Nan Ya Printed Circuit Board Corporation (8046.TW): The Quiet Substrate Maker Behind the AI Boom

I. Cold Open & Roadmap

In early September 2026, shares of 南亞電路板 Nan Ya Printed Circuit Board Corporation changed hands on the Taiwan Stock Exchange at roughly NT$1,195, valuing the company at about NT$772 billion — call it US$25 billion. Over the preceding fifty-two weeks the stock had traded as low as NT$177.50 and as high as NT$1,415.[^1]

Read that range again. Not the price. The range. In a single year, this company's equity has traded across an eight-fold band. That is not a growth stock finding its footing or a value stock re-rating. That is a violent repricing of what the market believes a business is, and it happened to a company most people outside Taiwan's electronics supply chain have never heard of.

Here is what Nan Ya PCB — 南電 to everyone in Taipei — actually does. When Nvidia builds a Blackwell-class accelerator, or AMD builds a data-center GPU, or Broadcom builds a custom AI chip for a hyperscaler, the silicon die does not get soldered directly onto a server motherboard. It cannot. The connections coming off a modern processor are spaced a few tens of microns apart; the wiring on a motherboard is measured in hundreds of microns. Something has to sit in between and translate. That something is an IC package substrate — a slab of resin, glass fiber and copper, roughly the size of a coaster, with a dozen or more layers of circuitry laminated inside it. Nan Ya PCB is one of five companies on earth that can make the hardest version of them at scale.

And here is the uncomfortable part. This same company, making this same indispensable component, saw its equity lose roughly nine-tenths of its value between 2022 and 2024. In the year ended December 2024, Nan Ya PCB generated NT$32.3 billion of revenue, a gross margin of approximately 1.1%, an operating loss of NT$1.27 billion, and net income of NT$204 million — earnings of NT$0.32 per share.¹ In June 2024, the whole company was worth NT$102.2 billion.² It is now worth more than seven times that.

So which is it?

That is the question this story is built around, and it is a harder question than the current narrative admits. The bull framing — advanced by Morgan Stanley among others, and echoed enthusiastically across Taiwanese financial media — is that advanced substrates have "broken free from traditional PCB cyclicality," that the industry has entered a structural, decade-long shortage driven by AI infrastructure, and that companies like Nan Ya PCB have therefore earned a permanently higher multiple.³ The bear framing is simpler and rests entirely on this company's own record: it has been here before, it responded the same way last time, and the result was a 90% drawdown.

At NT$1,195 against trailing twelve-month earnings of roughly NT$8.51 per share, the stock trades near 140 times trailing profits. Measured against the NT$3.01 the company actually reported for the full year 2025, it trades near 400 times.[^1]¹ Either number describes a security priced for a decade to go right.

What follows is an attempt to test that. We will explain what a substrate is and why a Japanese seasoning company sits at a chokepoint of the AI supply chain. We will go back to a PVC conglomerate in 1985 and explain, briskly, how a plastics business ended up in semiconductors. We will spend real time on the period nobody in the current bull case wants to discuss — 2016 through 2019, when this company lost money in three years out of four — and then on the 2020–2024 supercycle and bust, because that episode is the single most useful piece of evidence anyone has about how this management team behaves at the top of a cycle. We will look at the AI re-rating of 2025 and 2026 on its actual numbers rather than its headlines. We will war-game the competitive structure. We will examine a genuine governance transition at both the company and its parent group, and a controlling shareholder that sold into the rally. We will size the glass-substrate story honestly. And we will end with the two or three things that would actually tell an investor whether the golden-decade thesis is real.

Start with the object itself.

↑ Back to Top

II. What an IC Substrate Actually Is, and Why It's the Story

Hold a modern AI accelerator in your hand and you are holding a sandwich. At the top, a heat spreader. Below it, one or more slivers of silicon — the processor die, and increasingly stacks of high-bandwidth memory beside it. And beneath all of that, doing the least glamorous and most underestimated job in the package, sits a green rectangle of laminate: the substrate.

The substrate's job is translation. Think of the silicon die as a city where every street is one meter wide, and the motherboard as a country where every highway is a hundred meters wide. You cannot merge one into the other directly. The substrate is the interchange — a multi-layer structure that takes tens of thousands of impossibly fine connections coming off the bottom of a chip and fans them out, layer by layer, into something a circuit board can physically accept. It also has to deliver clean power to a device drawing a thousand watts, carry signals at frequencies where a copper trace behaves less like a wire and more like an antenna, and do all of it without warping when the package heats from room temperature to ninety degrees and back, thousands of times.

There are, broadly, three tiers of this business, and the distinction between them is where nearly all the economics live.

At the bottom sits the ordinary rigid printed circuit board — the motherboards, graphics cards and server backplanes most people picture when they hear "PCB." It is a real business, but it is a commodity one: thousands of competitors, low barriers, thin margins.

In the middle sits the BT substrate, named for the bismaleimide-triazine resin it uses. These are the substrates under memory chips, mobile application processors, camera sensors and system-in-package modules. Harder than a PCB, easier than the top tier.

At the top sits the ABF substrate, and ABF is where the story is. The letters stand for Ajinomoto Build-up Film, and yes — that is the same 味の素 Ajinomoto that invented monosodium glutamate and sells soup stock in Japanese supermarkets. In the 1990s, the company's researchers were applying amino-acid chemistry to epoxy resins. What emerged was an insulating film with an unusual combination of properties: very low thermal expansion, ease of laser processing, and a surface that accepts direct copper plating. It was first adopted by a major semiconductor manufacturer in 1999 and has since become, in Ajinomoto's own words, the material found "at the heart of most of the world's personal computers."⁴

The build-up process is worth understanding because it explains the industry's cost structure. Rather than drilling holes through a finished stack, the manufacturer starts with a core and builds outward, one layer at a time: laminate the film, drill microscopic vias with a laser, plate them with copper, pattern the circuitry, repeat. Every added layer multiplies the opportunity for defect. A single flaw in a single layer scraps the entire substrate — and by then most of the cost has already been incurred. That is why yield, not nameplate capacity, is the number that actually governs supply. And it is why the industry's tooling and process knowledge take years, not quarters, to transfer.

Which brings us to the dependency worth naming explicitly. There is essentially one qualified source for the film itself. Ajinomoto's position in ABF dielectric is close to a monopoly, and it is a monopoly nobody has successfully attacked in a quarter-century, because the qualification burden falls not on Ajinomoto's customers but on their customers' customers: to swap the dielectric, a chip designer would have to requalify an entire package. Every ABF substrate maker in the world, Nan Ya PCB included, therefore builds its highest-value product on top of a single-vendor input it does not control. That is not a hypothetical risk. It is a permanent structural feature of the industry, and it caps how much bargaining power any substrate maker can accumulate.

The final piece of context is why substrates became a bottleneck at all. Through the 2010s, the ABF market was tied to PCs and networking — steady, unexciting, chronically oversupplied. AI changed the demand curve in a way the industry did not anticipate. An AI accelerator package is not merely a bigger CPU package; it is dramatically bigger. Larger die, plus multiple stacks of high-bandwidth memory placed alongside, means substrate area per unit rises sharply, and layer counts rise with it. A high-end AI substrate can consume several times the panel area and processing steps of a PC substrate. So even flat unit growth in accelerators translates into steep growth in substrate area demand — and area, not units, is what fills a factory.

There is a commercial consequence to that physical fact worth making explicit, because it is what turns a boring laminate business into a violently operationally levered one. A substrate factory sells area. Its costs — equipment depreciation, clean-room overhead, plating lines, laser drills — are almost entirely fixed against that area. So the profit of a substrate maker is not really a function of how many chips the world ships; it is a function of how much of its panel capacity is full, and at what price. Fill the lines and the incremental panel drops to the bottom line at something close to full contribution. Empty them by twenty percent and the same factory produces losses. There is very little in between, which is why this industry's margins do not glide — they hinge.

