From Capacity Plans to Reported Earnings
On 7 September 2026 the holding group behind a major PCB manufacturer reported consolidated August revenue of 20.666 billion New Taiwan dollars, up 41.05 percent year on year and a record for the month. Within that figure, revenue from server, optical module and IC substrate business combined was more than three times the level of the previous year. The listed board manufacturer in the same group separately reported August consolidated operating revenue of 3.81 billion yuan, up 18.44 percent.
The distinction between holding company and operating entity matters for reading the numbers, but the signal is the same either way. The effect of AI computing on the board industry has moved from capital expenditure announcements into reported revenue. That shift changes what the industry argues about: not whether demand exists, but which products capture it and at what margin.
Three times growth in a single business line also changes internal priorities. Engineers who were assigned to consumer boards are redirected to server and optical work, and process development budgets follow them. Over a year that reallocation matters more than any single equipment purchase, because it changes which problems the factory knows how to solve.
Server and Optical Board Revenue Grew Faster Than Everything Else
Earlier disclosures from the same manufacturer put first-half AI server board revenue close to one billion yuan and high-speed optical module board revenue above six hundred million yuan, with the optical module part growing elevenfold year on year. Products in those categories have already entered 800G and 1.6T applications, with 3.2T versions in development.
An elevenfold increase in one product line inside a single year is unusual in a manufacturing business. It indicates that the customer base is not expanding gradually but stepping from one generation to the next, and that volume follows design wins rather than price competition. In that environment the constraint on growth is capability and capacity for a specific class of board, not sales effort.
There is a second reason margins hold: the qualification cost falls on the customer as well. Moving a 1.6T board to a new factory requires re-validating impedance, thermal performance and long-run consistency, which takes months and carries schedule risk. Buyers tolerate a higher price more readily when the alternative is a requalification programme in the middle of a product ramp.
Why High-Speed Boards Capture Disproportionate Value
A conventional multilayer board is a mature product with well understood costs and competitive pricing. A high-speed board for an 800G optical module or an AI switch is a different proposition. It requires low-loss laminate, tight impedance control, fine-line imaging, controlled lamination and a level of process control that only a subset of factories can hold in volume.
That process scarcity is what allows the price to hold. When a customer cannot easily move a design to a second source, the supplier’s pricing power depends on how difficult the board is to reproduce rather than on how much material it contains. This is the mechanism by which board capability becomes a source of margin rather than a cost centre.
Price increases along that chain also change how boards are designed. When laminate and copper foil costs rise, designers look for stackups that achieve the same electrical performance with fewer expensive layers, and they scrutinise whether every layer earns its place. Cost pressure, applied through the material chain, ends up reshaping electrical architecture.
The Material Chain Is the Binding Constraint
High-speed boards depend on a supply chain that has been tight for two years. Low-loss laminates, very low profile copper foil, low dielectric glass fabric and high-performance resins have all experienced capacity pressure, and several have seen multi-round price increases. Each of those materials sits upstream of the board manufacturer and is sourced from a small number of qualified suppliers.
That structure explains why capital spending in this sector has concentrated on materials as much as on drilling and imaging capacity. A board factory that can run fine lines but cannot secure qualified laminate cannot serve the customers driving the revenue growth, regardless of how many machines it owns. The reported numbers are, in effect, evidence that some manufacturers secured their material position early.
The same is true of the flexible section inside these modules, where the transition from rigid to flexible carries most of the manufacturing risk. Registration across that boundary is a different problem from registration on a flat panel, and only a limited number of suppliers have production experience with it at the required tolerance.
Optical Modules Are Now a Board Business
The optical module category deserves particular attention because it was historically viewed as a component business rather than a board business. That view is outdated. A 1.6T module contains a dense rigid board, a flex section and an optical engine, with differential impedance controlled to within a few percent across the whole channel and thermal management designed into the copper.
The capital transaction discussed elsewhere in the industry this year, in which an optical module leader bought into a thermal materials company, confirms that the module is now a system with several engineering disciplines inside it. For telecommunications hardware suppliers, that means the addressable content per module has grown substantially.
Substrate capacity is also slower to add than board capacity, because the equipment and the material qualifications are more specialised. That asymmetry means substrate availability can constrain an AI platform even when board supply is comfortable, and it argues for engaging substrate suppliers earlier than the obvious schedule suggests.
Substrates: The Third Leg of the Growth
IC substrates appear in the same revenue line as servers and optical modules because all three serve the same customer: the AI data centre. An advanced package needs a substrate that can carry fine-pitch interconnect between the die and the board, and the substrate faces its own material constraints, particularly in the availability of build-up films.
The convergence is commercially useful for manufacturers who are qualified in more than one of those categories, because a customer designing a rack needs boards, substrates and optical assemblies together. Being able to discuss all three with one quality system and one delivery schedule is a meaningful advantage when programmes are compressed.
There is also a risk in reading a strong quarter as a permanent condition. Capacity added by several suppliers at once can compress prices quickly once the first wave of demand is met. The manufacturers that withstand that cycle are the ones whose process position is difficult to replicate, not the ones with the largest installed base.
What the Numbers Do Not Show
Record monthly revenue is a lagging indicator of design wins that happened a year or more earlier. It says little about the next generation, where 3.2T optical interfaces, higher layer counts and larger package sizes will place further demands on the same process base. The competitive question in eighteen months is whether today’s leading suppliers can qualify the following generation without diluting yield.
That is where investment in quality systems and process control pays off, even though it does not appear in a revenue headline. A factory that improves its measurement discipline during a strong market is better prepared for the next qualification cycle than one that simply adds capacity.
Lead time behaviour is the most reliable indicator available to a buyer. When quoted lead times for high-layer-count boards extend while general-purpose boards remain short, capability is genuinely tight. When both are short, the constraint has moved elsewhere and negotiation leverage returns to the customer.
Geography matters too. Much of the growth described here is concentrated in a small number of manufacturers with the material access, the fine-line capability and the customer relationships to serve AI platforms. Buyers sourcing outside that group should verify capability with production evidence rather than assuming that a general multilayer supplier can step into high-speed work, because the gap between a capable line and a qualified line is measured in months, not in equipment lists.
Reading the Signal as a Customer
For procurement teams, the practical implication is that high-speed board capacity will remain tight and lead times will be governed by capability rather than by machine count. Booking capacity early, sharing forecasts honestly and providing complete design data reduce the risk of being displaced by a better prepared customer.
It is also worth assessing whether a supplier is single-source dependent on a critical laminate or a single fine-line line. Diversification of process, not just of vendor name, is the more meaningful hedge. Buyers who ask how a supplier built its AI hardware capability, and what would break it, learn more than a price comparison will ever reveal.
Finally, the numbers confirm a pattern that has held through several technology transitions in this industry. The board manufacturers that move fastest are the ones that invest in the ability to build the hardest board a customer needs, not the ones that optimise the cost of the easiest one. Volume follows capability in advanced electronics, and it usually arrives faster than capacity plans expect, which is why early preparation and disciplined volume production matter more than a favourable quarter.
