The ABF Substrate Supply Chain Behind AI Accelerators

September 16, 2026

Altsets

Research by Altsets Research

Share

AI accelerator growth is pushing the bottleneck beyond Ajinomoto's ABF film and into the high-layer-count substrate factories that convert the material into increasingly large, complex packages.

Data used:Altsets Supply Chain Intelligence: 90k+ entities, 400k+ relationships, 20+ years of history.

Key findings

  • IBIDEN approved an approximately ¥500 billion electronics investment program through fiscal 2028 centered on high-performance IC package substrates.
  • The first announced IBIDEN phase includes roughly ¥220 billion of investment around its Gama facility, with additional mass production expected from fiscal 2027.
  • Nan Ya PCB is developing next-generation ABF substrates for AI server ASICs and high-end networking equipment.
  • Ajinomoto describes ABF as a thermosetting film built from epoxy resin, a hardener, and inorganic microparticle filler.
  • The ABF bottleneck can move downstream into qualified substrate process capacity or upstream into specialized chemical inputs depending on which layer expands more slowly.

AI accelerators are creating a packaging constraint that sits between semiconductor materials and advanced packaging assembly. Ajinomoto supplies the build-up dielectric known as ABF, but the more immediate capacity question is increasingly downstream: whether companies such as IBIDEN and Nan Ya PCB can convert that material into the larger, more layered substrates required by GPUs and custom AI ASICs. At the same time, ABF itself depends on a less visible upstream chain of epoxy resins, curing chemistry and inorganic fillers.

The clearest signal is coming from IBIDEN. In February 2026, the Japanese substrate manufacturer approved an approximately ¥500 billion investment program for its electronics business through fiscal 2028, centered on expanding high-performance IC package substrate capacity for AI and other high-performance servers. The first announced phase includes roughly ¥220 billion of investment around its Gama facility, with additional mass production expected from fiscal 2027. [1]

That spending matters because the AI packaging constraint is not simply about producing more square meters of substrate. Each generation of accelerator is demanding more substrate manufacturing work per package.

The key distinction is between dielectric availability and qualified substrate process capacity. More ABF film does not automatically create more AI accelerator substrates.

¥500B
IBIDEN electronics investment program
Planned through fiscal 2028
¥220B
First announced Gama investment phase
High-performance IC package substrate expansion
FY2027
Additional mass production expected
IBIDEN capacity expansion timetable
3
Core ABF ingredient groups
Epoxy resin, hardener, and inorganic microparticle filler

ABF becomes more valuable as the package gets harder to build

ABF is the insulating layer inside high-density organic package substrates. Ajinomoto describes it as a thermosetting film built from organic epoxy resin, a hardener and inorganic microparticle filler. Substrate manufacturers laminate the film, create microscopic vias, deposit copper and pattern fine interconnects before repeating the process across multiple build-up layers. [3]

This puts Ajinomoto near the beginning of a long conversion chain.

Exhibit 1

ABF sits inside a longer conversion chain

The dielectric has to pass through specialized substrate manufacturing before it becomes part of an AI accelerator package

  1. 01
    Epoxy resin, hardener, filler
    Core ingredient groups identified by Ajinomoto for ABF.
  2. 02
    ABF film
    Ajinomoto converts the chemistry into build-up dielectric film.
  3. 03
    Package substrate
    Substrate manufacturers laminate, drill, plate, and pattern repeated build-up layers.
  4. 04
    Advanced package assembly
    The finished substrate joins logic, memory, and other package elements.
  5. 05
    AI accelerator
    The completed package becomes part of a GPU or custom AI ASIC system.

This is a simplified structural map. It does not imply that every supplier uses the same formulations, process sequence, or substrate architecture.

Source: Altsets research using Ajinomoto disclosures

The distinction matters for investors because additional ABF output does not automatically create additional accelerator substrate capacity.

Producing an advanced substrate requires repeated lamination, laser drilling, plating and fine-line patterning. As the package grows, more area has to pass through these processes. Additional build-up layers increase the number of manufacturing steps. Finer wiring raises process difficulty and yield sensitivity. Larger packages also make warpage and dimensional stability harder to control.

