PCB materials for AI servers: fabric, foil, resin

Electronic fabric, copper foil, and resin are rising together because AI servers demand all three at a higher technical grade at the same time. More layers and 800G or 1.6T links force lower dielectric loss in the glass fabric, lower Dk and Df in the resin, and lower surface roughness in the copper foil. Demand lands on a narrow slice of high end capacity, and that slice pulls kiln, drawing, weaving, and qualification capacity away from standard grades. That is why both ends of the range move at once.

What actually changed in the past quarter

Between late August and September 2026, laminated board makers pushed through a dense run of price letters. Copper clad laminate for all thicknesses went up 10 percent. Prepreg of 7628 fabric and above went up 10 percent, and prepreg thinner than 7628 went up 20 percent. One Japanese producer raised selected laminate grades by up to 30 percent from the start of September. A major glass fiber group lifted thick electronic fabric 15 percent and thin fabric 20 percent in the same month.

The moves cluster in one window, and the pattern inside them is worth reading closely.

Material	Producer tier	Move	Effective from
Copper clad laminate, all thicknesses	Laminates	plus 10 percent	late August 2026
Prepreg, 7628 fabric and thicker	Laminates	plus 10 percent	late August 2026
Prepreg, thinner than 7628	Laminates	plus 20 percent	late August 2026
Selected laminate grades	Laminates	up to plus 30 percent	1 September 2026
Electronic fabric, thick	Glass fiber	plus 15 percent	September 2026
Electronic fabric, thin	Glass fiber	plus 20 percent	September 2026
Electronic grade PPO, PPE, OPE, MPPO	Electronic resin	plus 15 to 20 percent	July 2026

Upstream of the fabric, prices moved further and faster. Mainstream G75 electronic yarn quoted at roughly 19,700 to 20,000 yuan per metric ton in August, up nearly 53 percent from the end of May. The 7628 electronic fabric monthly average reached about 10.2 yuan per meter, up roughly 121 percent from the start of the year, and quotes kept climbing into September.

Read that as a materials cost story and it looks like an ordinary input price cycle. Read it as a specification story and it looks different. The three materials are not climbing for the same reason, and they are converging on the same product at the same time.

Why AI servers pull on all three materials at once

The three layers of a copper clad laminate and what each one has to do

A copper clad laminate is the base material a printed circuit board is built from. The structure is simple in principle: copper foil, a resin system, and electronic glass fabric pressed together. Electronic fabric supplies mechanical support and dimensional stability. Resin supplies insulation, bonding, and dielectric behavior. Copper foil carries signal and current.

Mainstream consumer electronics have settled requirements for all three. AI servers do not. Fast SerDes links, GPU interconnect, switch silicon, and 800G or 1.6T optical interfaces keep raising transfer rates, while server boards add layers and line density at the same time. Tolerance for signal loss and impedance instability keeps shrinking.

That leaves high end laminate with three simultaneous problems. Glass fabric must not contribute much dielectric loss. Resin must reach lower permittivity and lower loss tangent. Copper foil must not be so rough that high frequency loss climbs. Fix one and the other two stay open.

The three material families and what the AI server grade demands

The three material families in the board stack:

Layer	Standard grade	AI server grade	The property that decides it
Electronic fabric	Standard E-glass yarn, 7628 fabric	Low-Dk and Low-CTE glass, ultra thin fabric, quartz fabric	Dielectric constant and dielectric loss of the glass
Copper foil	Standard foil with conventional roughness	VLP, then HVLP, then lower roughness grades	Surface roughness against peel strength
Resin	Standard FR-4 epoxy	Low loss modified epoxy, PPO and PPE, hydrocarbon resin, MPO systems	Dielectric constant Dk and loss tangent Df

The three columns are the whole story behind PCB materials for AI servers. Every price letter in the past quarter traces back to a row in this table.

Electronic fabric: the shortage is not a capacity shortage

What the August and September surveys found

Industry surveys through August found an electronic yarn shortage in parts of the high end range, alongside tight electronic fabric. High end electronic yarn was hard to obtain at any price, and some producers ran inventories close to zero. At the same time, producers shifted capacity toward ultra thin fabric, Low-Dk low dielectric fabric, and quartz fabric, all of which carry higher value per unit.

