what truly pushes PCB technology to its limits are AI-era products like NVIDIA AI server motherboards, high-speed switches, and 1.6T optical modules. These require not only ultra-high-layer-count PCBs but also incorporate high-density interconnect technology.

AI Infrastructure Boom Reshapes Industry Profit Distribution

Given that PCBs are ubiquitous and the technical barrier for standard boards is relatively low, this multi-billion dollar industry has long been characterized by a large market size but low concentration. According to authoritative data, there are over 2,800 PCB manufacturers globally, with 1,500 to 2,000 Chinese manufacturers alone having annual revenue exceeding RMB 20 million. Even by 2025, the combined market share of the global top ten PCB manufacturers is projected to remain below 40%. This characteristic is also reflected in the A-share market, where there are about 35 listed PCB companies, with only six expected to surpass RMB 10 billion in revenue by 2025. The vast majority of firms remain entrenched in the mature market for standard boards.

However, the AI infrastructure boom is fundamentally rewriting this industry's profit distribution logic. As demand for AI servers, high-speed switches, and optical modules surges, shipments of high-end computing power PCBs are climbing rapidly, driving up prices for upstream key materials like CCLs, copper foil, and various substrates. The issue is that standard PCB manufacturers face severe product homogeneity and weak pricing power, unable to smoothly pass on rising raw material costs to downstream customers. Consequently, a clear industry divergence is emerging. Leading PCB manufacturers that have successfully entered the high-end AI computing board market are achieving significant simultaneous growth in revenue and profit, thanks to higher technical barriers and stronger product pricing power. In contrast, numerous small and medium-sized manufacturers stuck in the red ocean of the standard board market face sustained pressure on their performance due to the dual squeeze of rising material costs and industry price competition. In the AI era, PCB companies are already on distinctly different development paths.

Breaking Through the Barriers to High-End Computing Power PCBs

Given the lucrative profit potential of high-end AI computing boards, why don't more manufacturers crowded into the standard board red ocean upgrade? The transition is not a simple one, as it is blocked by two core barriers: advanced materials and advanced processes. To understand the process barrier, one must first grasp why AI computing boards must employ ultra-high-layer-count and HDI technology. PCBs in common appliances are like single-story or low-rise buildings, with simple wiring structures satisfied by single, four, or eight-layer boards. However, NVIDIA GPU chips have thousands of pins at their base that require massive data exchange with high-bandwidth memory and switch chips. The limited board space cannot accommodate all traces in a single layer; the only solution is to stack more layers, creating an ultra-high-layer-count structure. Standard PCBs typically have low layer counts, while AI computing boards often have 24, 30, or even over 40 layers, and must employ semiconductor-grade processing standards to create precise circuits with trace widths/spacings of just tens of microns. This is the core circuit etching process.

Standard board etching is relatively crude, commonly using traditional subtractive processes. This involves covering the entire CCL surface with a protective film patterned for the desired circuit, then submerging the board in a chemical etchant to remove the unprotected copper, leaving the basic circuit after the film is removed. However, the traces on AI computing boards are extremely dense and fine, at just tens of microns. During strong acid etching, while removing excess surface copper, the etchant can erode the sides of the fine traces under the protective film—a phenomenon called undercut—easily causing high-frequency signal lines to narrow or break. Therefore, high-end computing boards must adopt a modified semi-additive process (MSAP) comparable to semiconductor chip manufacturing.

The core MSAP process involves: first retaining a thin layer of base copper on the substrate, applying a protective film to cover non-circuit areas and leave trenches for circuit formation, then using precise electroplating to allow copper ions to gradually grow, fill, and thicken along the micron-level trenches. After circuit formation, the protective film and underlying thin copper are removed, finally creating circuit patterns with tens-of-micron precision, completely avoiding the undercut defect of subtractive processes.

After etching, factories need to laminate dozens of CCL layers etched with micron-level precision circuits, along with prepreg and copper foil, in a high-temperature, high-pressure press. Since specialty resin, glass fiber, and copper foil have different coefficients of thermal expansion, even a micron-scale misalignment during lamination can cause multi-layer circuits to interfere or short, rendering the entire board scrap.