The direction of travel makes the leverage sharper still. Management has described ABF specifications heading toward formats exceeding 140mm on a side with thirty-plus layers after 2028, a step change from the substrates that dominated the PC era.⁸ Bigger and more layered means fewer units per panel, more process steps, more chances to scrap, and far more capital per unit of output. The good version of that story is higher revenue and higher barriers per wafer of demand. The bad version is that the fixed-cost base you must keep full gets larger every generation.

That is how the substrate joined CoWoS advanced packaging and HBM memory on the short list of things actually constraining AI hardware output. Not because it is exotic, but because it is physically large, difficult to yield, and takes eighteen months to add.

Which raises the obvious question: how did a company founded to make PVC pipe end up here?

↑ Back to Top

III. Origins: A Division of a Plastics Conglomerate (1985–1997)

台塑集團 Formosa Plastics Group is, in the Taiwanese imagination, less a company than a piece of national infrastructure. 王永慶 Wang Yung-ching and his younger brother 王永在 Wang Yung-tsai built it from the late 1950s out of PVC resin, and then did what great industrial families do: they integrated relentlessly, downward into feedstocks and outward into anything that could consume their own output. By 2009, after fifty years, the group had reached number 323 on the Fortune Global 500.⁵

Wang Yung-ching's management philosophy is often summarized in Taiwan by a phrase the company still uses: 追根究柢 — roughly, "get to the bottom of one thing." It meant something specific and slightly ruthless. Cost was not a budget line to be negotiated; it was a physical quantity to be decomposed until you understood every gram of material and every second of labor inside it. Formosa's culture prized diligence, frugality and vertical integration in a way that shaped every subsidiary it ever spawned — including this one.⁵

The path into circuit boards ran through materials, not electronics. 南亞塑膠 Nan Ya Plastics, one of the group's four pillars, already made epoxy resin, glass fiber cloth and copper-clad laminate — which is to say, it already made the physical substance of a printed circuit board. In 1985, as Taiwan's industrial policy pushed the island up the value chain from plastics and textiles toward electronics, Nan Ya Plastics formed a PCB business unit and began manufacturing boards.² The logic was straightforward: if you already make the laminate, making the board is one step of forward integration, not a leap into a new industry.

For twelve years it stayed a division. Then, in 1997, the board of Nan Ya Plastics approved establishing Nan Ya PCB Corporation as a separate reinvested company — legally independent, operationally and financially still very much part of the group.² Expansion followed quickly. In 1999 the company capitalized a Kunshan subsidiary in mainland China with US$29.8 million, and in 2000 it began manufacturing wire-bonding substrates — its first real step above the commodity board business. In 2002 it established Nan Ya PCB America. In 2006 it listed on the Taiwan Stock Exchange under ticker 8046.²

The 2006 listing is worth a beat of its own, because of what it did and did not change. Public shareholders got a liquid claim on the economics and quarterly disclosure. They did not get control, or anything close to it: the parent retained roughly two-thirds of the shares, a level it held for the next two decades. In practice that meant Nan Ya PCB was a public company with a private company's governance — which, in a business that would go on to require enormous and repeatedly mistimed capital commitments, is not a neutral detail.

Two features of that origin story still matter enormously three decades later, and both cut in both directions.

The first is vertical integration. Nan Ya PCB's parent makes the copper-clad laminate, the epoxy and the glass fiber cloth that go into its products. In a period when substrate raw materials are tight — and in 2026 they are — that is a genuine and unusual advantage, and the parent has said so explicitly.⁶ It is also, structurally, why the parent has an interest in this subsidiary that goes beyond the dividend.

The second is that the company has never been an independent capital allocator in the way a standalone technology firm is. Its chairmen have been group executives. Its expansion decisions have been shaped by group priorities. Its capital comes with group backing — and, as we will see, so do its factory leases. For an investor, that is neither good nor bad in the abstract. It simply means that when you evaluate management's capital discipline, you are evaluating a conglomerate's approach to a cyclical subsidiary, not an entrepreneur's approach to a business.

And by the mid-2010s, that conglomerate had a problem. Being able to make the board was not the same as being paid well for it.

↑ Back to Top

IV. The Lost Decade: Rigid PCBs, Intel, and Chronic Thin Margins (2001–2019)

Walk the floor of a substrate plant in Taoyuan in 2017 and nothing about it looks like a failing business. The clean rooms are immaculate. The laser drills are Japanese and expensive. The engineers are good. Panels move through plating, lamination and inspection on a line that cost billions of New Taiwan dollars to build, producing a product only a few dozen factories on earth could make at all.

And every one of those panels was, that year, being sold for less than it cost to produce.

There is a version of this company's history that gets told at investor conferences, and it goes like this: Nan Ya PCB moved into flip-chip substrates in 2001, embedded substrates in 2013, system-in-package substrates in 2016, and ABF substrates from 2019 — a steady, disciplined climb up the technology ladder toward the AI moment.²

The milestones are accurate. The implication is not.

Look at what the ladder actually produced. In the year ended December 2016, Nan Ya PCB reported revenue of NT$29.2 billion and a negative gross margin — it cost the company more to manufacture its products than customers paid for them — and a net loss of NT$694 million. In 2017, revenue fell to NT$26.6 billion and the net loss widened to NT$1.96 billion, or NT$3.03 per share. In 2018, revenue recovered to NT$28.8 billion and the company lost NT$584 million. In 2019, after four years of technology upgrades and a return to revenue growth, Nan Ya PCB earned NT$308 million on NT$31.1 billion of sales — a net margin of one percent, and earnings of NT$0.48 per share.¹

Three losses in four years. Then a year of barely breaking even. That is the base rate.

It is worth sitting with the mechanism, because it is the same mechanism that will matter in 2028. The company had genuine advanced capability by 2001 — flip-chip substrates were, at the time, the leading edge, and they qualified it into the high-value packaging supply chain of the PC era. But capability is not the same as economics. Through the 2010s, the mix of what Nan Ya PCB actually shipped drifted back toward lower-value rigid boards and commodity chipset substrates, in an industry that had far too much capacity chasing a PC market that had stopped growing. Revenue rose. Value did not.

This is the single most important pattern in the company's long-run record, and it recurs: Nan Ya PCB has repeatedly demonstrated that it can make the difficult thing, and has repeatedly struggled to be paid well for making it. Technical qualification is a ticket to compete, not a claim on profit. In a business where five companies can all make an acceptable substrate and customers requalify every design cycle, the leading-edge product of 2001 is the commodity of 2011, and the margin belongs to whoever has the newest capability at the moment demand exceeds supply — not to whoever had it first.

Depreciation compounded the problem. A substrate line is enormously capital-intensive; equipment gets depreciated over years whether or not the line is full. In 2019 the company carried NT$2.64 billion of depreciation and amortization against NT$385 million of pre-tax income.¹ When utilization fell, there was nothing to absorb the fixed charge, and margins went straight through zero.

Management's response, from roughly 2019 onward, was the pivot that set up everything since: rather than defend share in legacy boards, lean into the highest-difficulty tier. In 2019 the company began building an ABF substrate production line at its Kunshan campus, which started manufacturing in 2021; a second line followed at the Shulin 樹林 campus in Taiwan in 2022.² Strategically, this was the right call, and it deserves credit — it is genuinely hard for a company bleeding money to fund a capability build.

But notice the timing, because it becomes the central fact of the next section. The decision to build ABF capacity was made in a downcycle. The capacity arrived in an upcycle. And the next set of decisions — the ones that determined whether shareholders kept any of the upcycle's profits — were made at the top.