That shift produced a textbook crowding out effect in glass fabric supply. AI servers need more Low-Dk and Low-CTE fabric. High end lines consume kiln time, drawing capacity, and weaving capacity. Standard E-glass yarn and 7628 fabric draw on the same resources, so their supply tightens as the high end expands.

The result is unusual. It is not only high end electronic fabric that rose. High end demand pushed the increase down into standard electronic fabric as well.

Why the fabric line behaves differently from earlier glass fiber cycles

A September survey found the new round of fabric price increases landing well, with supply still short and first tier producers accelerating expansion. That is where this cycle separates from the glass fiber cycles of the past two decades. Those cycles tracked construction, wind power, and automotive demand, all of which are broad and cyclical. Electronic fabric now answers more directly to structural demand from AI computing infrastructure, and that demand does not move with the construction cycle.

The practical consequence is that a fabric producer cannot simply wait out this round. The specification mix has changed, and the mix is what the price is attached to.

Copper foil: what AI needs is flatter copper, not more copper

The skin effect sets the roughness requirement

Standard printed circuit boards care about copper foil conductivity, thickness, and adhesion, and copper foil prices have long been quoted on those terms. Once signal frequency reaches the tens of gigahertz range, surface roughness becomes the dominant variable. High frequency current concentrates near the conductor surface, an effect known as the skin effect. A rough surface lengthens the real path the current travels, and insertion loss rises with it.

That pushed high speed CCL away from standard foil toward very low profile foil, then extremely low profile foil, then grades below that, a shift that every AI server PCB now has to design around. A published estimate puts dedicated high end copper foil demand for AI servers at about 24,000 metric tons in 2026, roughly 260 percent higher than the year before, with 2027 demand potentially reaching about 50,000 metric tons. Those two numbers come from a single analysis, so treat them as direction rather than measurement.

Why HVLP is not a matter of swapping a machine

Extremely low profile foil cannot be produced by moving conventional foil production onto different equipment. It requires tight control of grain structure, surface morphology, and roughness starting from the raw foil, while preserving bond strength to the resin system. A surface too rough raises signal loss. A surface too smooth can fail peel strength.

So the real difficulty sits in the process window between low roughness and high adhesion. That is why conventional copper foil is not necessarily short at all, while high end extremely low profile foil stays persistently tight. The shortage is in a process capability, not in a metal.

Resin: high speed laminate is moving from mechanical to dielectric competition

The FR-4 epoxy system has reached its limit

Resin draws less attention than fabric and foil, but PPO and PPE resin systems are the other core variable in high speed laminate performance. Standard FR-4 leans heavily on an epoxy system, and as data rates climb, the dielectric loss of ordinary epoxy becomes the binding constraint. At M8 and M9 laminate grades, high speed materials increasingly use low loss modified epoxy, PPO and PPE systems, hydrocarbon resin, and other specialty systems to bring Dk and Df down.

In July 2026 one electronic materials producer raised electronic resin prices on its electronic grade PPO, PPE, OPE, and MPPO series by 15 to 20 percent. Laminate producers raised prices again in the same window, which shows electronic resin and laminate moving into the same upcycle.

Resin demand is not a capacity problem, it is a material substitution

The resin layer deserves separate attention because it does not behave like copper foil. It is not mainly a capacity shortage. It is a material system replacement inside high frequency and high speed boards. As M8 and M9 grades and lower loss laminate ramp, the beneficiaries are not epoxy resins in general. They are the high end electronic resin systems that satisfy low Dk, low Df, heat resistance, dimensional stability, and processability at the same time.

That is why the technical upgrade in electronic resin may matter more than the short term price increase. A resin that qualifies for an M8 stack is not a commodity anyone can source from a supplier with spare reactor time.

How the crowded out grade reprices PCB materials for AI servers

The three materials converge in one midstream product, the copper clad laminate, which is why copper clad laminate prices are the clearest transmission point in the chain. Prices have moved repeatedly since March 2026, and by late August a leading producer had issued its seventh increase of the year. Other producers adjusted quotes in parallel, and from September the laminate increase pushed through to board makers.