Following lamination, to enable electrical connection and data exchange between layers, laser drilling and via plating must be performed, tackling the third and fourth process hurdles. Traditional PCBs use mechanical through-hole drilling, which is completely unsuitable for AI motherboards with over 30 layers, as such holes would occupy and damage significant internal routing space. Therefore, high-end computing boards must use high-energy lasers to precisely drill hundreds of thousands of micron-level micro-vias between specified layers—the HDI process. After drilling, a stringent chemical process is required to uniformly deposit a copper layer on the micron-level via walls, turning insulating holes into conductive pathways and establishing inter-layer connections. If even one via fails during the entire production, the entire board is scrapped.

Understanding this microscopic production flow reveals the industry's harsh reality: extreme precision manufacturing rests upon an extremely high capital barrier. For the PCB industry, without high-end advanced equipment, advanced processes cannot be implemented. To achieve fine etching, precise lamination, and stable drilling, PCB manufacturers must heavily invest in core equipment like top-tier laser drilling machines, high-end lamination presses, and ultra-high-precision exposure systems, each costing millions. A single high-end HDI or ultra-high-layer-count production line can require upfront capital expenditure of hundreds of millions to billions. This is the first major mountain facing small and medium-sized players—a barrier of both process and the hard cash required for equipment, excluding most standard-board-focused firms from the AI computing power track. Therefore, those who secure a ticket to the AI computing power race are typically leading manufacturers with fixed asset investments reaching tens of billions.

This creates a divergent industry structure: while overall PCB industry concentration is below 40%, in the high-end circuit board segment, the global top ten players command over 90% market share. In high-end PCB and HDI sub-segments, Chinese manufacturers hold a competitive edge, with leading domestic PCB firms also ranking in the global top tier. Mainland China and Taiwan firms collectively hold eight of the top ten global PCB manufacturer spots, forming a dominant pattern. However, in higher-tier segments like substrates and IC packaging boards—chip-level packaging materials—Japanese and Taiwanese manufacturers maintain absolute leadership, while Mainland Chinese firms are still striving for breakthroughs.

Key Bottleneck Materials for High-End Computing Power PCBs

Beyond process and equipment gaps, a more fundamental bottleneck for Mainland Chinese manufacturers in top-tier chip substrates is the second major industry barrier: advanced materials. The core base material for PCBs is the CCL, and its three key upstream raw materials—specialty resin, glass fiber cloth, and electronic copper foil—are critical bottlenecks constraining the development of high-end computing power PCBs and advanced packaging. A capacity bottleneck in any of these three materials prevents the effective release of overall high-end computing PCB capacity, making overcapacity difficult to achieve. This logic is entirely consistent with the underlying dynamics previously seen in the optical module industry: the capacity scale of downstream assembly manufacturers is always constrained by the supply capabilities of upstream core material oligopolies.

Before deconstructing the barriers and competitive landscape of these three core materials, it's essential to understand the three core physical metrics that permeate the entire PCB industry. All material upgrades, technological breakthroughs, and enterprise value growth are essentially aimed at optimizing these metrics.

The first is dielectric constant (Dk), which determines signal transmission speed within the board. A lower Dk value means faster signal transmission, effectively reducing nanosecond-level data transmission delays in the AI computing era and shortening AI server response times.

The second is dissipation factor (Df), which determines the degree of signal attenuation during transmission. Simply put, the copper traces inside a PCB are like water pipes, and electrical signals are like water flow. If the performance of surrounding insulating materials like resin and glass fiber is inadequate, they can absorb the signal like a sponge. Such losses are negligible in common electronics, but high-speed PCBs in the AI era demand extremely high signal stability. An excessively high Df value causes rapid signal attenuation, necessitating ultra-low-loss materials for high-end computing boards.

The third is coefficient of thermal expansion (CTE), which determines the operational stability of high-end computing boards. These boards are laminated from over 30 layers of different materials (resin, glass fiber, copper foil) under high temperature and pressure. All materials expand when heated; if CTE values are mismatched, materials expand at different rates during heating, pulling on the hundreds of thousands of micron-level micro-vias and fine traces inside the board, causing misalignment, open circuits, and board failure. Therefore, the CTE of all raw materials for high-end PCBs must be highly matched and stable. Low dielectric constant, low dissipation factor, and low thermal expansion coefficient are the core R&D focus for upstream material companies over decades. Facing the extreme demands of AI computing power for high-speed transmission, traditional resins, glass fibers, and copper foils must undergo comprehensive iteration. Materials that cannot achieve these "three lows" cannot produce high-end computing boards recognized by the industry, even with top-tier processes.

