Goldman Sachs Cautious, Ming-Chi Kuo Strongly Backs TSMC's GCS Strategic Position
Alina Collins
Goldman Sachs upgraded Japanese fiberglass-cloth maker Nittobo to 'buy' while flagging caution on TSMC's glass core substrate (GCS) timeline; analyst Ming-Chi Kuo pushed back, arguing the market underestimates GCS's strategic weight — the real debate is how fast glass can replace today's packaging materials.
Why is Goldman bullish on Nittobo but cautious on GCS?
Goldman's July 6 report noted that ABF substrates — the current "baseboard" carrying chips in advanced packaging — are not expected to be replaced by GCS before 2030.
This means → existing fiberglass cloth stays in use for years, at least through Nvidia's Rubin Ultra generation. Nittobo, which holds over half the fiberglass-cloth market, keeps benefiting.
In plain terms = Goldman's logic is straightforward: GCS may be the future, but mass production is still far off — so buy the surer near-term bet.
Why does Kuo say Goldman's logic and a bullish GCS view don't conflict?
Kuo responded that Goldman's other reasons for upgrading Nittobo — larger substrates and thicker core layers driving higher fiberglass-cloth volume — do not materially conflict with a constructive multi-year GCS outlook.
In plain terms = making money on fiberglass cloth now and believing glass substrates will eventually replace them can both be true at the same time.
This reflects that the real GCS debate is not "whether to do it" but "how fast it can be done."
What is TSMC's GCS structure, and why does Kuo say "oS matters more than CoP"?
At the 2026 JPCA show in Japan, TSMC disclosed GCS R&D progress and announced partnerships with Ibiden and Innolux to advance development.
GCS uses a three-layer design: a glass core sandwiched between two ABF build-up layers — the "oS" (on-Substrate) portion of the CoPoS (Chip-on-Package-on-Substrate) architecture.
Kuo stressed that the market underestimates "oS." This means → within CoPoS, the substrate is more critical than the package itself — which is why GCS is being tested with existing CoW (Chip-on-Wafer), not with CoP (Chip-on-Package).
How expensive is GCS?
Kuo disclosed that GCS unit cost runs several multiples of current ABF substrates. Innolux-processed glass carries an extremely high unit price and is the most critical — and costliest — material in the entire scheme.
Beyond Nvidia, two other U.S. customers have expressed strong interest in GCS.
This means → despite the high cost, demand-side pull from top-tier customers is already real — the commercialization bottleneck sits on the manufacturing side, not the demand side.
Will upcoming earnings calls bring major GCS updates?
Kuo judges that TSMC is unlikely to make a major GCS/CoPoS disclosure on its upcoming July earnings call.
The reason: TSMC already revealed key findings at the Japan conference and, unusually, named its supplier partners publicly — what needed saying has been said.
Ibiden and Innolux are similarly unlikely to announce major GCS progress on their August earnings calls. In plain terms = investors waiting for a GCS bombshell on earnings day will probably be disappointed — the substantive commercialization milestone still awaits further validation.
Content is for reference only, not financial advice.