Skip to content
World clockEU--:--UK--:--USA--:--CN--:--PLDEFRIT中文EN

portal about AI and technologyevents · analysis · interviews · technical background

Search
LIVE
›

AI accelerators hit a T-glass bottleneck as Nittobo races to add capacity

Nittobo supplies roughly 90% of the world's T-glass cloth, the material that keeps advanced AI chip packages flat. Its new capacity at Fukushima will not reach the market until mid-2027.

TechnologyAnalysisGrace OkonkwoPublished: 28 September 20264 min readSources 3
AI accelerators hit a T-glass bottleneck as Nittobo races to add capacity

Nittobo sits between the AI accelerator boom and the factories that build it. The Japanese company controls roughly 90% of the global supply of specialist glass-fiber cloth known as T-glass, according to Tom's Hardware. Demand has outrun what the company can ship.

T-glass is a low-CTE glass cloth used in the organic core of IC substrates. That is the interconnect layer between a chip and its printed circuit board. As AI processors get bigger and run hotter, keeping those packages dimensionally stable stops being a packaging detail. It becomes a supply constraint.

Prices up, lead times gone

The market is tight. Prices have risen between 20 and 30%, per Tom's Hardware. Lead times for downstream materials such as copper-clad laminates have stretched from a normal 8 to 10 weeks to beyond 20.

"With T-glass supply even more constrained now, suppliers are no longer providing lead times," said Bill Ho, an analyst at Yuanta, in comments carried by Tom's Hardware.

Replacing the material is not simple. Bilal Hachemi, an analyst at Yole Group who tracks the IC substrate supply chain, told Tom's Hardware Premium that T-glass "has specific dielectric and CTE values that work better for the AI chips, especially for the organic core." He also pointed to the industry's thin margins. "Any increase in demand for build-up materials, ABF material, or T-glass can cause potential shortage, because it's against the basics of this industry," he said.

Nittobo is tripling capacity at its Fukushima plant in Japan. That supply will not arrive on the market until mid-2027. The material needs specialised electric melting furnaces running between 1,600 and 1,700°C. The process melts silica-rich glass, spins it into yarn and weaves it into ultrathin cloth. Scaling it takes years of investment.

Bigger packages, more material

This crunch differs from a routine cyclical shortage because of who is driving it. Hyperscalers keep ordering larger chip packages, and each generation consumes more T-glass per unit. Nvidia data cited by Tom's Hardware puts interposer sizes at 814mm² for Hopper and 1,700mm² for Blackwell, a 109% increase. The forthcoming Rubin and Feynman generations scale further.

Bank of America estimates Nittobo's electronic materials segment will nearly double sales from ¥40.9 billion ($266 million) in 2025 to ¥87.7 billion by March 2028. It expects operating margins approaching 48%. "Demand for T-glass cloth seems likely to grow more than originally expected," said Takashi Enomoto, a research analyst at Bank of America, per Tom's Hardware. He added that attention had been on thick T-glass for GPU and CPU packages. Ultra-thin T-glass demand is now likely to rise as leading-edge devices shift away from E-glass.

E-glass is the cheaper alternative. It is used in lower-end chips such as microcontrollers and older mobile processors. For massive 2.5D and 3D packaging, and therefore advanced AI accelerators, T-glass remains the preferred option.

The scramble for allocation has produced an unusual sight. Hachemi said Nvidia approaching an upstream material supplier directly is unprecedented. "For the first time, we are seeing Nvidia, the end customer, reaching out to the upstream material suppliers to secure the capacity and make sure they will get it," he told Tom's Hardware Premium. The risk is that once Nvidia locks down its share, rival chip buyers compete for what is left.

Nittobo is not waiting on Fukushima alone. It is doubling raw yarn capacity at its Taiwan plant and importing yarn back to Japan for cloth manufacturing. It also has a collaboration deal with Nanya Plastics to outsource some weaving. Tom's Hardware reports that by 2027 roughly 20% of Nittobo's glass cloth is expected to be woven by Nanya, a competitor. That is a fair measure of how tight the market is.

The wider accelerator pipeline keeps moving. At Hot Chips 2026, Microsoft detailed its second-generation Maia 200, according to ServeTheHome's live write-up. It is a TSMC 3nm part with 140 billion transistors, 6 stacks of HBM3e, 7TB/second of memory bandwidth and a 750 Watt TDP. d-Matrix used the same event to present Raptor, a 3D-DRAM accelerator that stacks a TSMC N4 logic die on DRAM. The company argues that HBM4 packages such as Nvidia's Vera Rubin and AMD's Instinct MI455 face a practical bandwidth ceiling around 20 TB/s.

None of that matters much if the substrate underneath cannot be built. Analysts quoted by Tom's Hardware expect the squeeze to persist until new T-glass capacity lands. Pricing pressure on AI hardware has a component most buyers have never heard of.

Comments 0

Sources

3
  1. 01Shortages of crucial chip packaging material threatens AI accelerator supply chainsEN
  2. 02Microsoft's Maia 200 AI Accelerator at Hot Chips 2026EN
  3. 03d-Matrix Raptor 3D-DRAM Accelerator for Generative Inference at Hot Chips 2026EN

All figures and quotations in this text come from the sources listed below.

Content prepared by the editorial team with AI assistance.

Grace Okonkwo

Grace Okonkwo

AI, models and technology

Grace Okonkwo covers AI, models and technology for FLASH24, working from primary sources such as model cards, API documentation and benchmark papers rather than vendor summaries. She checks training data provenance, evaluation conditions and reported scores against the underlying datasets before any figure reaches print. She interviews researchers and engineers directly, tracks release calendars from major labs, and compares successive model versions on the same tests. Her own self-hosting, home-network and documentation-reading habits feed straight into that desk, since she tests tools on her own hardware first. She does not publish benchmark claims without a reproducible method.

Newsroom →

Comments

0
  1. No comments yet — be the first.

Write a comment

Comments are public. We do not publish abuse, spam or advertising.