73 Percent of Capacity, 8 Percent of the High End: CCL Structural Gap

In mid-August 2026, research notes from several institutions described the same industry from two angles that do not appear to belong together. Chinese copper clad laminate capacity has reached 73.3 percent of the global total. Yet the global share held by domestically produced specialty copper clad laminate, the high-frequency and ultra-low-loss grades, is only about 8.3 percent.

The same notes contained growth forecasts that explain the tension. AI server high-density interconnect is projected to grow at a compound annual rate of 29.6 percent, and boards of eighteen layers and above at 33.8 percent, both far above the roughly 8.2 percent growth expected for the PCB industry overall. As new AI platforms move toward M8U and M9 class materials with 24-layer HDI, 44-layer and even higher interconnect structures, the basis of competition in the high-end segment is shifting from production capacity toward material and process capability.

Two Numbers Describing the Same Industry

The gap between 73.3 percent and 8.3 percent is not a measure of manufacturing weakness. It is a measure of where the two segments sit in the value chain.

Standard laminate is a high-volume product manufactured to well-established specifications. Competing in it requires scale, cost control, and reliable delivery, and Chinese producers have become highly effective at all three. Specialty laminate is a different business. High-frequency and ultra-low-loss grades are differentiated products whose performance depends on resin formulation, filler systems, glass fabric structure, and copper foil characteristics, and whose acceptance depends on qualification inside customer designs that may run for years.

Entering that segment is therefore not a matter of building a plant. It requires developing a formulation, proving its stability in production, qualifying it with laminate customers, and then qualifying it again inside finished boards at board manufacturers and their end customers. Each of those steps takes time and produces data that must satisfy an engineering review rather than a purchase specification.

That is why capacity leadership and high-end share can diverge so widely without contradiction. The industry has solved the problem of producing laminate in volume. It is still working through the problem of producing laminate whose electrical properties are predictable enough to be trusted in a 224 Gbps channel.

Why the Growth Rates Diverge by Board Class

The forecast spread between roughly 8 percent industry growth and 33.8 percent growth for eighteen-layer-and-above boards is the clearest available signal of what is actually happening to demand.

AI servers change the relationship between board area and board value. Increasing GPU counts, higher switching bandwidth, and more complex intra-rack interconnect raise the number of signals the board must carry, the power it must deliver, and the data rate those signals must sustain. The result is not simply more board area. It is the simultaneous increase of area, layer count, material grade, and process complexity, and each of those factors multiplies rather than adds to the others.

That is why price increases in this segment are not a conventional commodity cycle. A traditional server might use a well-established multilayer construction. A current-generation AI system moves toward 24-layer HDI, forty-layer-and-above high-speed boards, and system-level interconnect structures with more demanding requirements at every stage. The board’s cost and value rise because the manufacturing difficulty rises, and the difficulty rises in several dimensions at once.

Beyond layer count, the factors that determine value are registration between layers, the number of lamination cycles, back-drill accuracy, yield, and differential impedance control on high-speed channels. At 224 Gbps, holding differential impedance within roughly plus or minus five percent is not an aspirational target; it is the baseline condition for a channel to close. That requirement propagates backwards into material selection, because a laminate whose dielectric constant varies across a panel or between lots cannot support that tolerance regardless of how carefully the board is processed.

What M8U and M9 Material Actually Requires

The progression from M7 to M8, M8U, and M9 reflects a physical limit being approached rather than a marketing ladder being climbed.

As channel rates rise, dielectric loss grows with frequency, and the accumulated attenuation over the length of a high-layer-count channel becomes a first-order constraint on system performance. Material improvements reduce that loss, but the improvements come from formulation changes that alter other properties. Lower dissipation factor typically involves adjustments to the resin system and the filler content. Higher filler loading improves electrical performance and reduces thermal expansion, but it changes flow during lamination, hardness during drilling, and dimensional behaviour through pressing.

The consequence for board manufacturing is that a material change invalidates the existing process window. Press cycles have to be re-established because flow behaviour differs. Drilling parameters have to be adjusted because the material machines differently. Impedance models have to be recalibrated because the dielectric constant has moved. Etch compensation may change because the surface behaves differently. Each of those adjustments requires validation before production, and the validation produces the process capability data that a customer will ask for during qualification.

