Citi Research: AI Material Bottlenecks Spreading Upstream, Conservative Pricing by Japanese Firms Emerges as Core Tension
Alina Collins
Citi's on-the-ground research across six Japanese materials leaders finds AI hardware bottlenecks have spread from chips to upstream materials — glass cloth, InP substrates, copper foil — yet conservative pricing by Japanese firms is capping the profit upside their technology should command.
Where exactly have the bottlenecks spread?
AI servers, optical communications, and advanced packaging are pulling demand upstream simultaneously — glass cloth, CCL, low-Dk resin, InP substrates, copper foil, and tantalum powder all feel the strain.
This means → the chokepoint is no longer just chips and compute; it has climbed link by link into raw materials.
Citi surveyed business units and senior management at Mitsubishi Gas Chemical, Nittobo, JX Advanced Metals, AGC, Asahi Kasei, and Mitsui Mining & Smelting. All six point the same way.
Japanese firms hold the technology moat — why is profit elasticity still weak?
The core tension: supply is tight, products sit at critical nodes of the AI chain, yet Japanese firms still default to "modest cost pass-through" pricing.
In plain terms = your product is scarce, your customer can't leave — but you won't raise the price. You're capping your own margin.
Citi's verdict: whether these firms shift from "technology supplier" to "active price-setter" will decide who ultimately captures the value their technology creates.
Glass cloth: who controls the scarcest grade?
Citi's tightness ranking: T-glass (low thermal expansion) > NER glass (2nd-gen low-Dk) > NE glass (1st-gen low-Dk).
T-glass stands out most — the report says "essentially only Nittobo can supply it," putting the company under pressure to add capacity and renegotiate pricing.
NER glass had a surprise: resin-side advances lowered the Dk bar for some applications, so NER already suffices — no upgrade to 3rd-gen NEZ needed. This means → NER's addressable market may expand significantly.
Asahi Kasei is scaling up 2nd-gen low-Dk yarn; a low-thermal-expansion variant has begun sampling, with volume production expected between late March 2027 and March 2028.
InP substrates: why is JX Metals betting on a 10× expansion?
JX Advanced Metals announced it will expand InP substrate — indium phosphide, a core material for optical transceivers — capacity by up to 10× by March 2031, revising demand-growth expectations from ~25% to 30%–40%+.
The logic follows the optical-interconnect penetration path: in 2026, data-center optics mainly link rack-to-rack and site-to-site; after 2028, optics are expected to reach inside servers and racks, with electrical-to-optical migration sustaining demand.
Citi notes actual demand exceeding the 10× ceiling would not be surprising, but a cumulative ¥145 billion investment-recovery burden makes pricing strategy all the more critical.
JX also holds positions across the AI data-center materials map — sputtering targets, CVD/ALD precursors, lithography materials, tantalum powder, and titanium-copper.
Which materials ride the advanced-packaging upgrade?
Asahi Kasei's Pimel is the key material for RDL — the redistribution layer that routes signals between die and substrate in advanced packages. It holds a dominant share in leading customers' most advanced products, and supply already trails demand; capacity expansion is being pulled forward.
As CoWoS packaging evolves from CoWoS-S to CoWoS-R and CoWoS-L, layer counts rise and RDL material demand scales in step.
At Mitsui Mining & Smelting, optical-transceiver PCB demand is surging, creating a new growth driver for Micro Thin copper foil; HVLP5 holds roughly 80% share, serving the high-frequency, high-speed signal path where material upgrades are required.
How far has the low-loss resin race gone?
Mitsubishi Gas Chemical's new BT-resin grade RS is winning positive customer evaluations in FC-BGA and GPU applications; tight Low-CTE glass supply has opened a window for RS to be reassessed as an alternative.
AGC's latest CCL, the METEORWAVE ELL series, is being evaluated for 1.6T switches and routers; AGC is also developing a low-Dk fluororesin with a molecular structure close to PTFE, aiming to solve PTFE's inter-layer adhesion problem in multilayer builds.
Citi is cautious on PTFE itself: insufficient rigidity, poor surface adhesion, and weak CTE performance remain unresolved. PTFE looks more like a "composite-material upgrade" challenge than a drop-in replacement for existing systems on its own merits.
Content is for reference only, not financial advice.