Guo Mingchi: Glass substrates are the core of TSMC’s CoPoS—they are a “must-have,” not just “a nice addition.”

```

Renowned analyst Ming-Chi Kuo conducted an in-depth interpretation of TSMC’s recently leaked CoWoS glass substrate slide and reached a key conclusion: In the CoPoS packaging technology system, the glass core substrate (the "oS" part) is a necessary condition for successful AI chip manufacturing, rather than a dispensable optimization option. The market has significantly underestimated its strategic importance.

On June 11, 2026, TSMC delivered a speech titled "Advanced Packaging Technologies Essential for AI Evolution" at Japan's JPCA Show. One slide, "CoWoS Glass Substrate Development," soon spread online and drew widespread industry attention. In the speech, TSMC officially announced its cooperation with Ibiden and Innolux to develop glass core substrates, adopting a three-layer structure—glass core sandwiched between two layers of ABF stack, forming the "oS" in CoPoS.

Ming-Chi Kuo pointed out that the power integrity (PI) improvement data shown in the slide is the most commercially valuable information. Glass substrates are thinner, shortening the vertical conduction path of through glass vias (TGV), lowering conduction resistance and loop inductance, making power delivery more stable, enabling more transistors or higher clock frequencies, which directly translates to stronger AI computing power. This is the fundamental reason Nvidia and two other American clients show strong interest in this technology.

According to Kuo's industry chain investigation, if progress goes smoothly, TSMC aims to start mass production of glass substrates between Q4 2028 and Q1 2029 to match Nvidia’s AI chip iteration rhythm.

Three-Party Collaboration to Overcome Composite Material Structural Bottlenecks

The cooperation among TSMC, Ibiden, and Innolux has made key progress in mechanical structure verification. Industry chain investigation shows that the glass core substrate shown in the slide is cut from a full 250×250 mm substrate, and the ABF stack mainly uses Ajinomoto’s GL107, combining ABF-GCP, with test layers from 24 to 28—the mainstream ABF specification for AI chips from 2027 to 2028.

TSMC used CoW (Chip-on-Wafer) as the test carrier in experiments, pairing it with the "oS" glass core substrate for verification. This is sufficient to reproduce the most challenging mechanical structural problems in composite material processing. Kuo believes the positive test results mean the three parties have jointly overcome the key technical bottleneck.

For division of labor, Ibiden is currently responsible for cutting the 250×250 mm glass substrate. Kuo’s investigation indicates that the simulation stage for mass production of 510×515 mm substrates is expected in the second half of 2027. If Ibiden wishes to reduce production complexity to protect its ultra-high gross margin, it may transfer the cutting process to Innolux, which is more familiar with glass properties.

"Must-Have" vs. "Nice-to-Have": Distinct Roles of oS and CoP

Kuo clearly distinguishes the functional positioning of the two core components in the CoPoS system. CoP addresses production efficiency and cutting economy, influencing cost and price; oS deals with warping and durability, determining whether the chip can be manufactured and function normally.

He further notes that CoP is a "very-nice-to-have" optimization; lacking it means higher chip costs, but chips can still be manufactured. The oS is "must-have"; without it, successful chip manufacturing is questionable. This explains why TSMC chose to verify oS paired with existing CoW, not CoP—prioritizing verification of the most critical technical link.

Kuo specially clarified a common misunderstanding: the "COP" on the slide does not stand for Chip-on-Package, but refers to coplanarity, a technical index measuring structural flatness.

Power Integrity Improvement: The Key Reason Clients are Willing to Pay

Kuo called the PI improvement data shown in the slide "the real gold" and explained the client payment logic: production efficiency is TSMC’s basic responsibility, and clients won’t pay extra for it; but improvements in AI computing power directly affect their competitiveness and profitability, so they’re willing to pay. This is the fundamental reason Nvidia has a highly favorable view of glass substrates.

For TSMC, glass substrates increase packaging yield and reduce costs, while also boosting AI chip computing power and price. With both cost reduction and price increase effects, it positively impacts profitability and competitive position.

Additionally, in the post-speech Q&A, a listener asked about TGV details in glass substrates, and TSMC refused to answer. Kuo believes this refusal itself is a signal—TGV is the core key technology of glass substrates, and TSMC and Innolux currently hold the key intellectual property. TSMC is unwilling to disclose it. By contrast, when another listener asked about IVR, eDTC, and LSI integration solutions, TSMC responded in detail.

Cost Structure: High Unit Price Not Enough to Deter Adoption

The unit price of glass substrates is several times higher than current ABF substrates, and the glass processed by Innolux is the most critical and expensive single material. Nevertheless, Kuo believes high unit prices will not substantially dampen client adoption.

The reason is overall cost structure: currently, substrate cost only accounts for a low single-digit percentage of the AI chip BOM, while packaging yield loss can cost five to ten times as much as the substrate. So even if glass substrates cost several times more, their BOM proportion remains low, and they can significantly reduce packaging yield loss, which creates bigger savings. Overall, using glass substrates makes economic sense for clients.

Mass Production Target End of 2028, Ibiden Roadmap Divergence

According to Kuo’s industry chain investigation, if progress is smooth, TSMC will start mass production of glass substrates between Q4 2028 and Q1 2029 to meet Nvidia’s AI chip iteration timetable.

Widely circulated Ibiden earnings slides mark glass substrate mass production as 2030. Kuo interprets this as: Ibiden has always been conservative in public, and formally adding glass substrates to the roadmap further confirms the technology’s long-term development. However, he also notes that some details in Ibiden’s slides differ from known market information—for example, their photo mask timeline lags behind TSMC’s public statements by about a generation, and Rubin Ultra substrate sizes are much larger than the 90×90 spec marked for 2026-2027 in the slides. He reminds investors that they must cross-verify multiple sources when predicting future trends and not rely on a single source.

Risk DisclaimerThe market has risks, invest cautiously. This article does not constitute personal investment advice, nor does it take into account the special investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, viewpoints, or conclusions in this article suit their specific situation. Investing based on this is at your own risk. ```