For an investor, the honest reading of 2001–2019 is this: it establishes that Nan Ya PCB is capable of surviving a multi-year value-destroying downturn without dilution or distress, which is real and worth something. It also establishes that this company's long-run normalized earning power, absent a shortage, has been close to nothing. Any thesis that treats current margins as the new baseline is implicitly arguing that a structural break has occurred. That is a claim requiring evidence, and the company's own first supercycle is where we should look for it.

↑ Back to Top

V. The First Supercycle and Its Bust: 2020–2024

In the spring of 2021, the phrase circulating in Taipei brokerage notes was 載板三雄 — the "three heroes of substrates." Unimicron, Kinsus and Nan Ya PCB had spent the previous decade as unloved industrial cyclicals. Suddenly they were the hottest thing on the exchange. The pandemic had done something to demand that nobody had modeled: everyone bought a laptop at once, cloud providers built out capacity at once, and — for a strange eighteen months — cryptocurrency miners bid for the same GPU packages as everyone else. Substrate lead times went from weeks to a year.

The financial results were extraordinary. Revenue rose from NT$38.5 billion in 2020 to NT$52.2 billion in 2021 and NT$64.6 billion in 2022. Gross margin, which had been negative as recently as 2017, reached 40% in 2022. Net income hit NT$19.4 billion, or NT$30.05 per share — the best year in company history by a wide margin.¹

Now the important part: what did management do with the money?

They spent it on capacity. Cash capital expenditure ran NT$7.4 billion in 2020, NT$8.5 billion in 2021, NT$16.9 billion in 2022, and NT$11.8 billion in 2023 — roughly NT$28.7 billion in the two years spanning the cycle peak and the year immediately after it.¹ Unimicron and Kinsus did the same thing at the same time. So did Ibiden. So did AT&S. The entire industry, having just been humiliated by its inability to serve demand, resolved simultaneously never to be caught short again.

Then demand normalized.

The bust was fast and it was brutal. Revenue fell to NT$42.3 billion in 2023 and NT$32.3 billion in 2024 — a decline of half from the peak in two years. And the profit decline was far worse than the revenue decline, for a reason that is the single most instructive number in this entire story: depreciation and amortization rose from NT$4.34 billion in 2022 to NT$5.90 billion in 2023 and NT$6.46 billion in 2024, as all that boom-era equipment came into service.¹ Revenue halved. The fixed cost of the factories built to serve the boom went up by half. That is how a business with a 40% gross margin ends up with a 1% gross margin in twenty-four months.

By 2024 Nan Ya PCB was running an operating loss of NT$1.27 billion and free cash flow of negative NT$219 million, and it cut capital spending by 80%, to NT$2.38 billion.¹ The dividend told the same story with brutal clarity. On 2022's earnings the company paid NT$18.00 per share. On 2023's it paid NT$5.50. On 2024's earnings of NT$0.32 per share, it paid NT$1.00 — a payout that consumed more than three times the year's profit and had to come out of retained earnings.⁷

One under-appreciated feature of the collapse deserves a mention, because it bears directly on how much downside risk the balance sheet carries. The 2023 downturn was, in cash terms, far gentler than the income statement suggests. As revenue fell, receivables and inventory unwound, releasing roughly NT$6.9 billion of working capital and helping the company generate NT$16.5 billion of operating cash flow in 2023 even as profits fell by two-thirds. It used that cash to pay out NT$11.6 billion of dividends declared on the boom year and to fund the tail of the capacity program.¹ By 2024, with the release exhausted, operating cash flow fell to NT$2.16 billion and free cash flow turned slightly negative — and management responded by cutting capital spending rather than by raising capital.¹

That is the useful piece of the falsification exercise on balance-sheet strength: across the worst two-year deterioration in the company's modern history, Nan Ya PCB did not issue equity, did not undertake an emergency refinancing, and did not stop paying a dividend entirely. Whatever one concludes about the timing of its investment decisions, the company has demonstrated that it can absorb a severe cyclical trough without diluting shareholders. That is a genuinely differentiated attribute among capital-intensive component makers, and it is one of the few claims in this story that the historical record confirms rather than narrows.

What management said at the bottom

It is worth pausing on how management explained the collapse, because how a team narrates a bad year is a real piece of evidence about it.

At the August 2024 operating briefing, Nan Ya PCB was specific rather than evasive. First-half 2023 revenue had fallen 23.6% year on year, attributed to inventory correction in consumer electronics. First-half 2024 revenue fell another 33.0%, attributed to weak PC demand and continuing inventory correction in telecom. Operating profit fell by NT$5.56 billion in the first case and NT$5.61 billion in the second, each time explained by the specific end market responsible. And in one strikingly candid line, management noted that customers were delaying purchases of telecom products because of "budget cannibalization of AI" — that is, the AI boom that would later rescue the company was, at that moment, actively eating its networking business.²

That is a good disclosure. It names the end market, the mechanism and the magnitude, and it does not hide behind "macro headwinds." On the narrow test of whether this management team explains misses honestly, the 2024 record is a pass.

What the record does and does not establish

Here is the calibrated conclusion, because this is the piece of history that has to bear the most weight.

The claim under test is that Nan Ya PCB's management can be trusted to allocate capital sensibly through a cycle. The strongest disconfirming evidence in the company's own record is that it deployed nearly NT$29 billion into capacity across 2022 and 2023, at and immediately after the peak of the largest demand event in its history, and that the capacity landed into a demand air pocket that halved revenue and eliminated profitability. Depreciation from that spending is still on the income statement today.

Does that history reject the claim? Not entirely, and it is important to be precise about why. The same capacity built in 2022 and 2023 is the capacity now running above 80% utilization and driving the current re-rating.¹² Had the company not built it, it would have had nothing to sell into the AI cycle. There is a real sense in which the 2022 expansion was strategically correct and merely mistimed — and in an industry where a line takes twelve to twenty-four months to build, some mistiming is structural rather than a management failing.

What the history does establish is narrower and still important: this company has invested pro-cyclically in both of its last two cycles, and it has no demonstrated record of countercyclical capital discipline. It has never been observed slowing down at a peak. Every data point we have shows it accelerating into strength and then cutting hard into weakness — which is the standard behavior of capital-intensive cyclicals, and precisely the behavior that converts industry-level shortages into industry-level gluts.

The second thing the period establishes is a market-structure fact that no amount of AI demand repeals: substrate capacity is lumpy, slow and irreversible. You cannot build half a line. You cannot un-build one. When five companies all read the same shortage and all commit at the same time, the additions arrive together — and they arrive eighteen to twenty-four months after the moment that convinced everyone to build them, which is typically eighteen to twenty-four months after peak scarcity.

Hold that thought. It becomes extremely relevant when we get to what the industry committed to in August 2026.

↑ Back to Top

VI. The AI Re-Rating: 2025–2026

The recovery did not announce itself. It showed up quarter by quarter in the utilization data.

By the March 2026 investor call, Nan Ya PCB was reporting something it had not been able to report in three years: full-year 2025 consolidated revenue of NT$40.17 billion, up 24.4%, with operating profit swinging back into the black at NT$1.98 billion and fourth-quarter operating margin reaching 8.6%. Full-year earnings came to NT$3.01 per share, and the company proposed a cash dividend of NT$2.00.⁸¹ IC substrate utilization, management said, had exceeded 90% across the board.⁸

Then 2026 accelerated hard.

First-quarter consolidated revenue was NT$11.18 billion, up 32.1% year on year despite the Lunar New Year, with operating margin jumping to 12.1% from 3.8% a year earlier and gross margin expanding 10.8 percentage points to 15.8%. Earnings were NT$2.03 per share.⁹¹⁰¹¹ The second quarter was better still: net profit of NT$2.26 billion, up 72.8% sequentially and the strongest quarter since the second quarter of 2023, on a gross margin of 24.75% and an operating margin of 21.36% — both the highest in thirteen quarters. First-half earnings came to NT$5.53 per share.¹² July revenue reached NT$5.44 billion, up 50.2% year on year.¹³

Two things drove that margin expansion, and separating them matters.

