# Glass Substrates: Every Advantage Has Its Flaw Written on the Back

> Notes from the 2026-08-30 episode of the Taiwanese podcast Statementdog on glass substrates — the distinction between carriers and substrates, the drilling and metallization bottlenecks, the three-way competitive landscape, and the production timeline. Educational content, not investment advice; no stock recommendations or price targets.

Published: 2026-08-31
Locale: en
Tags: glass substrate, advanced packaging, semiconductors, industry analysis, investment thinking


![A deep cleanroom corridor where a single transparent glass substrate is held under an inspection lamp, light passing through it, rows of equipment receding into the distance](/covers/caibaogou-2026-08-30-549-cover.png)

> When the water is too clear, there are no fish; when a person is too exacting, there are no companions.
>
> — Dongfang Shuo, *Reply to a Guest's Objection* (Western Han, c. 1st century BCE)

Water that is too clean cannot hold fish. The line was written about not being too severe with people. Two thousand years later a sheet of glass demonstrated it precisely: the surface is so smooth that you can print extremely fine copper lines on it — and so smooth that those lines will not stick.

The 2026-08-30 episode of *Statementdog* picks up from the previous episode on panel-level packaging and moves on to glass substrates. What I most wanted to write down afterwards was not the production timeline but the structure that recurs throughout: every time an advantage of glass is described, the next sentence almost inevitably introduces the trouble it brings — and that trouble *is* the advantage.

## What This Episode Is About

The show opens by dismantling a common confusion: many people bundle "panel-level packaging" and "glass" into one thing, as if using either requires the other. In reality glass is a material and panel-level packaging is a process format. They pair naturally, which is why they get discussed together, but neither presupposes the other — one manufacturer has announced panel-level packaging heading into production without using glass at all.

With that cleared up, the main thread unfolds: what role glass actually plays inside a package, why it has suddenly become a hot topic, the three technical barriers it is stuck behind, who is fighting over it, and the earliest realistic date for commercialization. The tone is notably restrained — timelines are repeatedly flagged as "my personal estimate," and on interposers the host simply says "nobody knows."

## Key Takeaways

**1. Ask first *where* the glass sits, or you will hear two contradictory answers at once.**

Glass has two entirely different identities in packaging. The first is as a temporary carrier — a workbench, a table on which material sits during processing. When the job is done, the table is the table and the chip is the chip; there is no glass in the finished product. This use is already widespread; for both existing and broader wafer-level packaging flows, the mainstream carrier material has long been glass. The second identity is the one currently generating excitement: glass that genuinely becomes part of the packaged chip and stays there permanently.

So when you encounter both "glass is already in mass use" and "glass is still two or three years from maturity," both may be true — they are simply talking about different things. This is the most basic and most frequently skipped step when parsing industry news: confirm which layer the term refers to.

**2. Why so many layers between chip and board? Because the gap is a thousandfold.**

Chip wiring is at nanometer scale — three nanometers, two nanometers — while the printed circuit board works in micrometers, sometimes hundreds of them. That is a gap of over a thousand times. The episode offers a good analogy: it is like a mountain trickle trying to merge into a major river. You cannot connect the trickle directly; nature works by degrees — trickle to stream, stream to river, river to sea.

The interposer and the substrate are those degrees. A signal leaves the chip's internal metal layers, enters the interposer's copper traces, then the substrate's, and only then reaches the board. Someone once proposed removing the substrate layer and connecting the interposer straight to the board — appealing, since one fewer layer means one less stretch of routing loss — but it is not yet technically achievable, so the ladder stays.

Beyond the smooth fine-to-coarse transition, the interposer has a second reason to exist: as advanced-node scaling hits its limits, the industry has shifted toward chiplets, splitting functions across several dies and reassembling them. That requires a board whose wiring resolution approaches advanced-node levels to hold them together.

**3. Glass got picked for a focused reason: it does not warp, and it can *agree* not to warp.**

As package area grows, both interposer and substrate grow with it — and the traditional material is a composite organic resin. Multiple materials, mismatched coefficients of thermal expansion, and warpage on every heating cycle. Glass addresses this two ways. First, it is rigid and simply resists bending. Second — and this gets mentioned less — glass's coefficient of thermal expansion can be tuned by formulation. If your material expands a lot, glass can be tuned to expand a lot; if little, little. It is not merely "I am stiff so I do not move," but "I can move the way you move."

**4. Then the flip side arrives immediately: hard means brittle — and the issue isn't whether it shatters.**

The hardest problem for glass substrates is drilling. Why drill? Because both faces of the glass are wanted. On top you stack ten or twenty layers of copper wiring like a tower; underneath you want the same, like a basement. And the tower and the basement need a direct elevator between them — routing around means longer traces and more loss. That elevator is the through-hole.

