The most provocative recent claim in the PCB supply chain is not about price increases, but that order visibility for some high-layer-count AI boards now extends into the first quarter of 2027. It was followed by reports that lead times for high-end electronic-glass weaving looms have reached 18-24 months, preventing new capacity from coming online quickly.
Both claims have an industrial basis, but they should not be combined into a blanket statement that "all PCBs are in shortage." The genuinely constrained capacity is for customer-qualified AI server boards, accelerator cards, switch boards, and backplanes that use low-loss materials, have many layers, and require repeated lamination cycles. A basic single- or double-sided board, a consumer PCB, and a high-layer-count AI board all belong to the PCB industry, but their production capacity cannot be reassigned at short notice.
1. Orders Visible into Q1 2027 Do Not Mean an Industry-Wide Boom
The assertion that orders extend into Q1 2027 comes mainly from recent institutional and industry research. It is not a uniform disclosure of order backlogs by Chinese PCB manufacturers, and it does not mean that every unit of capacity is locked through the first quarter of 2027. Public information does confirm that utilization is high on some advanced lines and that new projects are still expanding or ramping.
Wus Printed Circuit has disclosed that more than 70% of the overseas data-communications customers for its Thailand operation have completed qualification. Utilization exceeded 90% in the first quarter of 2026, and targeted expansion was expected to begin contributing during the second quarter. SCC's Thailand plant and its Nantong Phase IV project were connected and entered production in the second half of 2025, but were still ramping capacity during 2026. Turning on equipment and achieving stable deliveries remain separated by yield, customer qualification, and product-mix adjustments.
Victory Giant Technology previously disclosed planned annual capacity of 1.5 million square meters for its high-layer-count PCB project in Thailand, with an estimated construction period of 24 months. Its AI HDI project in Vietnam is planned for 150,000 square meters per year with an estimated construction period of 36 months. Factory construction can be accelerated; customer qualification and advanced-product yield cannot be solved simply by adding construction crews.
The shortage, therefore, is not merely a lack of square meters. It is a shortage of capacity capable of high layer counts, low loss, high aspect ratios, and high reliability.
2. Why Can a Weaving Loom Constrain a 20- to 40-Layer AI Board?
An AI server PCB has to carry GPUs, CPUs, switch silicon, and high-speed interfaces. A public filing from Shengyi Electronics gives a typical comparison: AI server PCBs generally have 20-40 layers, versus 8-16 for conventional servers, while high-speed copper-clad laminate can cost 3-5 times as much as ordinary material. More layers require more copper-clad laminate and prepreg per board, together with more lamination, drilling, plating, and impedance-control operations.
Moving one step further upstream, a copper-clad laminate consists of resin, copper foil, and electronic fiberglass cloth. The glass cloth provides insulation, mechanical support, and dimensional stability. High-speed platforms additionally require lower dielectric constant, dissipation factor, and coefficient of thermal expansion. Shengyi Technology's Synamic9GN is a concrete example: its website specifies a Dk of 3.34 and a Df of 0.0014 for high-speed applications including AI servers, switches, storage systems, and routers.
High-end electronic cloth is not created simply by weaving ordinary fiberglass more thinly. Ultrafine glass yarn breaks easily, while warp and weft density, fiber opening, surface uniformity, and microscopic contamination all affect signal loss, impedance consistency, and the glass-weave effect. Toyota Industries states that its JAT910 air-jet loom can weave industrial materials including glass cloth for electronic circuit boards. Institutional research in 2026 placed lead times for high-end weaving equipment at 18-24 months.
An important boundary is needed: the 18-24 month range refers to industry research on high-end electronic-cloth weaving equipment. It is not a common lead time for every PCB machine, such as drills, presses, or plating lines. It is still enough to demonstrate that rapid expansion by a PCB manufacturer cannot conjure qualified Low Dk or Low CTE electronic cloth from nowhere.
