Without Compass and Square: Statementdog EP541 on Panel-Level Packaging, and a Technology Held Back by Its Own Translation

Though a man have the keen eye of Li Lou and the skill of Gongshu Ban,
without compass and square he cannot draw circles and squares;
though he have the ear of Music-master Kuang,
without the pitch-pipes he cannot set the five notes right.
—— Mencius, Book IV, “Li Lou I”

What this episode is about

Statementdog (財報狗) is a Taiwanese investing podcast. Its industry-deep-dive format gives one topic a full hour, and this one spends all of it on panel-level packaging. It starts with terminology — what the “level” in wafer-level and panel-level actually means — then works through why this path is accelerating in 2026, what physically goes wrong when you change a round carrier into a square one, why glass suddenly became the keyword, and finally splits the field into three groups of players: assembly-and-test houses, panel makers and wafer foundries, each with its own motives, capability gaps and timeline.

Original episode: Statementdog Podcast EP541, “Industry Deep Dive: Panel-Level Packaging — Outlook and Competitive Landscape” (published 2026-07-26, in Mandarin).

I expected a technical explainer and got something more useful than that. What the episode is really about is this: when a new technology’s essence is saving cost rather than being more advanced, it redistributes the profit pool of an entire supply chain — and the logic runs almost backwards from the leading-edge stories we’re used to. What follows are my own notes and extensions, not a recap of the show.

The notes I took

The word “level” is a translation accident; the real driver is geometry. The hosts open by clearing this up: the “level” in wafer-level and panel-level means category, not tier. Panel-level is not more advanced than wafer-level, nor less — they solve different problems. The real cause is the mismatch between round and square: dies are square, wafers are round, and cutting squares out of a circle always wastes the edges. When dies were small the waste was tolerable. But AI silicon doesn’t care about volume, only about compute, and package sizes have grown from roughly two reticles a generation ago to four and five and a half today, with TSMC’s roadmap reaching toward fourteen. The show’s arithmetic: at 5.5 reticles, a 300mm wafer yields about nine units, an area utilisation of roughly 58%. Same process, same cycle time, half the output — and the half you lost was eaten purely by shape.

Going from round to square isn’t hard because of the shape; it’s hard because of what the shape drags in. The dominant problem is warpage. A package isn’t one material but several stacked, each with a different coefficient of thermal expansion, and packaging processes cycle repeatedly through 250–300°C. The layer that expands faster gets held back by the one that doesn’t, and the whole thing bows. The show’s analogy is perfect: a book that got soaked and then dried out, all puffed and buckled. The trouble is that it doesn’t stop at “bowed” — every downstream step assumes a perfectly flat 2D surface. Once the surface curves, exposed circuit positions shift, layers stack on top of misplaced layers, vias fail to align, and in the end the signal simply doesn’t get through. Two side effects follow: coating relied on spin-induced centrifugal uniformity, and a square has no constant centre-to-edge distance; electroplating concentrates charge at sharp points, so the four corners plate thicker. You save area, and you pay for every bit of it in yield.

There are two kinds of glass here, and they keep getting conflated. This was the most valuable distinction in the episode. One kind is glass as a work carrier — 12-inch packaging has used round glass carriers for years, and all that’s changing is size. The other kind is glass that actually becomes part of the package, replacing the core of an ABF substrate or an interposer. The first isn’t news; the second is. Glass gets invited in because its thermal expansion coefficient is low enough that it stays flat even at large formats — a direct answer to warpage. But the host adds a sharper inference: TSMC’s move to glass substrates may not be primarily about area or warpage at all, since organic substrates can already be made large. The bigger motive is signal loss — the dielectric surrounding a copper trace affects loss too, and today’s organic materials lose meaningfully more as frequencies climb. There’s a lovely moment here: mid-explanation the host says that if you press an AI on the physics, “it’ll give you reasons you won’t understand either,” and his co-host replies, “well, that’s my one new fact for today.”

The chips in volume production first are the low-end ones — which is counterintuitive. If panel-level packaging was forced into existence by giant AI dies, you’d expect it to start at the top. It’s the reverse. What’s actually shipping today is power-management and RF silicon, parts whose line-width requirements are loose enough (mid-teens of microns is fine). The reason is simple: they’re not buying “advanced,” they’re buying cost. A small die that needs three passes on 12-inch gets done in one pass on an 800mm-square panel. At the high end, TSMC’s CoPoS pilot line matures around the middle of next year, and genuine volume inside customer products plausibly lands in the second half of 2028 or nearer 2029. The show flags something worth remembering: TSMC’s own stated window is “between 2025 and 2030” — which could mean the last day of 2029.

