AI supply chain at risk: between T-glass shortage and emerging glass core
How a Japanese glass fiber company, Nittobo, represents a materials risk scenario for the AI chips supply chain.
As AI demand surges, new bottlenecks are emerging deep within the semiconductor materials supply chain. In this new article, Bilal Hachemi, PhD, Senior Technology & Market Analyst, Semiconductor Packaging at Yole Group, examines how T-Glass, an ultra-specialized glass fiber largely controlled by Nittobo, could become the next critical constraint for advanced packaging and AI hardware.
Building on lessons from the 2021 ABF shortage, Bilal highlights growing supply risks, pricing pressure, and structural vulnerabilities.
Through its extensive collection of reports on semiconductor materials, substrates, PCB technologies, and supply chain dynamics, Yole Group helps industry leaders anticipate disruptions and seize emerging business opportunities.
Put yourself a coffee and explore a new perspective on the semiconductor industry!
A supplier monopoly: is Nittobo’s position similar to Ajinomoto’s?
In the semiconductor industry’s collective memory, the 2021 ABF material shortage was a harsh lesson on how a single material can paralyze an entire supply chain.
Ajinomoto’s ABF dielectric layer, which most designers or VPs had never considered a strategic asset, suddenly became a symbol of structural fragility when substrate manufacturers ran out of qualified capacity. But simply adding more build-up layers could not solve the problem: the real constraint was factory throughput, yield compounding, and long qualification cycles.
Four years later, the industry finds itself staring at a remarkably analogous situation.
Bilal Hachemi
PhD, is a Senior Technology & Market Analyst, Semiconductor Packaging at Yole Group
Only this time, the material in question is even less well known and the supplier monopoly even more extreme: T-Glass, a low CTE* fiberglass cloth produced almost exclusively by Japan’s Nitto Boseki, better known as Nittobo.
T-Glass matters because of a simple physics problem. As AI chip packages and substrates are increasing in size to accommodate higher compute power in an approach that includes more HBM memory and chiplets, the IC substrates that support them must also increase, but carefully to avoid the warpage trap, by using lower CTE fiberglass.
The parallel to Ajinomoto’s monopoly on ABF film is instructive but imperfect. The differences make the T-Glass situation even more precarious.
ABF was a proprietary dielectric film from a single supplier, and substrate makers could, and eventually did, qualify alternative build-up materials from Sekisui Chemical, Sumitomo, or startups like Thintronics and others, even though this took almost a year.
T-Glass, by contrast, is not a proprietary product. It is the output of an entire specialty glass-melting and fiber-drawing infrastructure that Nittobo has refined over decades.
Building a new glass furnace requires approximately two years of construction and commissioning. Producing yarn with the required uniformity, dielectric properties, and CTE also demands deep process expertise. Weaving that yarn into cloth with the precise thread counts and surface finishes needed for high-end CCLs adds yet another layer of complexity.
Nittobo currently has around 85% market share, and the company’s recent announcement that it will invest ¥15 billion to build a new Fukushima plant capable of tripling T-Glass output underscores both the scale of the deficit and the reality of the timeline. It is extremely unlikely that this facility will reach full production before 2027 or 2028.
The strategic position of Nittobo has not been lost on hyperscalers. The latter targeted Nittobo directly to secure capacity and T-Glass allocation. This new level of upstream engagement would have been unthinkable five years ago when procurement of glass cloth was delegated several tiers down the supply chain.
Nvidia, in particular, was reported to have pre-booked fiberglass cloth capacity more than a year ahead of volume shipments. And other players like Apple, Google, and AWS are locked in intensifying competition for whatever remains.
Most of Nittobo’s 2025 and 2026 T-Glass production is already fully allocated to these hyperscale customers, while IC substrate and PCB manufacturers find themselves unable to secure additional supplies, with lead times for some CCL products exceeding 20 weeks.
And the supply chain has, in effect, been inverted: where once customers dictated price and terms through quarterly cost-reduction reviews, today it is the material suppliers who hold pricing power, and the rest of the chain must toe the line.
T-glass shortage reaches the entire PCB playground
Bilal Hachemi
PhD, is a Senior Technology & Market Analyst, Semiconductor Packaging at Yole Group
The T-Glass shortage is not an isolated event. Its reach has exceeded the IC substrates landscape to the entire PCB materials supply chain, with cascading consequences that extend far beyond AI packaging.
As glass fiber manufacturers in Taiwan and Japan redirect furnace capacity, weaving looms, and engineering resources from standard E-glass to higher-margin T-Glass and other low Dk cloths, the supply of E-glass, the primary material for BT substrates and PCBs for memory, automotive, consumer electronics, and telecommunications, is being actively cut.
TUC has publicly announced the suspension of production across key mid-Tg, high-Tg, and halogen-free FR-4 platforms, including its widely used TU-662, TU-768, and TU-747 series, citing upstream glass cloth suppliers’ decision to halt E-glass production in favor of low-Dk weaves.
And Fulltech Fiber Glass, Asahi-Schwebel Taiwan, and even Nittobo’s Taiwanese subsidiary Baotek are following suit.