Specialty Resin: Ensuring Low-Loss Signal Transmission and Achieving Full-Chain Domestic Substitution

So, what iterations are required for traditional base materials? Which overseas chemical giants control core pricing power? In which segments have Chinese companies made breakthroughs and achieved domestic substitution? First, consider the core material determining signal loss: specialty resin. Its primary role in PCBs is to bond glass fibers, encapsulate circuit signals, and isolate interference. Standard boards commonly use epoxy resin (FR-4 material), but its molecular structure easily absorbs high-frequency electrical signals. In high-speed transmission scenarios of 112G and above for AI motherboards, epoxy resin acts like a water-absorbing sponge, causing rapid signal attenuation. To solve this, the industry adopted polyphenylene oxide (PPO) as a substitute. Industrial-grade PPO is widely used, but electronic-grade PPO for high-end PCBs requires stringent modification: extremely high purity, very low molecular weight, and specific molecular interfaces at chain ends for stable bonding with other materials. This technological leap from industrial to electronic grade excludes the vast majority of global chemical firms. For decades, synthesis patents and core production capacity for electronic-grade PPO resin have been monopolized by two oligopolies: Belgium-based Solvay and Japan's Asahi Kasei, with global high-end computing board core raw material powder largely dependent on these two suppliers.

However, possessing top-tier raw material powder doesn't guarantee production of top-tier boards; this depends on the core formulation technology of downstream CCL manufacturers. After purchasing electronic-grade PPO powder, CCL companies must add various curing agents, accelerators, and undergo complex chemical blending to produce specialized base materials, eventually processed into stable-performance CCLs. In this field, Japan's Panasonic holds absolute dominance. Its developed Megtron series (the industry-familiar M series), with iterations like M6, M7, and M8 low-loss materials, has become the de facto global industry standard. Specifications from leading companies like NVIDIA often directly specify the use of Panasonic M8 or equivalent materials. This long-cycle validation creates a formulation trust barrier deeper than equipment or process barriers.

Facing upstream powder monopoly and industry formulation standard monopoly, Chinese companies have achieved breakthroughs across the entire chain. At the most upstream raw material level, domestic firms like Dongcai Technology and Shengquan Chemical have successfully broken through the synthesis and purification technology for electronic-grade PPO resin, achieving mass production, breaking the overseas oligopoly on powder, reducing procurement costs for domestic CCL makers, and securing the supply chain for domestic computing base materials at the source. In the CCL formulation field, Mainland Chinese companies represented by Shengyi Technology have, through long-term independent R&D, conquered M8 and M9 grade material technology, with loss indicators meeting international advanced standards. They have passed stringent certifications from core domestic and international server manufacturers, formally entering the high-end AI computing market and capturing share from overseas giants. Industry trends show domestic resin firms and domestic CCL manufacturers are forming industrial synergy. As technology continues to be validated and iterated, this fully domestic chain from raw material powder to board manufacturing is competing head-to-head with overseas giants, sharing in AI industry profits.

Electronic Glass Fiber Cloth: Stabilizing Board Structure, Domestic Capacity Expansion Still Constrained

Solving the signal loss issue from resin gives the CCL basic signal transmission performance. However, to support 30 to 40-layer high-end AI computing motherboards, a core structural skeleton—electronic glass fiber cloth—is also needed. In the CCL structure, specialty resin acts like cement, encapsulating and protecting circuit signals; electronic glass fiber cloth acts like rebar, providing mechanical support for the entire board, with importance comparable to specialty resin. High-speed electrical signals travel through dense glass fiber grids, so high-end glass fiber cloth must have extremely low dielectric constant and dissipation factor to avoid interfering with or attenuating signals. Additionally, glass fiber cloth directly determines the thickness precision and thermal stability of high-end PCBs.

As AI computing board layer counts increase, thickness deviations in each layer accumulate, potentially causing the entire board to exceed thickness specifications. To accommodate more circuit layers in limited space, the structural glass fiber cloth must be extremely thin and light. The diameter of a single fiber in high-end electronic glass fiber is just a few microns, ten times thinner than a human hair. Furthermore, AI servers operate at high temperatures long-term. If the CTE of resin, copper foil, and glass fiber cloth are mismatched, the materials expand and contract at different rates, pulling against each other within the board structure, damaging the hundreds of thousands of micron-level micro-vias and fine traces, causing connection failure. Therefore, high-end glass fiber cloth must balance extreme thinness with extremely low CTE, precisely locking in PCB dimensional stability.