For the assembly-and-test houses this is money they must spend and won’t get paid extra for. This was the most investment-relevant stretch of the hour. The co-host’s question is blunt: you change packaging methods, your customers are the same customers, you don’t gain new ones — so why invest? The answer is that not investing kills you. Someone else does it cheaper, you lose competitiveness, and customers won’t pay a premium for a new method, so your only lever is cost. The technology therefore delivers no pricing power, only scale: the big house is bigger than you and cheaper than you — so who would invest in the small one? The named players are Powertech and ASE. Powertech announced its entry back around 2018–19, has coverage from low-end to high-end, and this year committed roughly NT$20 billion to expansion. ASE was vaguer for years and only committed clearly last year to a ~300mm-square line, with its larger 600mm-square line further out.

Panel makers are the genuine new entrants here, and their motive is that the day job is dying. Panel manufacturing and packaging are entirely different disciplines, but panel makers bring one asset nobody else has: experience handling large square substrates — thin-film deposition, coating, lamination, and moving big fragile things without breaking them. That’s exactly what the semiconductor world has to relearn when it goes from round to square. What they lack is fine line-width and multilayer build-up know-how, which is precisely what packaging houses and foundries have accumulated over decades. So the natural landing spot is the low end, helped by the ability to convert existing fabs and equipment at lower entry cost. The show is unsentimental about the motive: the panel business itself has no upside left — Taiwan’s “twin tigers” are on their way to being twin kittens — so they need an exit while they still have resources. Innolux is the most committed, with front-end volume production confirmed and a claimed 40-million-unit capacity — to which the host immediately adds that the number is partly marketing, since how many units you cut depends entirely on die size.

Don’t infer leadership from who discloses revenue; it’s close to an inverse indicator. Only Innolux quantifies its panel-level packaging revenue share on earnings calls, and it’s a low share. Powertech and ASE certainly have some, but it’s too small to break out. The host is honest about what this does and doesn’t tell you: he won’t conclude Innolux leads, and the undisclosed numbers may well be larger in absolute terms. On this topic, disclosure correlates better with whether a company needs a story than with how big the business is. As for why some customers chose a panel maker over an established packaging house, he refuses to invent a structural reason — it may simply be cheaper, or a willingness to take on projects the big houses found too small or too awkward. He notes how casually many Taiwanese founders describe how they got big: nobody else wanted to do it, so I did it, and that’s how I got the slot.

Where this takes me

1. When a technology is “do it or die” rather than “do it and win,” it doesn’t belong in the valuation.

That question the co-host asked the packaging houses is actually a clean test, and it generalises. Ask it of any new technology: what happens if you don’t do it?

If the answer is “we’d earn less,” it’s incremental, and you can talk about a higher multiple. If the answer is “our customers leave,” it’s maintenance spending — the fee for staying at the table. Panel-level packaging is plainly the second thing for the assembly-and-test houses: it doesn’t raise price, it lowers cost, and in an industry where technical leads get matched, lower cost almost always ends up passed to the customer. It changes who survives, not who earns more.

Follow that line and you reach an uncomfortable conclusion: the value this kind of technology creates mostly doesn’t stay with the people adopting it. It flows to two places — equipment vendors, because everybody has to buy, and the largest player, because a cost advantage only means something multiplied by volume. That’s exactly what the host’s line captures: the big house is bigger and cheaper, so who would fund the small one? So my practical use is this. When I see “company X is also moving into panel-level packaging,” I don’t treat it as a reason to mark anything up; I first check where that company sits on the scale curve — because in a maintenance technology, the mid-tier is in a worse spot than the tail: same bill, no scale dividend.

The one exception is the panel makers, for whom this is genuinely incremental revenue they didn’t have. But honesty requires the other half: sitting opposite that increment is a legacy business still deteriorating, and the two partially cancel. The show declines to call it and just says we’ll all watch — which I think is the right refusal. When a company has one business growing and another rotting simultaneously, what you need isn’t a more optimistic narrative, it’s the magnitude of each side. Until you have those magnitudes, neither story can carry a conclusion.

