Headline
T-Glass Shortage Threatens AI Accelerator Supply Chains
Key Facts
- What: Nittobo controls ~90% of global T-glass supply, a critical low-CTE glass-fiber cloth used in advanced AI chip packaging substrates.
- Impact: Surging demand from hyperscalers has caused 20-30% price increases and lead times for related materials to stretch beyond 20 weeks, with suppliers no longer providing firm delivery dates.
- Response: Nittobo is tripling capacity at its Fukushima plant, but new supply will not reach the market until mid-2027.
- Why it matters: T-glass is essential for maintaining dimensional stability in large, high-power 2.5D and 3D packages used by Nvidia, Apple, Google, and Amazon AI accelerators.
- Context: Interposer sizes have grown dramatically—from 814 mm² in Nvidia Hopper to 1,700 mm² in Blackwell—driving higher T-glass consumption per chip.
Lead paragraph
A little-known Japanese material supplier’s near-monopoly on a specialized glass-fiber cloth is emerging as a significant bottleneck for the AI industry. Nittobo controls roughly 90% of global production of T-glass, a low-CTE material required in the organic cores of advanced chip substrates. With demand from AI accelerators exploding, the resulting shortages are driving up prices and extending lead times, threatening to slow the rollout of next-generation systems from Nvidia, Apple, Google, and Amazon, according to a detailed report from Tom's Hardware.
What is T-glass and why is it critical?
T-glass is a high-performance glass-fiber cloth with a low coefficient of thermal expansion (CTE). It forms the core of organic IC substrates—the critical interconnect layer that sits between a high-performance processor die and the printed circuit board. As AI accelerators grow larger and run hotter, maintaining flatness and mechanical stability during manufacturing and operation becomes increasingly difficult. T-glass provides the precise dielectric and thermal properties needed to keep these massive packages dimensionally stable.
Unlike standard E-glass, which is used in lower-end chips such as microcontrollers and older mobile processors, T-glass is preferred for high-power 2.5D and 3D packaging architectures common in AI accelerators. The material is produced by melting silica-rich glass in specialized electric furnaces operating at 1,600–1,700°C, spinning it into yarn, and weaving it into ultrathin cloth. This complex, capital-intensive process cannot be scaled quickly, creating a structural vulnerability in the AI supply chain.
Nittobo’s dominant position and capacity expansion
Nittobo’s near-monopoly has become a focal point as AI demand surges. The company is tripling production capacity at its Fukushima plant in Japan, but the expanded output is not expected to enter the market until mid-2027. In the interim, the supply crunch has already produced measurable effects: prices for T-glass have risen 20–30%, and lead times for downstream materials such as copper-clad laminates (CCLs) have more than doubled from the historical 8–10 weeks to over 20 weeks.
“With T-glass supply even more constrained now, suppliers are no longer providing lead times,” said Bill Ho, an analyst at Yuanta, in comments reported by Tom's Hardware.
Analysts note that the IC substrate industry has historically operated on thin margins. Even modest spikes in demand for build-up materials, Ajinomoto Build-up Film (ABF), or T-glass can therefore trigger shortages. Bilal Hachemi, an analyst at Yole Group who tracks the IC substrate supply chain, told Tom's Hardware Premium that “it is not easy to replace the T-glass” because of its specific dielectric and CTE values that work better for AI chips, especially in the organic core.
Hyperscalers driving unprecedented demand
The current crunch is particularly acute because of who is consuming the material. Major hyperscalers—Nvidia, Apple, Google, and Amazon—are designing and ordering ever-larger chip packages. Each new generation of AI accelerator requires bigger interposers and more complex substrates, directly increasing T-glass usage.
Nvidia’s own data illustrates the trend: interposer sizes have grown from 814 mm² for the Hopper architecture to 1,700 mm² for Blackwell. Future generations, including the upcoming Rubin architecture, are expected to push these dimensions even further. This scaling directly translates into greater consumption of T-glass per unit, exacerbating the supply imbalance.
The situation mirrors earlier concerns over Nvidia’s dominance in GPUs and TSMC’s position as the leading-edge foundry, but it highlights a deeper, less-visible layer of the supply chain. While those companies have drawn significant public and geopolitical attention, the reliance on Nittobo for T-glass has largely flown under the radar until now.
Broader supply chain implications
The T-glass shortage is part of a larger pattern of material constraints affecting advanced packaging. Industry observers have described the competition for PCB materials—including copper foils, glass fabrics, and CCLs—as the “battle before the battlefield” that will determine success in the downstream AI hardware race. Similar pressures have been reported for high-bandwidth memory (HBM) and ABF film, creating a compounding effect on AI server production timelines.
Because new T-glass production lines require years of investment and specialized expertise, alternative suppliers cannot quickly enter the market. This leaves the industry dependent on Nittobo’s expansion schedule and any incremental capacity improvements other players might achieve. Reports indicate Nittobo has signed a strategic partnership with Nan Ya Plastics (part of Taiwan’s Formosa Plastics Group) to help address long-term demand, according to earlier coverage by Nikkei Asia and TrendForce.
Impact on developers, users, and the industry
For cloud providers and enterprises racing to deploy AI infrastructure, the shortage introduces new uncertainty into procurement and deployment planning. Extended lead times and higher material costs could delay the build-out of large-scale AI clusters and increase overall system prices.
Chip designers and packaging houses must now factor T-glass availability into their roadmaps, potentially forcing some to consider trade-offs in package size or performance. While E-glass remains an option for less demanding applications, it lacks the thermal and mechanical properties required for the largest, hottest AI accelerators.
The situation also underscores the fragility of the semiconductor supply chain’s upstream layers. Just as geopolitical tensions have spotlighted reliance on TSMC and specific countries for wafer fabrication, the T-glass issue highlights how a single specialized material producer can become a chokepoint for the entire AI ecosystem.
What’s next
Relief is not expected until at least mid-2027 when Nittobo’s expanded Fukushima capacity comes online. Until then, the industry will likely continue facing elevated prices and allocation challenges. Longer-term solutions may involve development of alternative low-CTE materials, increased investment in additional production capacity by other suppliers, or changes in packaging architectures that reduce T-glass consumption per chip.
However, given the technical requirements for next-generation AI accelerators and the time required to qualify new materials, T-glass is expected to remain the preferred solution for high-end applications for the foreseeable future. Companies such as Nvidia, Apple, Google, and Amazon are reportedly courting Nittobo and exploring ways to secure priority supply, according to industry reports.
The episode serves as a reminder that the AI boom’s success depends not only on cutting-edge silicon but also on the obscure, capital-intensive materials that make advanced packaging possible.
Word count: ~820
Sources
- Tom's Hardware - Shortages of crucial chip packaging material threatens AI accelerator supply chains
- Tom's Hardware - Glass cloth could be the next great AI shortage
- Nikkei Asia and TrendForce reporting referenced in Tom's Hardware coverage