The phase-outs, scheduled to accelerate through the second half of 2026, will remove mainstream CCL products from the Taiwanese market entirely, with affected orders expected to migrate to Chinese factories.
The implications of this shift are profound and multi-layered.
First on the pricing front. Indeed, the tightening has already produced cumulative price increases of almost 100% for lightweight glass fabrics and over 150% for heavy-weight fabrics in the current cycle. Resonac, a key CCL and prepreg supplier, announced a 30% price increase across its entire CCL and prepreg portfolio effective March 2026. A move seen far from just a routine cost adjustment but as a structural inflection point reflecting that pricing power has shifted upstream. Raw material cost escalation is not confined to glass: copper foil prices have hit record levels, with HVLP4-grade foils essential for AI server PCBs in particularly tight supply, tungsten carbide has surged 5x in the past year, and gold’s price increase has increased ENIG surface finish costs by 35–40%. The compounding effect of simultaneous shortages across glass, copper, tungsten, and precious metals is producing a cost environment that the IC substrates and PCB industries have not experienced in decades.
More strategically, the reallocation of Taiwanese and Japanese manufacturing capacity toward high-end materials is accelerating China’s consolidation of the mid- and low-end PCB materials market. TUC itself has confirmed that its China and Thailand factories are “full go on E-glass CCL” and could supply Taiwanese PCB manufacturers directly.
Bilal Hachemi, Yole Group explains: “This new reversal would deepen the global electronics industry’s dependence on Chinese supply for consumer products even as geopolitical tensions push governments toward supply chain diversification.”
The downstream effects are already visible in AI hardware delivery schedules. The supply chain from T-Glass through CCL through ABF films through to finished ASIC and GPU server modules operates as a tightly coupled serial process. A risk or constraint at any single phase propagates directly to final product shipments.
Unimicron has publicly stated that T-Glass delays have pushed out lead times for ABF-based IC substrates used in CoWoS packaging.
At Yole Group, analysts expect supply conditions to shift from relative balance to tightness in the second half of 2026, with price increases for high-speed PCIe Gen5 and Gen6 SSD controllers potentially starting as early as Q2 as inventories decline.
“The constraint is no longer at the silicon level. AI demand is now being limited by T-Glass capacity, specifically the output of glass furnaces in Fukushima and Chiayi,” comments Bilal Hachemi from Yole Group.
From ABF to T-Glass: the next supply shortage?
This material shortage is becoming a pattern in the semiconductor packaging supply chain. It reflects the existence of weak points and potential risks.
The question the industry must now confront is not whether the next bottleneck will emerge, but which material it will be and when.
The candidates are already visible:
HVLP4 and HVLP5 copper foils, where Mitsui has been pushed to triple its expansion plans within a single year, yet still cannot match demand projections,
Solder masks and photo imageable dielectrics required for finer-pitch interconnects,
Specialty resin solutions for M8-grade and beyond CCLs, and even the coated drill bits needed to process increasingly complex multilayer boards.
Each of these materials shares the same structural vulnerabilities that made ABF and T-Glass into potential risks of a full supply chain shutdown. Concentrated supply bases, multi-year lead times for capacity expansion, long and expensive qualification cycles, and an industry culture that historically treated upstream materials as interchangeable commodities rather than strategic assets.
The geopolitical dimension compounds the technical and commercial risks. Europe and the United States have, to date, made no meaningful moves to establish domestic or regional production of advanced glass fiber, specialty copper foils, or high-end CCLs, even though these materials are now, by any reasonable definition, strategic assets for national security and economic competitiveness.
And the Japanese government has recognized the stakes, approving Nittobo’s Fukushima expansion under the Economic Security Promotion Act and providing subsidies.
So, what are the next steps?
Stay ahead of the semiconductor materials for packaging curve with Yole Group
For years, Yole Group has been tracking the evolution of semiconductor manufacturing processes and especially the advanced packaging platforms. Yole Group’s coverage expands further with a strong portfolio of in-depth analyses, including the Status of the Advanced IC Substrate Industry 2025 report (2026 edition coming soon), as well as PCB Industry 2026 (coming soon). Yole Group’s analysts provide the market intelligence needed to understand technology choices, playground readiness, and scale-up dynamics.
w
Do not miss Yole Group’s analysts at BUSS 2026, May 19-20, 2026, in Silicon Valley, CA, USA. You can fix a meeting with them right now. Ask for a meeting!
And follow Yole Group on LinkedIn, to identify innovations & business opportunities that are shaping the future of the semiconductor industry!
w
About the author
Bilal Hachemi, PhD, is a Senior Technology & Market Analyst, Semiconductor Packaging at Yole Group.
He contributes daily to the analysis of packaging technologies and their related materials and manufacturing processes.
Previously, Bilal conducted experimental research in the field of nanoelectronics and nanotechnologies, focusing on emerging dielectric materials and their ferroelectric applications. He (co-) authored several papers in high-impact scientific journals and participated in several international conferences.
Bilal obtained a PhD in nanoelectronics in 2022 from the Grenoble Alpes University (France), and he studied at IAE Grenoble for a management master’s degree.