The first is volume and mix. The second — and this is the part that distinguishes a genuine shortage from a merely busy factory — is price. Management confirmed on the calls that it had been "actively negotiating with customers and appropriately raising prices," initially framed as passing through rising copper-clad laminate and precious-metal costs, and later as straightforward market-driven repricing.⁸⁹ Brokerage estimates in May 2026 put second-quarter increases on high-end products at 30% to 40% and above.¹¹ A component supplier raising prices by that magnitude, to customers as sophisticated as Nvidia and Broadcom, is not a supplier with a good quarter. It is a supplier its customers cannot replace on the timeline they need.

What is actually pulling the revenue

Here the popular narrative and the disclosed mix diverge in a useful way.

The headline story is Nvidia. The actual story is networking. On the 2026 calls, management put network communication applications at roughly 49% of revenue — the single largest and fastest-growing application — with AI and HPC at 16% to 18%, PCs and automotive roughly flat.¹⁰⁸ IC substrates account for approximately 85% of revenue, split ABF at 50% to 60% and BT at around 30%, with conventional boards at 10% to 15%.⁸⁹ By mid-2026, management characterized AI-related business broadly as more than half of revenue.¹²

The most specific competitive fact disclosed anywhere in this story concerns switches. Nan Ya PCB is estimated to hold market share exceeding 70% in the IC substrates used in 800G and 1.6T network switches.¹¹ That is a far more concentrated position than the company holds in any other product, and it explains why its recovery has looked different in shape from its peers': the AI data center does not only need accelerators, it needs the fabric connecting tens of thousands of them, and every generational step in switch bandwidth requires a larger, higher-layer-count substrate. Nan Ya PCB's edge is in a specific corner of the AI build-out, not in the accelerator socket everybody writes about.

Myth vs. reality: what has actually changed

Myth: The 2024 collapse was an idiosyncratic stumble that a better-run company would have avoided. Reality: Every major ABF supplier expanded into the same peak and absorbed the same downturn. That is worse news, not better: it means the behavior is structural to the industry rather than fixable by management, and there is no reason to expect a different collective outcome from the same collective decision.

Myth: Nan Ya PCB is an Nvidia supplier and the stock is an Nvidia derivative. Reality: Its disclosed mix is dominated by networking and switch substrates, with AI accelerators and HPC a smaller, growing slice. The correlation to Nvidia is real but indirect, and the more precise exposure is to data-center interconnect bandwidth — a related but distinct demand curve.

Myth: The company is enjoying a margin recovery back toward its 2022 peak. Reality: A 24.75% gross margin is excellent for this business and terrible relative to 2022's 40%. The company is roughly two-thirds of the way back on margin while the stock trades far above its 2022 level. The re-rating is running ahead of the operating recovery, not behind it.

Myth: Pricing power is now structural. Reality: Pricing is up because utilization is above 90% and no qualified alternative capacity exists this year. That is scarcity rent, and it is the same thing that happened in 2021. The structural question is what happens when the 2027–2028 capacity lands — and on that, the evidence is not yet in.

A word on what the calls do not contain, since disclosure quality is itself an input to the analysis. Nan Ya PCB does not give quantitative forward guidance — no revenue range, no margin target, no unit assumptions. It describes direction ("sequential quarterly growth"), not magnitude.¹⁵ It does not name customers in its own materials; the Nvidia, AMD and Broadcom attributions that dominate coverage of this stock come from trade press and brokerage work, not from company disclosure. It has declined to quantify the size of its price increases, saying only that it negotiates with customers according to market conditions.¹⁵⁹ None of that is unusual for a Taiwanese component supplier bound by customer confidentiality, and none of it is evasive in the sense of contradicting prior statements. But it does mean that an investor's model of this business is built substantially on inference, and that the precision of the bull case exceeds the precision of anything the company has actually said.

The record capex bet

Which brings us to August 6, 2026.

On that date, Nan Ya PCB's board approved a new capital expenditure program of NT$46.8 billion to build a smart factory for large-format, high-layer-count advanced substrates, funded from retained earnings, bank borrowing and other financing.¹² To put that in scale: it is more than the company's entire 2025 revenue, and roughly two and a half times the NT$16.9 billion it spent in its previous record year.¹ General manager 呂連瑞 Lu Lien-jui framed it explicitly as a bet on the next ten to twenty years of AI, high-performance computing and advanced packaging demand.¹⁴¹⁵

The same board meeting approved something less discussed and arguably more revealing: a lease of land and factory buildings in Shulin from parent Nan Ya Plastics. Lu described this as a deliberate strategy — 租廠房、重設備, "lease the plant, invest in the equipment" — designed to concentrate capital in advanced process equipment rather than real estate.¹⁴ It is a sensible allocation choice on its face, and it is also a related-party transaction between a subsidiary and the 61% shareholder that supplies its raw materials and now its buildings. The terms were not disclosed in the announcements reviewed here. An activist would ask for them.

Nan Ya PCB was not alone. Within the same window, Unimicron approved NT$19.7 billion of new capital expenditure plus NT$17.6 billion of long-lead equipment commitments, and Kinsus raised its annual capital budget by roughly NT$19.6 billion to a total near NT$53.7 billion.¹⁶

Read that paragraph again alongside the previous section. Five companies, having spent 2023 and 2024 absorbing the consequences of building together at a peak, are now building together at a peak. The industry-level supply gap that everyone forecasts widening through 2028 is being addressed, simultaneously, by everyone.

Management's own framing acknowledges the timing without acknowledging the pattern. New high-end ABF capacity is expected to come online at the end of 2026 or in the first half of 2027, with the company forecasting that the supply gap will not narrow but become "increasingly significant" in 2027 and 2028.⁹¹⁰ That forecast may prove right. But it is worth noting exactly what it is: a company that is spending NT$46.8 billion telling investors that supply will remain short precisely through the period when its own new supply, and everyone else's, arrives.

The evidence-based conclusion for an investor is not that the capex is wrong. It is that its correctness is unknowable today and will be adjudicated in 2028 — and that the shape of the risk is precisely the shape of the risk that materialized last time.

↑ Back to Top

VII. Industry Structure & Competitive Position

If you wanted to design an industry that produces spectacular booms and equally spectacular busts, you would design this one.

Start with the buyers. Advanced ABF substrates go into a startlingly small number of packages designed by a startlingly small number of firms: Nvidia, AMD, Broadcom, Intel, Marvell, and the hyperscalers commissioning custom AI silicon. These are among the most sophisticated procurement organizations on earth, and they are structurally larger and more profitable than any of their substrate suppliers. Today, with utilization above 90% and no alternative capacity, the bargaining power sits with the supplier — hence 30%-plus price increases. The moment a single qualified competitor has open capacity, that reverses, and it reverses quickly, because these buyers dual- and triple-source deliberately.

Now the sellers. The top five ABF substrate makers — 欣興電子 Unimicron, イビデン Ibiden, AT&S, Nan Ya PCB and 新光電気工業 Shinko Electric Industries — held roughly 74% of the global market as of 2024, with Unimicron alone at approximately 22%.¹⁷ Behind them sits a credible second tier: 景碩科技 Kinsus and 臻鼎科技 Zhen Ding, both pushing into high-layer-count niches. Taiwan is the center of gravity for production; Japan retains the lead in the underlying material chemistry.¹⁷

It is worth knowing who these firms actually are, because they do not compete on identical terms. Unimicron is the scale player and the broadest — the most complete high-end ABF capability for AI servers, and the Taiwanese name most often cited as furthest along on next-generation substrate technology. Ibiden and Shinko are the Japanese incumbents, historically the highest-precision suppliers and long anchored to Intel-era CPU packaging; their advantage is process maturity, their constraint is that expansion in Japan is slow and expensive. AT&S is the only major European player with meaningful production capacity, which gives it a geographic-diversification argument no Asian competitor can make to a Western customer worried about supply concentration.¹⁷ Kinsus, spun out of the ASUS orbit, has been the most aggressive of the second tier on capital — its 2026 budget after an increase of roughly NT$19.6 billion reached about NT$53.7 billion, larger in absolute terms than Nan Ya PCB's headline program.¹⁶ Zhen Ding is the wild card: sell-side work has projected its IC substrate revenue tripling by 2028 on the back of qualification for Google's custom AI silicon.³

Read that list from a chip designer's chair and the picture is clear. There are five to seven credible counterparties, none of which can serve the whole market, all of which are expanding, and at least two of which are actively trying to break into programs they do not currently hold. That is not a comfortable position for a supplier to be in over a full cycle, however comfortable it feels this year.

