Fiberglass: The Hidden Champion Pulled Into The Computing Power War By AI

Over the past two years, the "wealth-creating myth" of the AI ​​computing power industry has mainly focused on star sectors like GPUs, optical modules, and advanced packaging. However, as information becomes more transparent and the opportunities diminish, we've discovered a "hidden yet highly profitable" industrial chain quietly raking in profits.

 

That chain is fiberglass-more precisely, electronic-grade fiberglass.

 

If this name sounds unfamiliar, you've certainly heard its other name: electronic cloth. In the PCB (printed circuit board) industry, it's known as the "steel bar," the core framework indispensable to every circuit board.

 

How can a "fabric" hold AI servers so tightly? The answer lies deep within this industrial chain.

 

I. From Ore to Circuit Board: What Does Fiberglass Go Through?

 

The raw materials for fiberglass are simple: pyrophyllite, quartz sand, limestone, and other ores. Through high-temperature melting, drawing, and weaving processes, the ore is transformed into an industrial material with diameters measured in micrometers.

 

The type used in the electronics field is called electronic yarn and electronic cloth. Electronic yarn has a diameter of less than 9 micrometers, and the resulting electronic fabric can be as thin as 100 micrometers, appearing indistinguishable from ordinary cloth to the naked eye.

 

But don't underestimate it. From the perspective of its position in the upstream industry chain, high-purity silica sand accounts for 50% to 60% of the total cost of electronic yarn, representing the "first pot of gold" in the entire value chain. Because of this, leading domestic companies like Feilihua have already established a complete closed-loop industry chain from "quartz sand purification-quartz fiber-Q-cloth," firmly controlling upstream bargaining power.

 

After glass formulation optimization and surface treatment, the electronic fabric is delivered to copper clad laminate (CCL) companies. In the cost structure of CCL, the electronic fabric accounts for 25% to 40%. The CCL is then laminated with copper foil, circuitry, etc., ultimately becoming a PCB board, supplied to various industries such as consumer electronics, automobiles, and communications.

 

The depth of this industry chain lies in the fact that the more high-end the product, the more it relies on the support of special glass fibers. When ordinary E-type glass fiber dominated the market, no one paid it much attention. However, once it enters 5G communications, AI servers, and high-speed data centers, low dielectric constant (Low-Dk) and low coefficient of thermal expansion (Low-CTE) electronic fabrics become standard. The material changes, and the entire value chain changes.

 

II. How AI Ignites the "Non-linear" Demand for Electronic Fabrics

Let's start with the structure of AI servers.

 

An NVIDIA GB200 NVL72 rack houses 72 GPUs. Compared to the previous generation H100's eight-card rack, the number of GPUs per rack has increased nearly ninefold. With GPUs doubling, the number of supporting storage and interconnect chips must also double-directly driving a multiple-fold increase in demand for ABF/BT packaging substrates.

 

What is the "skeleton" of the packaging substrate? It's the electronic fabric, especially the low-CTE T-glass electronic fabric. Its function is to maintain dimensional stability of the packaging substrate when the chip operates at high temperatures, preventing warping. Previously, only three companies globally, including Nittobo in Japan, truly possessed mass production capabilities, resulting in a long-term oligopoly.

 

This is the core transmission chain: AI computing power stacking → chip area increasing → packaging substrate layers increasing and size increasing → electronic cloth usage increasing exponentially.

 

On the demand side, the demand for Low CTE electronic cloth is approximately 6.7 million meters in 2025, rising to 18 million meters in 2026, and further expanding to 33.6 million meters in 2027. From 2026 to 2028, the global T-cloth compound annual growth rate will exceed 70%. Even greater variables come from Low-Dk second-generation cloth and quartz cloth: in 2025, the supply-demand gap in the Low-Dk and quartz high-end product market has widened to 25%–30%, with prices increasing by as much as 250%–300% year-on-year.

 

III. Supply-side restrictions at every level: "cloth shortage" will become the new normal. With demand increasing several times over, looking at the supply side reveals a story of "forced slowdown."

 

The first bottleneck: reliance on imported weaving machines.