The chain now reads as a fairly complete transmission path. AI server volume rises, high speed board demand rises, high end laminate demand rises, Low-Dk fabric and extremely low profile foil and low loss resin tighten, high end production crowds out standard capacity, materials prices rise, laminate prices rise, and board makers re-quote. Each link shows up in the price data rather than being inferred.

That is what separates the current round from a plain cost push. Copper and resin feedstock costs are part of it. Product mix upgrade and capacity reallocation are the larger part.

Why upstream materials drew more attention than board making

Through the first half of 2026 and into September, one pattern repeated in capital markets. Upstream PCB materials outperformed parts of mainstream board manufacturing.

The industrial logic is not complicated. Board capacity expansion is comparatively easy to understand and comparatively fast to execute. Effective capacity for high end electronic fabric, extremely low profile copper foil, and low loss resin releases much more slowly. Electronic fabric alone runs from glass formulation through kiln, drawing, and weaving, and each stage carries a long construction cycle. Extremely low profile foil takes long process accumulation. High end resin involves formulation work and customer qualification.

Once AI servers ramp, value concentrates in the links where supply is hard to expand and the technical barrier is high. That is why electronic fabric, copper foil, laminate, and electronic resin move together as a group.

How long the price cycle lasts, and why that is the wrong question

Two variables decide this round, and neither one is the direction of the fabric price.

The first is how fast new high end capacity arrives. Electronic fabric producers are expanding, and some new projects land between late 2026 and 2027. Extremely low profile foil and high speed laminate capacity are being added too. Once effective new capacity releases quickly, tightness in standard grades may ease first.

The second variable matters more. Will AI server material grades keep climbing? If the next platform pushes to higher frequency, more layers, and lower loss, then even with total capacity growing, demand keeps migrating from standard E-glass, standard foil, and FR-4 toward Low-Dk glass fabric, extremely low profile foil, and higher grade low loss resin.

Put plainly, the price cycle can end while the material upgrade cycle continues. The two have to be tracked separately, because they run on different clocks. A fabric mill that adds standard capacity in 2027 may find the grade it added is the grade the market has left behind.

What PCB materials for AI servers mean for buyers

If you buy or specify PCB materials, four decisions follow from the data above.

• Track the specification mix, not the headline price. A 20 percent move on thin prepreg and a 10 percent move on thick prepreg are the same event seen from two positions in the stack.
• Treat fabric, foil, and resin as one procurement problem. A board stack that relaxes on one layer usually has to compensate on another, and the compensation is rarely cheaper.
• Qualify the second source early. Extremely low profile foil and M8 or M9 grade resin both sit behind customer qualification, and qualification time is the real constraint rather than order lead time.
• Watch the standard grade as the leading indicator. When high end capacity crowds out standard production, the standard grade tightens first and the effect shows up in quotes before it shows up in laminate prices.

Where the PCB materials for AI servers data stops being reliable

Three cautions apply to this analysis.

First, the copper foil demand figures for PCB materials for AI servers, about 24,000 metric tons for 2026 and about 50,000 metric tons for 2027, come from a single published analysis rather than a measured industry total. Treat them as direction rather than level.

Second, the source material for this review gives no market size in currency, no capacity tonnage, and no utilization rates for electronic fabric, copper foil, or resin. Published estimates for these categories diverge substantially between providers, so no market value is stated here. The physical quantities and prices quoted above are the ones with a traceable basis.

Third, price movements in this chain pass through negotiated contracts rather than public exchanges. A published increase of 15 or 20 percent is a list move, and realized prices depend on volume commitments, grade mix, and the buyer’s position. Anyone budgeting against these numbers should confirm them with their own suppliers.

Bottom line

The three materials that make up PCB materials for AI servers are not rising together because they share a feedstock. They are rising together because one customer, the AI server, re-rated all three specifications in the same cycle. Electronic fabric has to reach lower dielectric loss, copper foil has to reach lower roughness without losing adhesion, and resin has to reach lower Dk and Df while staying processable. The scarce item is not the material. It is the qualified high end capacity behind it, and that is what the market is repricing.

FAQ

Why are electronic fabric prices rising when capacity is being added?

Because the added capacity is not landing on the grades that are short. Much of the new investment goes to ultra thin fabric, Low-Dk and Low-CTE glass, and quartz fabric, which carry higher value per unit. Those lines consume kiln, drawing, and weaving capacity that standard E-glass yarn and 7628 fabric also need. Standard electronic fabric tightens as a side effect of the high end expansion.