Manufacturing high-end electronic glass fiber cloth has a high barrier, requiring breakthroughs in two core challenges: melting/drawing and weaving. First, special glass ore is melted in a high-temperature furnace and drawn into ultra-fine electronic glass yarn. This yarn is then woven into a high-precision, uniform grid structure using precision looms. Any minor parameter fluctuation during production directly affects material precision and performance. For a long time, Japan's Nitto Boseki nearly monopolized the global high-end electronic glass fiber cloth market, dominating from upstream low-loss yarn to downstream ultra-thin electronic cloth. Most high-end CCL manufacturers designed new products around Nitto's material parameters. In the early stages of the AI industry boom, the capacity and iteration speed of high-end glass fiber materials directly constrained the supply capability of downstream computing motherboards.

Facing long-term monopoly of foundational materials, Chinese companies, after over a decade of deep cultivation, have achieved bottom-up breakthroughs across the chain. In the upstream melting/drawing segment, domestic glass fiber giants like Jiuding New Material have independently developed melting and drawing technology for low-Dk, ultra-low-loss specialty electronic yarn, breaking the overseas monopoly and achieving stable domestic supply of high-end yarn. In the downstream weaving segment, domestic companies like Honghe Technology have conquered precision weaving technology for ultra-thin electronic cloth, gradually reducing dependence on imports for high-end micro-thin cloth. Companies like Sinoma Science & Technology are also advancing the industrialization of low-Dk, low-loss specialty glass fiber, continuously providing domestic substitution solutions for the high-end computing market. However, although the domestic glass fiber industry has achieved technological breakthroughs, capacity expansion remains constrained by overseas factors. High-end electronic cloth production heavily relies on jet looms from Japan's Toyota Industries and Tsudakoma. Lacking these top-tier imported core machines, even with core weaving technology, large-scale capacity expansion remains limited.

Electronic Copper Foil: Enabling High-Speed Data Transmission, Domestic High-End HVLP Foil in Mass Production

After examining the "flesh and bones" of specialty resin and glass fiber cloth, we come to the highest-cost core material in PCB manufacturing: electronic copper foil. Resin protects signals, glass fiber provides structural support, and copper foil undertakes the most critical signal transmission function. All data from GPUs and switch chips is transmitted at high speed through circuits etched from copper foil. Copper foil itself has excellent conductivity, but as AI server transmission rates increase to 112G, 224G, and beyond, a defect of traditional copper foil becomes apparent: excessive surface roughness.

To ensure bonding stability between copper foil and resin, traditional foil production intentionally roughens the surface, creating microscopic protrusions. The industry uses Rz value to measure roughness, with traditional foil typically above 3 microns. In low-speed transmission eras, this roughness had minimal impact. However, in high-frequency, high-speed scenarios, electrical signals no longer travel solely inside the copper wire but as electromagnetic waves along the interface between the circuit surface and the resin—the skin effect. A rough copper foil surface is like a winding mountain road, exacerbating signal collision losses, compounded by the signal-absorbing effect of resin, causing significant high-frequency signal attenuation.

Therefore, high-end AI PCBs must use ultra-smooth copper foil, minimizing surface roughness. Currently, the very low profile (HVLP) copper foil adapted for AI servers has Rz values controlled below 1 micron, with top global labs pushing towards a 0.4-micron physical limit. However, a core engineering paradox exists: the smoother the copper foil surface, the weaker its adhesion to resin. The M8-grade high-end resin used in AI computing boards is itself extremely smooth. Paired with mirror-smooth copper foil, these two smooth materials cannot bond tightly and would delaminate during high-temperature lamination, scrapping the board.

To solve this, the industry achieved a breakthrough through surface chemical treatment technology: after polishing the copper foil to extreme thinness and smoothness, a nanoscale coupling agent is plated onto its surface. This material's molecules bond chemically with the copper oxide layer on one end and undergo cross-linking reactions with the resin during high-temperature lamination on the other end. Macroscopically, the copper foil is smooth and protrusion-free, ensuring smooth high-frequency signal transmission. Microscopically, chemical molecular bonds achieve extreme adhesion and stable locking between copper foil and resin. This is the most technologically challenging and profitable core step in HVLP foil manufacturing, with surface treatment solution formulas being closely guarded secrets that directly determine a company's market competitiveness.