Five players holding three-quarters of a market sounds like a rational oligopoly. It is not behaving like one. A rational oligopoly restrains capacity to protect price. What this industry actually does — demonstrated in 2022, and again in 2026 — is add capacity in unison whenever price signals turn favorable. Oligopoly discipline requires either explicit coordination, which is illegal, or a shared belief that restraint will be reciprocated. Neither exists here. What exists instead is five management teams, each terrified of losing a qualification slot at a marquee customer, each individually rational, collectively producing gluts.

Then the suppliers. We have already met the binding one: a single qualified source for the ABF dielectric film itself. But it does not stop there. On the 2026 calls, management flagged tightness in T-glass — the specialized low-expansion glass fiber cloth used in high-frequency substrates — and in precious metals, and noted that Japanese suppliers, Nittobo 日東紡 among them, dominate the copper-clad laminate materials for T-glass and E-glass applications.⁹¹⁰ Nan Ya PCB's own group makes copper-clad laminate, copper foil and glass fiber cloth, which mitigates part of this and is a genuine differentiator versus non-integrated peers.⁶ It does not mitigate the ABF dependency at all.

Substitutes are the weakest of the five forces today, and we treat them properly in the next section. Barriers to entry are the strongest: a greenfield high-layer-count ABF line costs tens of billions of New Taiwan dollars, takes eighteen to twenty-four months to build, and then has to be qualified package by package with each customer — a process measured in quarters. Nobody is entering this industry opportunistically. That is why the current shortage is real and why it will take years to resolve. It is also why, once resolved, it over-resolves.

Where Nan Ya PCB's advantage actually is — and where it isn't

Run Hamilton Helmer's 7 Powers over this business and most of them fail immediately. There is no network effect — substrates do not get better as more people use them. There is no meaningful brand power; no data-center operator has ever chosen a server because of whose substrate was inside. There is no counter-positioning, because Nan Ya PCB's business model is essentially identical to Unimicron's and Ibiden's. There is no cornered resource — the one genuinely cornered resource in this value chain belongs to Ajinomoto. Scale economies are real but shared: all five majors are large enough to amortize R&D and equipment.

What is left is process power and its cousin, switching costs — and they are real. Once a chip designer has qualified a specific substrate supplier's specific line for a specific package, changing suppliers mid-program means re-running reliability qualification, re-validating warpage and signal integrity, and accepting schedule risk on a product with a two-year lifecycle and enormous revenue at stake. Nobody does that casually. Nan Ya PCB's 70%-plus position in 800G and 1.6T switch substrates is exactly this power in action: it won those programs, and the customers are locked in for their duration.¹¹

But be precise about what process power protects. It protects existing programs. It does not protect the next one. Every new package generation opens a fresh competition in which Unimicron, Ibiden and Shinko bid on capability, capacity availability and price. A supplier with 70% share in a product generation can have 40% in the next one if a rival qualifies a better process or simply has open capacity when the designer needs it.

And we have direct evidence of exactly this happening to this company. Recall that Nan Ya PCB was producing flip-chip substrates from 2001 — leading edge for its era — and yet by the mid-2010s its mix had drifted back toward lower-value products and its income statement showed losses in three of four years.²¹ Process power did not compound. Position in one generation did not translate into position in the next. That is the strongest available counter-evidence to any framing of Nan Ya PCB's customer relationships as a durable moat, and it comes from the same business, the same capability, and the same corporate parent.

The calibrated conclusion: the moat claim survives, but only in a narrowed form. Nan Ya PCB has a genuine, evidence-backed advantage in currently qualified high-layer-count networking substrates, protected by real switching costs for the life of those programs, and reinforced by group-level materials integration that peers lack. It does not have a general moat in advanced substrates, it is not the technology or share leader — Unimicron is larger — and its own history shows the advantage decaying across generations unless continuously re-won. The KPI that would confirm or falsify the durable version of the claim is whether that switch-substrate share holds through the next bandwidth transition, and whether the company converts its position into named design wins on custom hyperscaler AI ASIC programs rather than remaining a networking specialist.

Which leads directly to the question of who is making those decisions, and what we know about them.

↑ Back to Top

VIII. Current Management: From Family Chairman to Professional Manager

On May 27, 2025, at the annual shareholders' meeting held at the company's Luzhu facility in Taoyuan, 吳嘉昭 Wu Chia-Chao stepped down as chairman of Nan Ya PCB after nearly three decades in the role. The board elected 鄒明仁 Tsou Ming-Jen in his place. Wu said he would remain a director of Nan Ya Plastics and that his responsibilities there "will remain connected" to Nan Ya PCB's operations.¹⁸¹⁹

On its face, this was a routine succession. It was not. It was the visible end of a chain of events happening one level up, in a family conglomerate that had been quietly rewiring how it governs itself for a quarter of a century.

Tsou Ming-Jen is not an outside hire. He was already a Nan Ya PCB director, is concurrently general manager of Nan Ya Plastics, and sits on the boards of both Nan Ya Plastics and Nanya Technology.¹⁸ The transition is continuity of group control expressed through a different individual — which is exactly what you would expect, and worth stating plainly rather than dressing up as a governance revolution at the subsidiary level.

That interlock is worth pausing on rather than passing over, because it is the structural fact that governs everything else about how this company is run. The chairman of Nan Ya PCB is simultaneously the chief operating executive of the company that owns 61% of it, supplies its copper-clad laminate, copper foil and glass fiber cloth, and — as of August 2026 — leases it factory buildings. His predecessor in the chair now runs the entire group. There is nothing irregular about any of this by Taiwanese conglomerate norms, and the arrangement plainly delivers real operating benefits in a tight materials market. But an investor should be clear that Nan Ya PCB's board is not an independent check on its controlling shareholder in any meaningful sense; it is an extension of it. Minority shareholders here are along for a ride whose direction is set at group level.

The revolution, such as it is, happened in August.

The first non-Wang president

On August 15, 2025, Formosa Plastics Group announced that 王文淵 Wang Wen-Yuan, aged 78 and the son of co-founder Wang Yung-tsai, was stepping down as president of the group's Administrative Center — the top operating job across the four core companies — and that Wu Chia-Chao, aged 81, would succeed him. Wang moved to chair the Management Center, retaining a supervisory role.²⁰

Wu thereby became the first president in Formosa Plastics Group's history who is not a member of the Wang family.²⁰ For a Taiwanese conglomerate whose founders' names are effectively synonymous with the postwar industrial economy, that is a genuinely significant break, and the group presented it in exactly those terms — as the realization of 所有權及經營權分治, the separation of ownership from management, and a move away from 家天下, "rule by the family."

The structure this completes was designed in 2001, when the founders established a two-tier system: an Administrative Center composed of the operating heads of the four core companies, and a Management Center through which the founding family could continue to supervise without running day-to-day operations. From 2017 onward, the four Wang family members active in operations transitioned from the Administrative Center into the Management Center. Wu's elevation was the last step.