What does AI server demand change about copper foil?

It changes the surface roughness requirement rather than the volume requirement. Standard boards care about conductivity, thickness, and adhesion. Above roughly ten gigahertz the skin effect concentrates current near the surface, so roughness lengthens the signal path and raises insertion loss. That pushes laminate toward very low profile foil. Lower roughness weakens peel strength, so the process must hold both at once.

Why is HVLP copper foil hard to scale?

Because it is a process capability rather than a machine purchase. Producing it means controlling grain structure, surface morphology, and roughness from the raw foil stage onward, while keeping bond strength high enough to survive board fabrication. The window between a surface too rough for signal integrity and one too smooth for adhesion is narrow, so standard foil stays available while the high end grade does not.

Why does resin matter as much as the other two materials?

Because resin sets the dielectric constant and loss tangent of the laminate, and ordinary epoxy has reached its limit. At M8 and M9 grades the dielectric loss of standard FR-4 epoxy binds channel performance, so high speed designs move to low loss modified epoxy, PPO and PPE systems, hydrocarbon resin, and other specialty systems. Qualifying a new resin takes formulation time and customer approval.

What is the crowding out effect in electronic glass fabric?

It is the mechanism that transmits high end tightness into standard grades. AI servers consume more Low-Dk and Low-CTE fabric, and producing it occupies kiln, drawing, and weaving capacity. Those resources are shared with standard E-glass yarn and 7628 fabric. When the high end expands, the standard segment loses access to the same equipment. Both ends of the range then rise in one cycle.

What is the difference between standard copper foil and HVLP foil?

The difference is surface roughness at high frequency, and it is large. Standard foil is specified mainly on conductivity, thickness, adhesion, and cost, and it works at ordinary board frequencies. Very low profile and extremely low profile foil are specified on roughness, because roughness dominates insertion loss once the skin effect is active. They also keep peel strength high, since smooth surfaces bond less readily.

Why is laminate the clearest price transmission point in the chain?

Because electronic fabric, copper foil, and resin all feed into it, so every upstream move appears in one product. Laminate prices are therefore the fastest visible summary of the three material families. That is also why the current round shows so clearly in laminate letters while individual material contracts move more quietly. When all three inputs tighten at once, laminate turns the combined effect into a number.

Why do upstream materials outperform board manufacturing?

Because effective capacity releases at very different speeds. Board capacity expansion is faster and better understood. High end electronic fabric runs from glass formulation through kiln, drawing, and weaving, and each stage carries a long build cycle. Extremely low profile foil needs years of process accumulation, and high end resin needs customer qualification. Value concentrates where supply cannot follow.

Can the price cycle end while material upgrades continue?

Yes, and that is the central distinction in this round. New electronic fabric projects land between late 2026 and 2027, and foil and laminate capacity is being added, which can ease tightness in standard grades. But if the next AI platform pushes frequency higher, demand keeps migrating toward Low-Dk fabric, extremely low profile foil, and higher grade resin. Price cycles revert. Qualification cycles do not.

What should a buyer watch first?

Watch the standard grade rather than the high end one. When high end production crowds out standard capacity, the standard segment tightens first, and the effect appears in quotes before it reaches laminate prices. Then watch qualification schedules rather than lead times, because extremely low profile foil and M8 or M9 grade resin are gated by customer approval. Realized prices depend on volume commitments and grade mix.

Why do price increases differ between thick and thin fabric?

Because the two sit in different parts of the stack. Thick fabric such as prepreg 7628 carries more of the mechanical burden, and thin fabric sits where dielectric behavior matters most. In the current round thin prepreg moved 20 percent against 10 percent for thick grades, and thin electronic fabric moved 20 percent against 15 percent for thick fabric. The steeper thin side move shows where spec pressure concentrates.

Does this mean all PCB materials are in short supply?

No, and that distinction matters for sourcing. Standard copper foil is not necessarily short at all, while high end extremely low profile foil stays tight. The same split applies in electronic fabric, where standard E-glass yarn tightened because high end demand crowded it out rather than because capacity ran out. The constraint sits in qualified high end capacity and specific process windows.