In this field, Japan's Mitsui Mining & Smelting holds an absolute monopoly, not only producing ultra-thin, ultra-smooth foil but also mastering exclusive surface treatment formulas compatible with top-tier PPO resin, monopolizing the core copper foil supply for NVIDIA's high-end computing boards. Other Japanese firms like JX Metals and GF Metal follow closely, with these three Japanese companies essentially locking down the global supply of copper foil for high-frequency, high-speed CCLs.

Facing absolute overseas monopoly, Chinese companies have achieved a collaborative breakthrough. Domestic electronic copper foil manufacturers like Defu Technology and Hongyuan Copper Foil have not only achieved mass production of ultra-thin, ultra-smooth foil in terms of physical processes but have also collaborated with downstream CCL companies to independently develop core coupling agent formulas tailored to domestic specialty resin systems. Currently, domestic high-end HVLP copper foil has achieved mass production and successfully entered the global AI computing core supply chain.

Thus, the three foundational materials determining the performance ceiling of AI computing boards have been outlined: specialty resin ensures low-loss signal transmission, glass fiber cloth stabilizes board structure, and mirror-finish copper foil enables high-speed data conduction. It is the comprehensive domestic breakthrough in these three core materials that has allowed China's high-end CCL and PCB manufacturing industry to break free from low-end red ocean competition and gain the core confidence to impact the global computing power market.

Three Core Materials Form the Copper-Clad Laminate (CCL)

After high-temperature lamination, the three core materials form the CCL. Compared to the fully competitive downstream PCB manufacturing segment, the midstream CCL industry exhibits higher market concentration. While there are nearly three thousand PCB companies globally with concentration below 40%, the top ten CCL manufacturers command the vast majority of market share, a trend even more pronounced in the domestic Chinese market. The core reasons lie in differences in business models and competitive advantages. PCB manufacturers' core barriers are yield control, process optimization, and cost management, focusing on precision manufacturing. CCL manufacturers' core barrier is proprietary material formulation. Subtle differences in resin ratios, curing systems, cross-linking processes, and temperature control curves directly affect the board's signal loss, thermal stability, and lifespan. This formulation barrier, built over long-term accumulation, cannot be quickly replicated by purchasing equipment. Furthermore, after high-end CCLs enter the supply chains of top computing firms like NVIDIA and AMD, they undergo long-term certification lasting one to two years. Stable product performance and supply chain trust mean customers are reluctant to switch suppliers easily. The combination of formulation barriers and long-cycle certification barriers results in concentration and profit margins in the high-end CCL industry far exceeding those in the PCB industry.

Currently, the global high-frequency, high-speed CCL market forms a three-tier structure. US-based Rogers, Japan's Panasonic, and Taiwan's Taimide Technology firmly occupy the first tier, with Taimide becoming a core CCL supplier for NVIDIA's AI server supply chain through continuous iteration of high-speed materials. Within the Mainland Chinese camp, Shengyi Technology, thanks to continuous breakthroughs in formulation technology, stands as one of the few domestic core companies capable of competing directly with overseas leaders.

Ongoing PCB Industry Upgrading

The logic of underlying materials for traditional computing motherboards is now clear, but industry upgrading continues. As mentioned earlier regarding advanced processes, to achieve precise etching of ultra-fine circuits with widths of tens of microns, high-end PCB manufacturers have abandoned traditional subtractive processes, fully adopting the MSAP semi-additive process comparable to semiconductor manufacturing. This process iteration is a core watershed, signifying that the manufacturing logic of high-end PCBs is increasingly converging with advanced semiconductor packaging processes, blurring the boundaries between the two fields.

As AI chip computing power skyrockets, with transistor counts and chip pin counts growing exponentially, the line precision of traditional PCBs can no longer meet the mounting accuracy requirements of top-tier GPUs. Therefore, the industry has introduced an intermediate carrier layer between the chip and the motherboard. Its underlying manufacturing principle is consistent with PCBs, essentially a miniaturized high-end PCB. Leading PCB and CCL companies mastering microscopic precision processing capabilities like MSAP have naturally extended into the semiconductor packaging domain, giving rise to two core new product categories.