Why Wu? He had the longest tenure of the four subsidiary chairmen, having led Nan Ya Plastics since 2013 after serving as its general manager from 2006. He was, by all accounts, deeply trusted by Wang Wen-Yuan and acceptable to both the family and the other subsidiary heads. Since August 2024 he had chaired the group's corporate transformation unit, covering product innovation, decarbonization and digitalization.²⁰

It is worth noting, without snark, that this transition involved a 78-year-old handing the top job to an 81-year-old. "Professional manager governance" here means governance by career executives rather than by founding-family members. It does not mean generational renewal, and an investor evaluating succession risk should hold both of those facts simultaneously.

The credibility test that has not happened yet

What do we actually know about how the current leadership allocates capital?

Almost nothing — and that is the honest answer. Tsou Ming-Jen took the chair in May 2025, at the bottom of a cycle. Every quarter since has been an improving one. The NT$46.8 billion program approved in August 2026 is the first genuinely consequential capital decision made under him, and it will not be adjudicable for two to three years.¹²

What we can assess is behavior. Three observations, in ascending order of usefulness.

First, narrative consistency has been good. The story management told in March 2026, June 2026 and August 2026 is the same story: utilization above 90%, AI and networking driving mix, prices being renegotiated upward, capacity expansion focused on higher specifications, gap widening through 2027–2028.⁸⁹¹⁰¹² There have been no unexplained strategy shifts and no quiet reframings of prior guidance.

Second, the company declines to give precise forward guidance, and it declined to disclose capital expenditure figures at the May 2026 shareholders' meeting on the grounds that board approval had not occurred — which is a legitimate regulatory constraint, and which it then honored by disclosing the number promptly on approval.¹⁵¹² That is a defensible pattern, if a frustrating one for anyone trying to model the business.

Third — and this is the one to hold onto — management's stated rationale for the largest investment in company history is a demand forecast running ten to twenty years out.¹⁵ Nobody can forecast substrate demand ten years out. The 2022 expansion was justified by a demand outlook that proved wrong within eighteen months. The appropriate posture is not that the new plan is wrong, but that "we see demand for the next decade" is precisely the sentence that preceded the last mistake, and it should be weighted accordingly.

The controlling shareholder sold

Between March 19 and May 15, 2026 — as the stock ran from the mid-hundreds toward four figures — parent Nan Ya Plastics disposed of 19.385 million Nan Ya PCB shares at an average price of NT$689.78, realizing approximately NT$13.37 billion. That left it holding 393,974,977 shares, or 60.97% of the company, down from roughly 67%. On May 22 it filed to transfer a further 16.874 million shares over a window running from May 25 to June 24.²¹²²

Wu Chia-Chao, in his capacity as chairman of the parent, gave two reasons. The first was float: Nan Ya Plastics' holding was unusually high even after the sales, and increasing the shares in public circulation would allow more outside shareholders to participate. The second was funding: Nan Ya Plastics itself needs capital to expand the upstream copper-clad laminate, copper foil and glass fiber cloth capacity that supplies Nan Ya PCB's substrate production.⁶

Both reasons are plausible and neither is evasive. A 67% controlling stake genuinely does suppress liquidity, and the vertical-integration argument is consistent with everything else the group is doing.

But present the fact neutrally and let it sit: the entity with the deepest possible information about this business — its controlling shareholder, whose chairman had run it for three decades and now runs the entire group — was a net seller of roughly NT$13 billion of stock during the sharpest re-rating in the company's history, at an average price around NT$690, while the public-facing message described a decade-long structural opportunity. That is not evidence of wrongdoing. It is a data point about revealed preference at a given price, and any investor buying above NT$1,100 should know that the best-informed holder was reducing at roughly 60% of that level.

The dividend policy claim, tested

One more governance detail, small but instructive. Nan Ya PCB's investor relations materials describe a policy of returning cash generously, with distributions "around 90%" of earnings each fiscal year.⁷

The disclosed record does not match that description. Payout ratios were approximately 63% on 2020 earnings, 61% on 2021, 60% on 2022 and 61% on 2023 — consistent, respectable, and nowhere near 90%. On 2024's earnings of NT$0.32 per share, the NT$1.00 dividend represented a payout of 312%, which is not generosity but a token distribution funded from reserves.⁷¹

The gap between the stated policy language and the actual record is not large enough to be alarming, but it is exactly the kind of thing worth checking rather than accepting. The practical conclusion for an investor: this is a roughly 60%-payout company in good years and a near-zero-payout company in bad ones. The dividend tracks the cycle; it does not cushion it.

↑ Back to Top

IX. The Optionality Bet: Glass-Core Substrates

Every organic substrate has the same fundamental problem, and it is a problem of physics rather than engineering. Resin and copper expand when heated at a very different rate than silicon does. Package a large, hot die onto an organic laminate and the assembly wants to bend — what the industry calls warpage. As AI packages grow toward and beyond 100mm on a side, and layer counts climb past thirty, warpage stops being a yield nuisance and starts being a hard ceiling.

Glass fixes this by changing the core material. A glass-core substrate has a coefficient of thermal expansion of roughly 3 to 5 parts per million per degree Celsius — very close to silicon's — so it stays flat. It has extremely low dielectric loss, cutting signal attenuation at high frequencies. And because you can drill vertical holes through glass more precisely than through resin, it supports interconnect densities roughly an order of magnitude higher. Those holes are the through-glass via, or TGV: conductive channels finer than a human hair, drilled through a sheet of glass and filled with copper.²³

If it works at scale, it is the successor technology. That is why every major player in the packaging chain is working on it.

Now the reasons it has not happened yet, which are unglamorous and stubborn. Glass is brittle: drilling and thermal cycling create microscopic cracks. Filling deep, narrow vias with copper by electroplating is slow — historically measured in days per panel, which destroys throughput economics. Maintaining consistent via diameter across high aspect-ratio holes is difficult. And automated optical inspection, the workhorse of defect detection, struggles with a transparent, highly refractive medium.²³ None of these is unsolvable. All of them are the difference between a working sample and a profitable factory.

The industry consensus timeline has 2026 as the year of sample submission and verification, 2027 to 2028 as the first genuine volume ramp, and process maturity — yields in the mid-nineties — arriving toward the end of the decade.²³ Some analysts think commercialization slips past 2027 entirely, with most technical challenges unresolved until closer to 2030.²⁴

Where Nan Ya PCB actually stands

Here the honest answer is: less far along than the "glass substrate concept stock" label implies.

Nan Ya PCB is grouped by Taiwanese market commentary among glass-substrate beneficiaries on the basis of its advanced IC substrate and build-up line capability — the adjacent skills would transfer. But it has not clearly disclosed commercialization progress, and it is not the Taiwanese name most frequently cited as furthest along; that distinction generally goes to Unimicron.²⁴²³ No named design win, no disclosed yield data, and no capital expenditure line item specifically identified as glass has appeared in the materials reviewed for this piece.

More telling is what management itself has said. On the March 2026 call, discussing next-generation materials, the company's position was that low-Dk/Df organic materials would remain mainstream for the next three to five years, notwithstanding the glass-substrate conversation, with ABF specifications heading toward formats exceeding 140mm and thirty-plus layers after 2028.⁸ That is management bounding its own optionality story: the company is telling investors that the technology it is spending NT$46.8 billion on is the technology that matters through roughly 2030.

Given this company's specific history — leading-edge flip-chip capability in 2001 that did not durably convert into leading-edge economics a decade later — the appropriate framing for glass is a call option to be tracked, not a pillar of the investment case. It becomes interesting the moment three specific things appear: disclosed glass-specific capital expenditure, a named customer program, and any yield or throughput data. Until then it is a capability the company will probably need eventually and has not yet demonstrated it will win.

One related and more immediate supply-security item belongs alongside it: the company has been working to qualify second and third sources for glass fiber cloth and other raw materials, reducing dependence on the small group of Japanese suppliers that dominate T-glass and high-end laminate inputs.⁹¹⁰ That is prudent housekeeping in a tight materials market, not a new business line.