The subtlety is that the real bar is not "will it break." As the episode makes clear, total shattering is actually unlikely. What is fatal is a barely visible micro-crack beside the hole wall: harmless today, but it grows across repeated thermal cycles until the whole piece fails. So the standard is not "unbroken is good enough" — it is "not a single crack anywhere," which is a full order of magnitude harder.

**5. Too smooth is the same story: you can draw fine lines that refuse to stay.**

Copper traces are formed by photolithography, and light has no idea whether the surface underneath is rough — it assumes flatness. So on a rough surface, the straight line you designed lands on peaks and valleys and comes out distorted, with accuracy gone. Glass is far flatter than organic materials like ABF or BT resin, so in principle it supports finer lines and taller stacks — you can only build high on a level foundation.

The price is adhesion. Dust that lands on asphalt sticks; dust that lands on an ice rink slides away. Glass is the ice rink: the fine lines can be drawn but struggle to hold. That is the flat surface. The hole wall is worse — a narrow, deep hole whose interior is equally smooth, which metal must fill evenly. It may look well adhered at first, and later reveal thick patches, thin patches, and outright discontinuities. Surface metallization is hard; via metallization is harder still.

**6. The third barrier is the easiest to overlook: glass invalidates the entire inspection method.**

Existing packaging inspection leans heavily on reflected-light imaging — illuminate, reflect, image, camera-style, to see whether anything inside is wrong. The whole method rests on an assumption nobody states out loud: the material is opaque.

Glass is transparent. The assumption disappears, and the method fails with it. Manufacturers must move to scattering, X-ray, or infrared approaches — meaning new equipment, rebuilt procedures, and a fresh accumulation of stability and yield experience. This cost never appears on a material quotation, yet it is the real brake on adoption speed.

**7. Industry has inertia — and AI is what moved "unworkable" forward in time.**

The glass substrate concept was proposed over a decade ago; the shift from copper to optical interconnect has been discussed for just as long. The technology is not new. The industry simply behaves like people do: switching materials means buying equipment, spending money, relearning everything. If the old way still works, use the old way — usually until it genuinely stops working.

The episode's phrasing is apt: "usually you only switch when it really can't be done anymore." What is moving now is AI, pressing the old approach from two directions at once — packages must be larger (warpage surfaces) and signals must be faster (loss surfaces). Glass happens to help with both. So the variable worth tracking is not "is the technology mature" but "when does the cost of not switching exceed the cost of switching."

## Going Further

### 1. It's in the news every day — am I too early?

This is the honest thought after any thematic episode. And this one has already supplied the skeleton of an answer; it is just scattered across the timeline.

Sort the mentioned milestones by stage: the temporary-carrier use is **selling right now** and will grow naturally as chips get larger; glass substrates enter commercial products at **small volume in 2027**, with meaningful volume only from **late 2028 into 2029**; glass interposers have no timeline at all — the estimate offered is possibly after 2030, contingent on whether TSMC and Intel choose to adopt.

Three buckets, three completely different arrival times on a financial statement. The judgment worth practicing is this: after any thematic story, sort what you heard into "already shipping," "about to enter production," and "still in papers and demos" — then ask which bucket today's price is discounting. If a company's entire imagined upside sits in the third bucket, what you are buying is not a product but a three-to-five-year gap in time, and time gaps are easily defeated by delay.

There is also a genuinely useful cross-check buried in the episode: **upstream and downstream timelines can be checked against each other**. Material production must lead end-customer adoption by more than a year. So if the end customer says late 2028 to 2029, the material side should be announcing shipments around 2028 — and indeed Samsung Electro-Mechanics and America's Absolics target 2027, Dai Nippon Printing 2028, and Nippon Electric Glass 2029–2030. When those numbers line up, the timeline has credibility. When one company's schedule runs far ahead of everyone up and downstream with nobody moving in step, that is usually marketing language rather than a capacity plan.

An aside worth noticing: Korean players' schedules run over a year ahead of Japanese ones, and their share price moves have been correspondingly more energetic. That does not necessarily mean better technology — it may only mean greater willingness to announce. But it does remind us that "difference in progress" is something observable and verifiable, not merely something to believe.

### 2. Everyone claims to be positioned — so who actually wins?

The second classic pain point. Every earnings call describes an entry into this market, everyone sounds like a contender, and you end up unable to distinguish anyone.

The episode's decomposition is clean: break the thing into process steps, then see who stands at each. Glass substrates involve three critical steps — the glass itself (including the CTE-tuning formulation), drilling and metallization, and the layer-by-layer build-up wiring on top. Three types of player correspond: glass makers (Corning, SCHOTT, AGC, Nippon Electric Glass — essentially a four-way oligopoly), the display supply chain (already experienced with large glass sheets, and desperate for a growth market since panels have become brutal), and incumbent substrate makers (build-up is their home turf).

