Written by: Trend Research

In the past month, the most compelling narrative in the global semiconductor market has been "TSMC CoPoS + Glass Substrate."

On June 4, TSMC revealed at the shareholders' meeting that the trial production line has been completed. On June 11, the JPCA Show in Japan leaked slides validating the glass core substrate in collaboration with Ibiden and Innolux, and on June 16, the details of the collaboration were officially disclosed. A series of signals acted like a catalyst in the market, driving Taiwanese tech stocks, A-share glass substrate concepts, and even Corning ($GLW) across the ocean into a frenzied state.

On July 3, the cold water hit.

The Anger of the Supply Chain

The Taiwanese tech self-media "Tech Taiwan" released an exclusive report based on interviews with several key figures in the TSMC CoPoS supply chain. The emotions of the interviewees were not about "clarifying" but rather "angry"; some said their blood pressure "soared" every time they saw false rumors in the market, while others expressed their "indignation" openly.

They wanted to correct two core messages: TSMC's first-generation CoPoS likely does not use a glass substrate in its technical roadmap; TSMC has never considered a glass interposer.

These two messages are fundamentally at odds with the overwhelming narrative in the market over the past month that "glass substrates are core to CoPoS."

SemiWiki simultaneously reposted this exclusive report with a headline that stated plainly, "Why TSMC's First-Generation CoPoS May Be Glass-Free."

The report also disclosed that Samsung Electro-Mechanics (SEMCO), Toppan in Japan, and other substrate manufacturers from Japan and South Korea have joined the R&D race for glass core substrates and have recently started submitting engineering samples to TSMC. In other words, the competition for glass core substrates in the supply chain is indeed heating up, but this is for "preparation," not "launch."

What Did the Market Confuse?

The reason this news is impactful lies in the fact that most market participants fail to distinguish between the appearances of the word "glass" in three completely different contexts, corresponding to three distinct technical paths, value amounts, and timelines for implementation.

Glass Carrier is a temporary support board that carries the chip and RDL through the process during manufacturing and is then removed, not remaining in the final product. This is like scaffolding on a construction site—it is taken down once the building is completed. CoPoS indeed uses a 310×310mm glass carrier, which all sources confirm, but the value of this glass is low and is not what the market is speculating on.

Glass Core Substrate is the underlying support substrate of the packaging structure, located beneath the interposer layer, responsible for mechanical support and connection to the PCB. Kuo Ming-Chi's industry survey on June 11 described its structure: glass as the core layer, sandwiched by ABF multilayer coatings, forming a sandwich structure. The technical challenges of TGV (through glass vias) drilling and copper filling refer to this layer. This is the high-value segment that the market is truly speculating on, and it is also the object of validation for TSMC in cooperation with Ibiden and Innolux.

Glass Interposer is located between the chip and the packaging substrate, traditionally handled by a silicon interposer in the CoWoS architecture, responsible for high-speed interconnections between GPU and HBM. Some people interpret CoPoS as "replacing the silicon interposer with glass," but Kuo Ming-Chi explicitly listed this common mistake on June 11: "CoPoS uses a glass interposer" is the most typical market misreading. In the CoPoS architecture, this interconnection role is jointly undertaken by the RDL layer on the chip side and the TGV/ABF multilayer on the glass core substrate side. TSMC has never placed glass in the position of an interposer.

Three types of glass, three locations, three completely different investment logics. The market stirred them into a confusing mixture and then slapped a unified label on this mix: "glass substrate concept stocks."

Cross-Verification: What Do Various Sources Say?

Arranging the key sources from the past month reveals a clear timeline.

TSMC CEO Wei Zhejia confirmed on June 4 at the shareholders' meeting that the CoPoS trial production line has been built, expecting production to reach a significant scale in 2-3 years. This statement points towards 2028-2029.

Kuo Ming-Chi provided a more specific judgment on June 11: CoPoS is expected to enter mass production in the second half of 2028, with NVIDIA's Feynman AI chip likely being the first adopter. He elaborated on the two uses of glass in CoPoS—temporary carrier and glass core substrate—and corrected three common misunderstandings.