The first is substrate-like PCB (SLP), an intermediate product transitioning from traditional PCB to IC substrate, with line precision far exceeding traditional PCBs. It is widely used in high-end smartphones and other consumer electronics, representing the first high-end new track broken through by leading PCB firms. The second is the higher-technology-barrier IC packaging substrate, the core base material directly carrying the chip. Based on insulating materials, IC substrates are mainly divided into two systems: BT substrates, using BT resin as the core material, widely used in mid-to-low-end packaging for mobile phone chips, memory chips, etc.; and ABF substrates, the core packaging material for the AI era, a mandatory base material for NVIDIA GPUs, high-end CPUs, and AI chips.

The core production material for ABF substrates is ABF insulating film. The core patents and the vast majority of production capacity for this material have long been monopolized by Japan's Ajinomoto. All global high-end substrate manufacturers must purchase Ajinomoto's ABF film, meaning this traditional food company controls the core entry point for global AI chip packaging. The competitive landscape downstream in manufacturing is also highly concentrated. Global top-tier ABF substrate capacity is mainly held by a handful of companies like Japan's Ibiden and Shinko Electric Industries, and Taiwan's Unimicron Technology and Nan Ya PCB. Compared to traditional PCBs, the core barrier for ABF substrates is not manufacturing process but stringent long-term certification. Failure of a standard PCB causes minor losses, while failure of an ABF substrate can directly destroy a high-end GPU or CPU chip worth tens of thousands of dollars. Therefore, moving from sample validation to volume production for ABF substrates can take several years, creating an extremely high barrier to supply chain entry. Once a supplier enters a leading customer's supply chain, a long-term, stable partnership forms, which is the core reason the ABF substrate industry maintains high profitability and high concentration.

Giants Fully Deploy Next-Gen Core Substrate—Glass Substrates

Currently, Mainland Chinese manufacturers like Shennan Circuits have achieved mass production breakthroughs in the relatively mature BT substrate field. However, in the higher-technology ABF substrate field, constrained by dual monopolies in materials and manufacturing, they remain in the critical stages of capacity ramp-up and certification by leading customers. Meanwhile, ABF substrates are not the ultimate form of industry development. As AI chip sizes continue to increase and power consumption rises, issues like warpage, heat dissipation, and dimensional stability in ABF substrates become more prominent, pushing towards physical performance limits. Therefore, advanced packaging giants like Intel and Samsung are fully deploying the next-generation core substrate—glass substrates—using glass material with stronger thermal stability and higher dimensional precision to replace traditional organic materials, representing the core direction for future industry upgrades.

The progression from standard PCBs to SLP, from IC packaging substrates to next-generation glass substrates, is both an evolutionary process of continuously shrinking physical dimensions and increasing precision in electronic base materials, and a process of the electronics industry's profit center continuously migrating closer to the chip. The closer a segment is to the core chip, the more it can capture a high share of AI industry profits.

PCBs are not a new industry. For decades, the sector has been labeled as labor-intensive, overcapacity, and low-end manufacturing. However, the AI era is fundamentally reshaping this traditional industry. GPU data transmission rates are iterating from 112G to 224G and 448G; motherboard layer counts are stacking from 8 to 40 layers; trace widths are compressing from hundreds of microns to tens of microns; and the boundary between PCBs and advanced packaging continues to blur. Circuit boards are no longer simple basic components but core convergence carriers integrating materials science, precision manufacturing, advanced packaging, and semiconductor processes.

From electronic-grade PPO resin, ultra-thin glass fiber cloth, and HVLP mirror-finish copper foil, to high-end CCLs, ABF substrates, and next-generation glass substrates—this is not a single-point upgrade of one industry but a collective leap of the entire AI computing power underlying value chain. The common perception is that GPUs are the core of the AI industry chain. However, tracing the underlying industrial logic: without HBM memory, GPUs cannot unleash computing power; without ABF substrates, advanced packaging cannot be realized; without high-end CCLs and PCBs, AI servers cannot operate; and without the three foundational materials of specialty resin, glass fiber, and copper foil, all AI hardware would be impossible.

If the AI infrastructure buildout is likened to a pyramid, what supports the entire structure are these long-overlooked foundational materials, core equipment, and precision manufacturing enterprises. They may not have the ultra-high market capitalization of top tech firms or the dazzling prominence of industry leaders, but they are the core foundation for the steady development and continuous iteration of the entire AI industry.