2. Bottlenecks don’t get solved, they relocate — and where they land determines the next set of beneficiaries.

I habitually look at industries through the bottleneck layer: when demand doubles, which layer breaks first. What makes this episode interesting is that the thing breaking first isn’t a component — it’s geometry. A round carrier can’t hold increasingly large square dies. It’s the first time I’ve seen a bottleneck whose form is a shape.

And the episode traces its migration completely: round becomes square → warpage becomes the new binding constraint → warpage is a thermal-expansion problem in materials → so low-expansion glass gets pulled in → but glass brings its own failures, breaking during handling, microcracks from through-hole drilling, and sometimes cracks that only appear in later steps rather than at the drill. Every layer you solve pushes the bottleneck one step upstream or one step deeper into materials, and each relocation mints a new group of beneficiaries and a new set of failure points.

That changes what to track. Rather than asking who wins panel-level packaging, ask where the bottleneck currently sits. On this episode’s evidence it’s parked at the processing and yield of large-format glass — which happens to be the hardest layer to verify, because it announces itself through press releases. The show mentions Intel claiming it has achieved glass with no cracking at all, and the host’s reaction is the part worth internalising: being able to make one is not the same as being able to make them at volume. Between a single flawless panel and stable repeatable output lies an entire yield curve.

So for me this isn’t a conclusion, it’s a question to check the answer on later: what to watch next isn’t “we’ve broken through” announcements, but output numbers, yield, and whether any of it converts into shipped customer product. The nice property of that question is that it expires on its own — the answer arrives on schedule, and I don’t have to keep re-adjudicating it.

3. Disclosure isn’t leadership; capital expenditure is talk that has already been paid for.

The episode hands you a ready-made ranking trap. Sort by “who quantifies panel-level packaging revenue on an earnings call” and you get Innolux first and the big packaging houses unranked — but the host explicitly says he doesn’t believe that’s the real order. The big houses stay quiet because the share is immaterial, not because it’s zero. On this topic, willingness to disclose tracks how badly a company needs a story, not how large the business is.

So what’s harder? Capital expenditure. Powertech went years without major expansion and then committed roughly NT$20 billion this year; ASE is building too. The host’s inference is simple: nobody expands that aggressively without something concrete behind it, so demand has surfaced clearly enough to justify writing cheques. Statements are free; capex has already been spent.

But this needs an immediate guardrail or the test gets abused. Capex proves conviction, not returns. Taiwan’s panel industry is itself a monument to conviction — everyone expanded then too, and it produced exactly the situation that now has those companies crossing into someone else’s industry to survive. In this same episode, panel makers are selling off idle fabs and Innolux is eyeing glass-material processing; those are the invoices from the last round of conviction. So I keep the two uses separate: use capex to confirm that industry demand has genuinely arrived, but not to confirm that any particular company will capture it. The first is an industry-level signal; the second needs gross margin and utilisation, and neither has happened yet.

Following that logic, the signal actually worth waiting for is clear enough: the day Powertech or ASE starts breaking this business out on its own, it has become material to the parent. Until then, only the small player uses it as a headline — and that is itself the evidence that it’s still small.

Worth reading alongside

• Original episode: Statementdog Podcast EP541, “Industry Deep Dive: Panel-Level Packaging — Outlook and Competitive Landscape” (2026-07-26, Mandarin)
• The hosts mention their own site carries an industry report and supply-chain map for panel-level packaging, if you want the named-company version
• To follow the technology: TSMC’s and Intel’s public packaging roadmaps and technology disclosures — their glass strategies visibly diverge
• To follow actual adoption: Innolux, Powertech and ASE investor-relations pages, especially the capital expenditure plans
• The lines from Mencius at the top are my own footnote to the episode, not part of the show

---

Disclaimer: This post consists of personal listening notes and study material. It is educational content and does not constitute investment advice, an offer, or a solicitation. Companies and industries mentioned come from publicly broadcast content and public information; no specific security is recommended and no price targets are given. Figures cited are estimates or claims made on the show and have not been independently verified — refer to companies’ official disclosures. Investing involves risk; make your own judgement based on your financial situation and risk tolerance, and consult a qualified professional where appropriate. Copyright in the quoted material belongs to the original podcast; please listen to the original and support the creators.

Raw Markdown

This article is an educational discussion of investment method. It is not advice to buy or sell any individual security, offers no target prices, and does not analyze any current holding. Investing carries risk; make your own decisions or consult a qualified professional.