↑ Back to Top

X. The Bull Case

Strip away the price action and the bull case rests on four legs, three of which are supported by evidence that would be hard to fake.

The demand signal is unusually credible. Component suppliers do not normally get paid before they ship. In this cycle, they have been. Across the ABF sector, customers have signed long-term agreements running three to seven years and made advance payments to lock in allocation — an arrangement that gives the substrate makers multi-year visibility of a kind this industry has essentially never had.²⁵ By April 2026, ABF capacity at Unimicron, Kinsus and Nan Ya PCB was reported effectively sold out.²⁶ Prepayment is expensive for a buyer and only rational if the buyer believes the alternative is not getting the part. When Nvidia, AMD and Broadcom behave that way toward a laminate manufacturer, the constraint is real.

The demand is structurally different in composition, not just larger. In 2015, ABF demand was dominated by consumer electronics and PCs. Morgan Stanley projects that by 2030, AI GPU, AI ASIC and networking applications will account for close to 75% of total ABF demand.³ That matters because those applications consume dramatically more substrate area and layers per unit than a PC processor, and because their demand driver — data-center capital expenditure — is set by a handful of well-capitalized firms with published multi-year build plans, rather than by hundreds of millions of consumers replacing laptops. The volatility characteristics of that demand base are genuinely different from the ones that produced the 2023 bust.

The supply response is slow, and the gap forecasts are wide. Even accounting for every announced expansion, sell-side estimates put the global ABF supply-demand gap widening from roughly 15% currently toward 22% by 2030, with some projections putting the shortfall as high as 29% by 2028.³²⁵ If those forecasts are directionally right, pricing stays firm for years rather than quarters — and Morgan Stanley has forecast ABF price increases of 15% to 20% in 2026 with potential for more than 20% in 2027.³ A supplier that raises price for three consecutive years while running at full utilization compounds earnings very quickly from a low base.

Nan Ya PCB has a specific, defensible position in the fastest part of the market. Its share of 800G and 1.6T switch IC substrates is the single hardest competitive fact in this story, and switch bandwidth is arguably the most reliably growing requirement in the AI data center.¹¹ Combined with group-level integration into copper-clad laminate, copper foil and glass fiber cloth — a supply-security advantage none of its Taiwanese peers can fully replicate — and the operating leverage that comes from having already depreciated a large slice of its boom-era equipment, the earnings power on a full order book is substantial. The move from a 1.1% gross margin in 2024 to 24.75% in the second quarter of 2026 is what that leverage looks like in one direction.¹¹²

To which one may add a fifth, softer leg: governance is plausibly improving. Replacing a chairman of nearly thirty years, at the same time the parent group installs its first non-family president and increases the subsidiary's public float, is a more shareholder-oriented configuration than what preceded it.

Here is what the bull case is really claiming, stated plainly so it can be tested: that AI infrastructure demand has changed the amplitude and duration of the substrate cycle enough that a business which earned NT$0.32 per share as recently as 2024 should be capitalized as a multi-year compounder. It is not an absurd claim. It is a claim that requires the next two capacity cycles to behave unlike every prior one.

↑ Back to Top

XI. The Bear Case

The bear case does not require anything to go wrong operationally. That is what makes it serious.

Start with valuation, because it is a separate risk from the business. At roughly NT$1,195 the stock trades near 140 times trailing twelve-month earnings and close to 400 times reported full-year 2025 earnings.[^1]¹ Multiples like that do not merely require the golden-decade thesis to be right; they require it to be right on schedule. A year of flat pricing, a two-quarter digestion pause in hyperscaler orders, or a delayed ramp at the new Shulin facility would leave the operating business intact and the equity badly exposed. The bull case and the stock price are not the same proposition, and conflating them is the most common error in this name.

The company's own history is the strongest bear argument available. We have already traced it: pro-cyclical expansion into the 2022 peak, revenue halving by 2024, depreciation from the boom-era build rising into the trough, gross margin collapsing to approximately 1%, and a dividend cut to a token NT$1.00.¹⁷ There is no observed instance of this management team throttling capital spending at a peak. The NT$46.8 billion program is being executed by a chairman with fifteen months in the role and no track record on this specific test.

Competitive capacity is arriving, together, on a knowable schedule. Unimicron's NT$19.7 billion plus NT$17.6 billion of long-lead equipment, Kinsus's NT$19.6 billion increase to a roughly NT$53.7 billion 2026 budget, and Nan Ya PCB's own NT$46.8 billion are not independent decisions — they are five firms responding to the same price signal at the same moment.¹⁶¹² Nan Ya PCB's new high-end ABF capacity is guided to come online at the end of 2026 or in the first half of 2027.⁹ Everyone else's arrives in the same window. The forecast supply gap is what justifies the spending; the spending is what closes the gap. Sell-side forecasts of a widening 2028 shortfall are, in most cases, built on demand curves rather than on a bottom-up census of what every player has just committed to build.

Note also the framing analysts themselves are using: the current pricing environment has been explicitly compared to 2020–2022.¹¹ It is worth remembering how that comparison ended.

Buyer power is dormant, not absent. Porter's framework is useful here precisely because it separates current conditions from structural ones. Rivalry among five capable firms is intense and disciplined only by scarcity. Buyer concentration is extreme and the buyers are vastly larger than the sellers. The moment utilization drops from the low nineties to the low eighties, the same customers now prepaying for allocation will be running competitive bids, and the 30%-plus price increases of 2026 become the give-backs of 2028. The three-to-seven-year contracts help, but contract length locks volume, not necessarily price, and terms have not been disclosed.

Customer concentration and requalification risk are the same risk viewed twice. A 70%-plus share in one product family is a strength and a concentration. If a hyperscaler changes switch architecture, or if a rival qualifies a superior high-layer-count process for the next bandwidth generation, that share is contestable in a single design cycle — and this company's own drift out of high-value chipset substrates through the 2010s is proof that such drift happens here.²¹

The controlling shareholder was a seller into the rally. Approximately NT$13.4 billion realized at an average of NT$689.78, with a further tranche filed weeks later.²¹ Stated reasons are credible. It still cuts against a purely promotional reading of the story.

Geopolitics and concentration. Nan Ya PCB manufactures in Taiwan and in Kunshan, mainland China, with Kunshan slated for double-digit percentage capacity growth.¹⁵ For a company embedded in the American AI hardware supply chain, cross-strait tension and US–China technology policy are genuine tail risks — not acute today, but the kind of exposure that gets repriced instantly rather than gradually.

A second-order item worth noting rather than dramatizing. The August 2026 board approved leasing land and buildings in Shulin from the 61% parent, extending a "lease the plant, invest in the equipment" model the company has used for several years.¹⁴ The economics may well be favorable — the parent has land, the subsidiary needs it, and concentrating capital in equipment is defensible. But an activist would want the lease terms, the rent escalators, and an independent valuation, because this is a controlled subsidiary transacting with its controller in raw materials, real estate and, indirectly, financing simultaneously. Related-party density is not itself a problem; opacity around it is.

The activist's summary question. If you are running a business with genuinely improving fundamentals, a controlling shareholder selling stock, an unprecedented capital program justified by a twenty-year demand forecast, an untested chairman, and an equity trading at over a hundred times earnings — where exactly is the margin of safety? The bull answer is: in the order book and the qualification barriers. The bear answer is: there isn't one, and there wasn't one in 2022 either.

Neither answer can be settled by argument. Both resolve into a small number of observable quantities, and it is worth being disciplined about which ones.

↑ Back to Top

XII. What to Watch: KPIs and Forward Tests

Most of what gets written about this company is noise. Three metrics carry nearly all of the signal.