Then comes what I think is the episode's most counterintuitive and most portable judgment: **the hardest step today may be the one with the least defensibility later.**

The reason is crowding. Drilling and via-wall metallization is the acknowledged hardest stretch — and precisely because it looks hardest and most like an opportunity, all three player types want in. Glass makers offer to drill it for you; substrate makers plan to integrate upstream once they scale; the display supply chain says this is exactly what they are best at. When that many parties attack one barrier simultaneously, the barrier rarely holds long. The defensibility instead sits at the two ends: glass formulation is a decades-deep four-firm oligopoly, hard to attack from below; and stacking twenty or thirty build-up layers depends on long yield-tuning experience and on being one of the few survivors of an industry shakeout, hard to attack from above.

The general rule this yields is handy: when judging whether a step can earn money over the long run, "how hard is it now" matters far less than "how many people are attacking it" and "once solved, how copyable is it." Difficulty gets diluted by crowds. Accumulation does not.

There is a finer distinction the episode makes well: **being able to do something is not wanting to do it.** Corning's technical disclosures suggest it clearly could drill, but it has announced no schedule — it can simply say "if customers want me to, I will; if not, I will focus on selling material." TSMC is judged uninterested in the substrate layer: low margin, far from its core, and it has consistently said publicly that it hopes packaging partners will take that work. Yet the same company's stance on glass interposers could be the opposite — because the hard part of an interposer is not only the glass layer but the redistribution-layer stacking above it, which requires semiconductor-like equipment and experience. That is precisely where a wafer manufacturer is strongest and others struggle to enter.

So news that "a major player has the technical capability" is separated from "it will enter this market" by an entire business decision. Conflating the two is the most common source of overestimation in stories like this. Intel is worth remembering here: it has been the most vocal advocate of bringing glass into packaging over the past ten to fifteen years, and this wave of enthusiasm is partly its doing — yet its pattern is to develop the technology and then outsource manufacturing. ABF was the same: Intel developed it first, then worked with the supply chain over years to raise several suppliers. The technology leader and the volume beneficiary are frequently not the same party.

### 3. I understood the good news — so why couldn't I hold?

Because what you understood was *that* it will happen; what you could not hold through was *when*. This episode supplies material for closing that gap.

The glass substrate concept has existed for over a decade. Feasibility was never the missing piece; the missing piece was the moment of "no choice." So rather than tracking technology news, track the trigger conditions — and the episode gives two, which are independent of each other. One is warpage from growing package area (tracks the progress of panel-level packaging). The other is loss in high-speed signal transmission (tracks networking and optical interconnect speed upgrades). Glass helps with both, meaning either one accelerating will pull it along — structurally sturdier than a theme with a single driver.

Then there is the bellwether customer. The episode says it plainly: volume depends on TSMC; without TSMC, volume is hard to reach. The value of that statement is not stock selection but that it supplies a **falsifiable observation point** — no daily guessing, no parsing of ambiguous earnings-call tone. Watch one concrete thing: has that key customer announced adoption? If not, the story is still in its small-volume phase. If yes, the timeline has genuinely started.

A good investment hypothesis should look like this: not only "I think this will happen," but "here is the specific event I am waiting for, and here is what would make me admit I was wrong." The episode does not discuss methodology directly, but the way the host handles timelines — flagging estimates as personal, naming the bellwether customer, admitting nobody knows about interposers — is itself a demonstration of the habit.

## Where to Look Next

- *Statementdog* podcast, episode 549 (2026-08-30), "Glass Substrate Industry Outlook and Competitive Landscape," together with the preceding episode on panel-level packaging. The warpage thread only makes full sense across both.
- For technical background, start from three keywords: TGV (through-glass via), CTE matching, and RDL (redistribution layer). They map respectively onto the drilling, warpage, and stacking problems described above.
- The earnings calls and technical disclosures of the companies named in the episode are the most direct primary source for verifying timelines — especially whether material-side and end-customer production dates line up.

## The One Thing to Take Away

**One idea**: an object's strength and its weakness are often not two things but two faces of one property. Glass does not warp because it is rigid, and it is brittle because it is rigid. It supports ultra-fine lines and tall stacks because it is smooth, and nothing adheres to it because it is smooth. You cannot take only the first half. The skill is not finding something with advantages only, but knowing exactly what is written on the back of the advantage you chose — and arranging for that back side in advance.

**One exercise you can do today**: pick the trait you most admire in yourself — conscientiousness, decisiveness, softheartedness, high standards, a good memory, any of them. Then write down one specific piece of trouble that trait caused you in the past week. Not "what are my flaws," but a case where the trouble was **caused by precisely that strength**: conscientious, so the work never finished; decisive, so you missed what someone was telling you; softhearted, so you agreed to something you shouldn't have.

Once it is written down, you tend to notice that the effort you have been making — keep the strength, fix the flaw — often does not apply, because it is one sheet of glass. What you can actually do is prepare for the back side in advance: the conscientious person sets a time limit; the decisive person adds a "ask one person first" step. This has nothing to do with investing, but it trains the same muscle: the moment you see an advantage, turn it over and look at the back.