 

The weaving equipment for high-end electronic fabrics, especially high-end air-jet looms, has long been exclusively supplied by Toyota of Japan. The cycle from ordering to delivery of a single loom is nearly two years. Since 2025, the demand for new looms has exceeded Toyota's annual production capacity, and the shortage will continue in 2026, with a further widening expected in 2027.

 

The second bottleneck: Platinum costs drive up investment barriers.

 

The price of platinum spinnerets, a core consumable in the wire drawing process, rose from approximately 230 yuan/gram in January 2025 to approximately 672 yuan/gram at the beginning of 2026. This has increased the investment cost of a production line by more than 40%, making it almost impossible for small and medium-sized enterprises to expand production.

 

The third bottleneck: Specialty fabric production capacity squeezes out ordinary fabric.

 

High-end products have much higher profit margins than ordinary fabrics, leading companies are shifting their production capacity towards specialty fabrics. Since 2025, the supply of ordinary electronic fabric has been passively contracted, with inventory dropping to "just over a week's worth," and the cumulative increase in price for 7628 type thick fabric reaching 49.4%. Conversely, the oligopolistic structure of "one dominant player and many strong players" has become a ballast for stable supply: China Jushi, Sinoma Science & Technology (Taishan Fiberglass), and International Composite Materials together account for 70% of the domestic production capacity, effectively controlling the market and making supply clearing relatively manageable.

 

IV. The Window for Domestic Substitution Has Been Opened

 

Approximately 85% of the global market share for T-glass electronic fabric is held by Japan's Nittobo, forming a hard barrier. However, domestic companies have gradually achieved technological breakthroughs:

 

(1) Honghe Technology is the only manufacturer in mainland China capable of supplying ultra-thin T-glass fabric, with performance approaching that of Nittobo, and has successfully entered the high-end consumer electronics and AI server supply chains.

 

(2) Sinoma Science & Technology has also achieved technological breakthroughs and is supplying in large quantities.

 

(3) International Composite Materials' low-dielectric glass fiber products have dielectric losses as low as 0.0002-0.0004, and have been applied in cutting-edge fields such as AI servers, 5G-A base stations, and vehicle networking.

 

(4) Feilihua has established a presence in the quartz electronic fabric market, having completed the entire industrial chain from quartz sand purification to Q-fiber fabric. Based on the current pace of capacity expansion, new global production capacity will not be gradually released until after 2027. The reality that Nittobo will not achieve effective supply volume until 2027 provides domestic companies with a window of opportunity of about one year to enter the high-end market.

 

V. Two Minor Reflections

 

Through this "AI breakthrough" in the fiberglass industry, at least two insights can be gained:

 

First, the hidden value of any core technology link is revalued during the AI ​​implementation process.

 

Before AI took off, electronic fabric was almost ignored in the capital market; but when computing power explodes and reaches physical limits, who will provide a stable substrate? Who can solve the problem of "dimensional stability under high power consumption"? Who can achieve better low dielectric properties?

 

Fiberglass is no longer a rustic "building material," but an indispensable part of the entire computing power foundation. From an inconspicuous industrial product to a strategic resource that even Nvidia cannot bypass, this role reshaping process applies to almost every underlying aspect related to AI.

 

Second, supply and demand structure and capacity bottlenecks are the core variables for judging price cycles.

 

The price increase of electronic fiberglass fabric is not simply driven by an event, but stems from three structural constraints: limited raw material supply, reliance on imported high-end weaving machines, and conservative expansion by Japanese companies. Chinese companies have a high degree of certainty in breaking the Japanese monopoly in the specialty fiberglass field and increasing their global market share. For entrepreneurs, whether in the downstream of AI hardware, 5G communications, or automotive electronics, the "de-Japaneseification" of the supply chain is underway. Early positioning in domestic substitution segments of related industrial chains may be one of the most certain trends in the coming years.

 

Finally, no matter how advanced the technology, it ultimately manifests as that inconspicuous piece of fabric on the PCB board. The convergence of fiberglass and AI is not accidental, but an inevitable result of the continuous expansion of the computing power landscape.

 

No one could have predicted that a basic material used for laying circuits would become a sought-after commodity for giants like Nvidia and Apple. This supply-demand gap shows no signs of significantly narrowing before 2027. Within this window of opportunity, whoever can enter this low-profile yet crucial industrial chain will stand on the high ground of becoming another AI "water seller."