3. Grace Fabric Is in Volume Production, Taishan Is Expanding, and China Jushi Is Still Sampling
Grace Fabric Technology has disclosed that its first- and second-generation low-dielectric electronic cloth and its Low CTE cloth have been used by customers in volume since 2025. The average selling price of all its electronic-grade fiberglass cloth was RMB 5.32 per meter in 2025, up 42.25% from RMB 3.74 per meter in 2024. That establishes a change in product mix and market conditions, but it is an average for all electronic cloth, not a separate quotation for specialty Low Dk or Low CTE products.
Taishan Fiberglass has larger expansion plans, but project progress shows that supply will not appear instantaneously. At the end of 2025, its project for 35 million meters per year of specialty fiberglass cloth was 45.25% complete; a separate 35-million-meter low-dielectric fiber-cloth project was 19.01% complete; and a 24-million-meter ultra-low-loss, low-dielectric fiber-cloth project was only 0.04% complete. The disclosed construction period for the 24-million-meter project is 18 months. Even after construction, equipment tuning, yield ramp, laminate matching, and customer qualification remain.
As of March 2026, China Jushi publicly described its low-dielectric fiberglass and electronic-cloth products as undergoing development, qualification, and customer sampling. It can reasonably be identified as a potential source of incremental supply and a domestic evaluation candidate. The available evidence does not support upgrading that description to volume supply into the AI server chain.
This distinction is where the market most often goes wrong: planned annual meters are converted directly into assumptions about next year's deliveries. Nameplate capacity exists in a project schedule. Effective capacity exists only after the material has passed qualification, maintained lot consistency, and been consumed by a laminate producer.
4. The Most Dangerous Sourcing Move Is to Reserve a PCB Factory Without Reserving the Material Code
When AI-board lead times extend, the instinctive purchasing response is to secure board-factory capacity. For high-layer-count, high-speed boards, however, a PCB capacity agreement alone is insufficient. Procurement also needs to lock the copper-clad laminate and prepreg grades, fiberglass-cloth class, copper-foil roughness, resin content, stack-up, and permitted second sources.
"Synamic9GN," "second-generation Low Dk," and "Low CTE" cannot remain labels on a quotation. A material change requires renewed checks of insertion loss, impedance, delay skew, Z-axis thermal expansion, CAF, peel strength, thermal stress, and warpage. Even when Dk and Df appear similar, different combinations of glass cloth, resin, and copper foil can change actual board-level behavior.
Another risk is stocking the wrong material. AI platforms evolve quickly, and customers may revise the stack-up, material grade, or approved-vendor list. Long lead times do not make every long-term order safe. A more robust approach is for the server OEM, PCB fabricator, laminate producer, and electronic-cloth supplier to confirm material codes and changeover rules together, retain the qualified design as a fallback, and then increase the second source's share gradually.
Conclusion: Q1 2027 Is Not the End of a Shortage, but a Test of Expansion Execution
It is more accurate to interpret "orders visible into Q1 2027" as increased visibility for advanced capacity, not as an indiscriminate shortage across the PCB industry. A conventional board maker does not automatically gain the ability to fabricate a 20- to 40-layer low-loss server board simply because AI demand is strong.
The real supply constraint is a chain formed by high-end air-jet looms, Low Dk and Low CTE electronic cloth, CCL formulations, high-layer-count manufacturing yield, and customer qualification. Solving only one link can still leave capacity stranded at the factory, equipment, or sample stage.
The first quarter of 2027 is better viewed as an observation window. Whether new lines at Wus, SCC, and Victory Giant ramp successfully; whether Grace Fabric can sustain higher output; whether Taishan's projects generate effective capacity on schedule; and whether China Jushi advances from sampling to volume qualification will determine whether the constraint eases or propagates further upstream.
Disclaimer: This article is for industry discussion only and does not constitute investment, procurement, or product advice. Orders, capacity, and supply status are subject to formal company disclosures and customer qualification results.