This is the reason a material supply announcement does not immediately create board capacity. A grade that becomes available does not become usable until a factory has developed and verified a process for it, and that work is specific to the combination of material, stackup, and product. The bottleneck in the high-end segment is therefore not one thing but a sequence: formulation, material production, process development, and customer qualification, each gated by the previous one.

Substitution Is a System, Not a Supplier Change

Domestic substitution in specialty laminate is often described as a sourcing question. In practice it is a systems problem, which is why progress is slower than capacity figures suggest.

Consider what happens when a board manufacturer considers a new laminate grade. The material must be characterised electrically, including dielectric constant and loss across frequency and across lots. A test coupon panel must be built alongside product so that properties can be measured without scrapping boards. The stackup must be re-verified for pressed thickness, since different flow behaviour changes the finished dielectric thickness and therefore the impedance. Registration must be confirmed through the lamination sequence. Drilling and plating parameters must be checked for hole quality and barrel integrity. Finally, the assembled board must be tested, because a material that behaves well in a coupon can still interact with an assembly process in unexpected ways.

Each of those steps produces data, and the data is what allows a customer to approve the change. For an AI infrastructure programme, where the loss budget is consumed deliberately and impedance tolerance is tight, approval also requires that the alternative not merely be close to the incumbent but demonstrably inside the design window across the production range. That is a higher bar than functional equivalence, and it explains why substitution advances fastest where designs retain margin and slowest where they do not.

Suppliers can accelerate the process by arriving with evidence rather than samples. Documented process capability on lamination, registration, drilling, and impedance, measured across lots and expressed as distributions, shortens the customer’s evaluation because it answers the questions the customer would otherwise have to ask. That is what a substantive capability statement contains, and it is why the material and process data a manufacturer maintains becomes a commercial asset during a supply-constrained period.

Where the Layer Ladder Ends

The frontier of board complexity has moved well beyond the range that defined high-end boards a few years ago. Boards up to 104 layers have been discussed in the context of advanced system interconnect, and current AI systems already use structures approaching 78 layers in the most complex backplane and midplane applications.

At those layer counts, the manufacturing problem is dominated by accumulation. Every lamination cycle introduces a small possibility of misregistration, and those errors sum across the stack. Finished thickness raises the aspect ratio of every plated through hole, which makes uniform barrel plating more difficult and makes a thin or voided section more consequential after thermal cycling. Back drilling, required to control via stub on high-speed channels, adds a controlled-depth operation across thousands of holes in a single panel. And because the panel value rises with each completed step, a defect discovered late costs far more than the same defect discovered early, which shifts the emphasis from inspection toward process control.

The corresponding supply chain question is not how many layers a factory can nominally achieve, but whether it can achieve them repeatedly with the required materials. A factory with presses and drills but without qualified access to M8U or M9 class laminate cannot deliver the boards, and a factory with material but without a developed process for it cannot deliver them to specification either. For customers building AI hardware, the practical implication is that supplier evaluation should cover material sourcing strategy and process validation evidence alongside equipment capability, and that this review is best conducted during manufacturing planning rather than at the point of quotation.

The structural gap between 73.3 percent capacity share and 8.3 percent specialty share will narrow as formulations mature and qualifications complete. What will persist is the underlying lesson: in the high-end segment, competitiveness is determined by the ability to combine material, process, and evidence. Capacity alone, however large, does not answer the question a 224 Gbps design asks.

Frequently Asked Questions

Why is Chinese CCL capacity share so high but specialty share so low? Standard laminate is a scale business that has been optimised over decades, while specialty grades require formulation development, stable production and multi-stage qualification inside customer designs, which takes considerably longer to establish.

Why do 18-layer-and-above boards grow much faster than the industry? AI servers raise signal count, power and data rate simultaneously, so demand migrates toward higher layer counts and higher material grades rather than increasing proportionally across all board types.

What does M9 class material change for manufacturing? It changes flow behaviour, drilling response and dielectric properties, which invalidates existing press cycles, drill parameters and impedance models. Each must be re-established and validated before production.

Can a specialty laminate be substituted easily? Not without validation. Electrical characterisation, coupon measurement, stackup verification, drilling and plating checks, and assembly-level testing are all required, and AI designs with tight loss and impedance budgets demand evidence across the full production range.

What should a customer ask about material strategy? Which grades are qualified and in supply, whether alternatives have been electrically validated, how material consistency is characterised across lots, and how process capability on lamination and impedance is documented.