1. Blended gross margin, read together with the ABF share of revenue. This is the single most informative number Nan Ya PCB discloses. It captures utilization, pricing and mix simultaneously — the three variables that determine whether the company is climbing the value ladder or sliding back down it. The historical range tells you how to interpret any given print: approximately 1% at the 2024 trough, 40% at the 2022 peak, 24.75% in the second quarter of 2026.¹¹² Watch it alongside the ABF share of the revenue mix, which management has run between roughly 50% and 60%.⁸⁹ Margin rising with ABF share rising means genuine mix improvement. Margin rising with ABF share flat means the company is simply enjoying a price cycle — which is fine, and which will end.

2. Capital expenditure versus depreciation, and any sign the plan is being throttled. This is the direct test of whether this cycle's capital discipline differs from 2022's. Depreciation ran NT$6.89 billion in 2025 against cash capital spending of NT$2.42 billion — a company harvesting its prior build.¹ The NT$46.8 billion program reverses that decisively.¹² The specific thing to watch for is not the absolute number but management's behavior at the first sign of demand deceleration: if hyperscaler AI capital expenditure growth slows in 2027 and Nan Ya PCB's spending continues on the original schedule anyway, that is the 2022 pattern repeating in real time. If the company visibly phases or defers a tranche, that is new information about this management team and would materially strengthen the bull case.

3. Utilization across Shulin, Jinxing 錦興 and Kunshan. The fastest available read on whether the shortage narrative is intact. Above 90% means pricing power persists; a drift toward the low eighties is the leading indicator that the 2027–2028 capacity additions are landing ahead of demand.⁸¹²

Read together, these three tell a coherent story in either direction. Utilization holding above ninety while margin and ABF share both climb, with capital spending tracking the announced plan, is the golden-decade thesis playing out as advertised. Utilization slipping into the eighties while margin flattens and spending continues regardless is the 2022 sequence beginning again, and it would be visible in the quarterly numbers a year or more before it showed up in the annual result.

Two secondary tells are worth tracking without elevating them to KPIs. The first is named design wins on custom hyperscaler AI ASIC programs — the test of whether the company broadens beyond its networking concentration. The second is Nan Ya Plastics' shareholding: further disposals, or a halt, is the cleanest available signal of how the best-informed holder views the price.²¹²²

Deliberately excluded: monthly revenue, which Taiwanese companies report and which financial media over-interpret. A single month tells you about shipment timing, not about the business.

↑ Back to Top

XIII. Durable Lessons for Investors

Three things this story teaches that generalize well beyond one Taiwanese substrate maker.

Being indispensable is not the same as having pricing power. Nan Ya PCB is, right now, genuinely hard to replace. Its customers are prepaying to secure its output. And yet the very same indispensability, present in identical form in 2021, produced record profits followed within twenty-four months by an operating loss — because indispensability in a capital-intensive component business triggers the industry's own supply response. When a component is scarce, every qualified maker builds, and they build at the same time, because they are all reading the same price signal and all afraid of losing the same customers. Capital-intensive suppliers to booming end markets have a strong historical tendency to compete away their own windfalls. The question is never whether demand is real. It is whether the supply response is slower than the demand growth for long enough to matter.

Technical qualification is not revenue, and this company is the case study. Nan Ya PCB was producing flip-chip substrates in 2001, embedded substrates in 2013, SiP substrates in 2016 — a genuinely impressive capability ladder that coincided with losses in three of four years between 2016 and 2019.²¹ The gap between "we can make it" and "we are paid well to make it at scale" is where this company's narrative has repeatedly outrun its economics. Apply the same discipline to the glass-substrate story: a sample is not a design win, a design win is not a ramp, and a ramp is not a margin.

A governance improvement is worth tracking, but only the right test counts. Formosa Plastics Group's move to a non-family president and Nan Ya PCB's first chairman in three decades are real changes, and the increase in public float is a genuine minority-shareholder positive. But none of that speaks to the one question that determines returns in this business: does the company slow down when the cycle turns? Management transitions get evaluated on biography and governance structure because those are visible immediately. They should be evaluated on capital allocation at a peak, which takes years to observe. As of September 2026, that observation has not yet been made — and an investor should be clear-eyed that the current price already assumes the answer.

References

1. Nan Ya PCB Corporation — Investor Relations, Financial Reports (annual and quarterly consolidated statements) ↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩

2. Nan Ya PCB Corp. — 2024 H1 Operation Briefing (PDF), 2024-08-22 ↩↩↩↩↩↩↩↩↩↩

3. ABF Supply Gap Widens to 22%: Morgan Stanley Sees "Golden Decade" for Substrates — BigGo Finance ↩↩↩↩↩

4. Ajinomoto Build-up Film (ABF) — Innovation Story, The Ajinomoto Group ↩

5. Nan Ya PCB Corporation — Company Profile / Vision ↩↩

6. 南亞兩考量 賣南電、南亞科股 — 經濟日報 Economic Daily News ↩↩↩

7. Nan Ya PCB Corporation — Dividend History ↩↩↩↩

8. 2026/03/17 南電 (8046.TW) 法說會：AI 與高階應用帶動，產能利用率突破 90%，2025 由虧轉盈 — vocus, 2026-03-17 ↩↩↩↩↩↩↩↩↩↩

9. 2026/06/12 南電 (8046.TW) 法說會：26Q1營益率跳升至12.1%，AI帶動ABF載板供不應求 — vocus, 2026-06-12 ↩↩↩↩↩↩↩↩↩↩

10. 南電法說會｜需求強勁，首季營收年增逾3成，ABF載板供需缺口恐擴大至2028年 — 鉅亨網 Anue ↩↩↩↩↩↩

11. ABF 載板進入定價上升週期！外資喊買南電 — TechNews 科技新報, 2026-05-08 ↩↩↩↩↩↩

12. 南電新增資本支出468億元 第2季獲利創13季高點 — 中央社 CNA, 2026-08-06 ↩↩↩↩↩↩↩↩↩↩↩↩

13. 南電 (8046) 2026 年 7 月營收 54.40 億元，月增 16.13% 年增 50.24% — Win 投資 ↩

14. 468億大擴廠！南電祭史上最大投資案 延續「租廠房、重設備」神操作 — 工商時報, 2026-08-07 ↩↩↩

15. 南電股東會／今年投資規模將會是史上最大 營運看季季增 — 經濟日報 Economic Daily News ↩↩↩↩↩↩

16. 載板廠全面追加資本支出！南電新增468億元最多 欣興、景碩也分別追加197億元及196億元 — 知新聞 knews ↩↩↩

17. ABF (Ajinomoto Build-up Film) Substrate Market Outlook 2025–2032 — Intel Market Research ↩↩↩

18. 南電傳承交棒 鄒明仁新任董事長 — 中央社 CNA, 2025-05-27 ↩↩

19. 吳嘉昭交棒 鄒明仁接南電董事長 — 工商時報, 2025-05-28 ↩

20. 台塑接班內幕！首位非王姓總裁出線，王文淵為何交棒吳嘉昭？ — 遠見雜誌 Global Views Monthly ↩↩↩

21. 南亞擬再賣南電股票 申報轉讓1.68萬張 持股將降至3.96億股 — NOWnews 今日新聞, 2026-05-22 ↩↩↩

22. Nan Ya PCB Corporation — 股權結構 (Stock Structure), Investor Relations ↩↩

23. 3分鐘秒懂玻璃通孔技術 TGV，四大量產瓶頸及台股概念股全攻略 — 槓桿學院 ↩↩↩↩

24. 半導體封裝用玻璃基板商用化恐延至 2027 年後 — 財報狗 statementdog ↩↩

25. ABF Supply-Demand Gap Could Surge to 29% by 2028 — BigGo Finance ↩↩

26. AI chip rivalry escalates: ABF substrate sells out for Unimicron, Kinsus, Nan Ya PCB — DigiTimes, 2026-04-20 ↩