Pre-market on July 6th, semiconductor research firm SemiAnalysis posted six consecutive tweets on platform X, directly stating that NVIDIA's Kyber NVL144 rack is facing a "major delay." Additionally, the Rubin Ultra version was reportedly scaled back from a 4-chip to a 2-chip design, and the adoption timeline for CPO (Co-Packaged Optics) also faces uncertainty. The news quickly spread, causing related stocks in China's Taiwan region, including CCL (Copper Clad Laminate), PCB (Printed Circuit Board), substrate, and test interface sectors, to fall by an average of 8-9% that day.

The Taiwan technology team at Bank of America Securities (including analyst Mike Yang et al.) promptly released a research report stating that while technical concerns are reasonable, the stock price decline actually presents a buying opportunity.

Concerns Are Reasonable, But the Logical Framework Differs: Demand Scaling is Not Demand Disappearance

The analysts argue that potential delays or cancellations in the CCL/PCB/substrate fields essentially represent "demand scale-down in response to supply constraint," not a reversal of the demand trend. They liken this to previous market concerns over DRAM demand for the Vera CPU—on that occasion, worries also triggered stock price volatility, but the underlying demand trend remained unchanged.

Why might this sell-off be an overreaction? The analysts provide a key rationale: CCL/PCB/substrate account for a relatively low portion of the total BOM (Bill of Materials) cost for an AI server, only in the low to mid single-digit percentage range. This low cost contribution means lower price elasticity—when customers cut or delay orders, they typically prioritize reducing high-value, high-proportion components (like memory) first, not these foundational materials.

This logic is supported by supply-side factors. The analysts indicate that high-end CCL (M8 grade and above) and ABF substrates are expected to remain in a supply shortage at least until the end of 2027, fundamentally due to lengthy equipment delivery cycles and structurally growing demand driven by server CPUs and accelerators. In other words, even if short-term demand pacing is disrupted by the Kyber delay, supply-side capacity expansion cannot catch up with demand in the short term.

Therefore, Bank of America characterizes this stock price drop as an "enhanced entry point" and maintains its Buy rating on the related sectors.

Q-Glass-Based M9 to Outperform PTFE, Sub-Sectors Show Varying Resilience

The manufacturing challenges for the PCB mid-board in the Kyber rack directly point to a technical route competition for high-end CCL materials. This backplane involves M9 grade CCL, where the industry currently has two main approaches: one based on quartz fabric (Q-glass) and the other on PTFE (Polytetrafluoroethylene). Both can improve the dielectric constant (Dk) and dissipation factor (Df), but their manufacturing bottlenecks lie in different stages.

The analysts note that the production difficulties for the two schemes are "located in different links"—the challenge for the PTFE approach lies in the production side of the CCL itself, while the challenge for the Q-glass approach lies on the PCB processing side (yield improvement). It is precisely this difference that leads Bank of America to conclude that Q-glass-based M9 grade CCL will achieve faster penetration in AI servers.

If the yield improvement progress for the Q-glass approach at the PCB stage is moderate, the upside potential for high-end CCL demand (i.e., the supply-demand gap) could actually be larger. This logic presents a potential benefit for CCL suppliers—the slower the yield ramp-up, the harder it is to meet effective demand, leading to higher scarcity premiums.

However, not all sub-sectors are equally exposed to the impact of the Kyber delay. The analysts provide a clear ranking of cyclical resilience: test interface/substrate > CCL > PCB. The reason is that solutions closer to the chip end are more directly driven by platform specification changes and are more deeply tied to the AI server upcycle.

For CCL, the report differentiates the advantages of TUC and EMC:

TUC: Has greater upside potential in market share expansion and pricing in the short to medium term.

EMC: Possesses long-term scarcity advantages stemming from its substrate material and M9-Q technology barriers.

Regarding substrates and test interfaces, Bank of America believes the negative impact of the "2-chip design" on the content per individual substrate can be offset by an overall increase in shipment volumes. Furthermore, the report specifically points out that market concerns about memory clients pressuring substrate prices are "unwarranted." The rationale is that the supply of BT substrates (Bismaleimide Triazine) continues to contract, glass fabric costs remain high, and memory demand remains robust. These three factors collectively support pricing resilience for substrates.