On June 16, Kuo Ming-Chi further interpreted the slides leaked from TSMC at the Japan JPCA Show, making a critical judgment: In CoPoS, the importance of oS (substrate) is greater than CoP (panel). CoP addresses production efficiency and cutting economics, it's a "nice optimization option"; oS addresses warping and durability, which is a "must-have." He positioned the glass core substrate as TSMC's "must-have" and pointed out that aside from NVIDIA, two US clients have already expressed high interest.

TrendForce's report on June 17 highlighted a critical time gap overlooked by the market: the mass production node for CoPoS itself is trial production in 2027 and mass production in the second half of 2028; however, the commercial scale production of glass core substrates "may not occur until after 2030."

This indicates there is a 1-2 year window between the mass production of CoPoS and the introduction of glass core substrates. During this window, the first generation of CoPoS could very well continue to use traditional organic substrates as a transitional solution.

Tech Taiwan's supply chain leak on July 3 aligns perfectly with TrendForce's timeline.

$GLW: Is the Plunge Coincidental or a Harbinger?

Corning ($GLW) plunged 13.3% on July 1 and dropped another 10.8% on July 2, falling from a 52-week high of $271.78 to about $197. Over the past year, this stock has risen by 391%, with a TTM price-to-earnings ratio exceeding 105 times.

It should be noted that Taiwan Tech's report was released on July 3, later than GLW's two-day plunge. The recent drop of GLW was triggered more directly by passive buying exits resulting from the FTSE Russell Index reclassification, massive insider selling in June (more than $54 million cashed out within three months), and profit-taking due to its extremely stretched valuation.

However, the narrative layer of risk is accumulating. Corning has been labeled by the market as "AI glass substrates," and institutions predict 2026 to be the year of commercialization for glass substrates, with a global market size of about $18.6 billion, and a compound annual growth rate of 67% after 2028. If the first generation of CoPoS doesn't use glass core substrates at all, then those parts of the forecasts strongly tied to CoPoS will have their realization timelines significantly delayed. For a stock with a PE over 100 times, the difference in pricing between a one-year delay and a two-year delay can be 30% or more.

At the same time, it is necessary to see that Corning has strong business support beyond semiconductor packaging: its optical communications business benefits from the fiber demand of AI data centers and has recently secured a multi-year fiber supply agreement with Amazon; its "Glass Bridge" optical interconnect components are entering the CPO (Co-Packaged Optics) track. These business lines are unrelated to CoPoS's glass substrate. The valuation bubble of GLW is an issue of its own and cannot simply be attributed to the "collapse of the glass substrate narrative."

What Does This News Really Say?

Put aside the emotional expressions; the report from Taiwan Tech conveys signals that are much more moderate than they seem on the surface.

It does not say "glass substrates are a scam." On the contrary, manufacturers like Samsung Electro-Mechanics and Toppan are submitting engineering samples of glass core substrates to TSMC, which itself indicates that the industrialization process is accelerating.

What it states is: "It is later than you think."

The core value of the first-generation CoPoS lies in "transforming round into square"—increasing the material utilization rate of a 12-inch round wafer of less than 70% to over 90% for a square panel. This matter does not rely on glass. Glass core substrates solve the issues of warping and signal integrity, making CoPoS perform better in ultra-large size packaging, but they are not a prerequisite for whether it can operate.

For investors, what truly needs to be differentiated are three completely different logics: temporary glass carriers are already in use, have a low value, and do not constitute an investment theme; the equipment supply chain for CoPoS panel-level packaging is in the verification phase of 2026-2027, with the highest certainty and is the most immediate beneficiary segment; the introduction cycle of glass core substrates spans 2028-2030, with the highest value, but TGV yield remains a bottleneck issue and belongs to medium- to long-term layout.

The problem in the market is not that it misread the direction, but that it discounted revenues for 2030 into 2026 stock prices. In the semiconductor field, the gap between pricing two years early and realizing one year late is enough to kill a significant rise. Conversely, when the market panics and compresses the timeline too short, those who understand the real rhythm can find better entry positions.

Fear and greed are always two sides of the same coin. This coin is currently flipping in the air.