IBIDEN has explicitly linked AI server growth with larger and more multilayered substrates. Its 2026 investment program therefore represents considerably more than ordinary capacity replacement. It is an expansion of the manufacturing infrastructure required to turn dielectric film and copper into something an advanced GPU or ASIC can actually use. [1]

Exhibit 2

IBIDEN is expanding the conversion capacity, not just the material supply

The announced program targets high-performance IC package substrates for AI and other high-performance servers

MeasureReported valueWhy it matters
Electronics investment programApproximately ¥500 billionLarge multi-year expansion centered on high-performance package substrates
Program horizonThrough fiscal 2028Shows that qualified substrate capacity expands on a multi-year timetable
First announced Gama phaseRoughly ¥220 billionTargets additional manufacturing infrastructure around a key substrate site
Additional mass productionExpected from fiscal 2027Capacity arrives later than the current AI demand signal

The figures come from IBIDEN's February 2026 investment announcement. They describe planned capital spending and production timing, not guaranteed future output.

Source: IBIDEN

Nan Ya PCB is describing the same transition from another part of the market. In its May 2026 investor material, the company said it was developing next-generation ABF substrates for AI server ASICs as hyperscalers expanded data center capital expenditure. It is also targeting substrates for high-end networking equipment used to move data between increasingly large AI systems. [2]

The important signal is that both GPU growth and custom ASIC growth converge on the same substrate manufacturing processes.

The bottleneck extends in both directions from Ajinomoto

Ajinomoto is unusual because ABF sits deep inside the package while remaining technologically important to fine-pitch routing. But treating Ajinomoto as the entire ABF investment story misses two different dependency layers.

Downstream, substrate manufacturers including IBIDEN and Nan Ya PCB have to convert ABF into qualified, high-layer-count products. Other major FC-BGA and advanced substrate producers, including Unimicron, Kinsus and Shinko Electric, occupy similar positions in this manufacturing layer. Their exposure is not simply to semiconductor unit volumes. It is increasingly tied to package area, layer count, wiring density and the amount of processing required for each accelerator.

Exhibit 3

The downstream bottleneck is substrate process capacity

AI package complexity increases manufacturing work per package

Demand driverWhat increasesWhy substrate capacity tightens
Larger package areaMore substrate area per packageMore material and more area must pass through each process step
Higher build-up layer countMore repeated lamination and patterningEach additional layer adds manufacturing steps
Finer wiringProcess precision requirementsYield sensitivity rises as interconnect dimensions tighten
Greater package complexityWarpage and dimensional control burdenLarger structures become harder to keep within manufacturing tolerances
GPU and custom ASIC growthDemand across multiple accelerator architecturesDifferent AI chips can converge on the same substrate manufacturing layer

The table describes the process drivers identified in the article. It does not estimate company-specific throughput or yield.

Source: Altsets research using IBIDEN and Nan Ya PCB disclosures

That makes substrate demand capable of growing faster than accelerator shipments themselves. A package generation that becomes larger and requires more build-up layers consumes more substrate process capacity even if the number of finished chips does not increase proportionally.

Upstream from Ajinomoto, the dependency becomes more chemical. Ajinomoto's own description of ABF identifies epoxy resin, a hardening system and inorganic microparticle filler as fundamental ingredients. [3] Those materials have to deliver electrical insulation, low thermal expansion, processability and mechanical stability at increasingly demanding geometries.

Exhibit 4

ABF also has an upstream materials dependency

The film itself depends on a smaller set of chemical functions

Ingredient groupRole in the ABF systemInvestment relevance
Epoxy resinOrganic resin baseProvides part of the dielectric and mechanical foundation
Hardening systemCuring chemistryControls formation of the thermoset structure
Inorganic microparticle fillerFunctional fillerSupports dimensional, thermal, and process performance

The ingredient groups are drawn from Ajinomoto's public description of ABF. This does not identify specific upstream suppliers or imply that formulations are interchangeable.

Source: Ajinomoto

This upstream layer is easier to overlook because the raw material value per package can be small relative to the finished accelerator. That does not make it economically irrelevant.

A useful supply-chain distinction is the difference between cost share and production criticality. A material can represent a modest portion of a substrate maker's total costs while still being difficult to replace inside a qualified manufacturing process. Conversely, a substrate manufacturer can buy relatively inexpensive dielectric and chemical inputs but create significant economic value through manufacturing precision, yield management and customer qualification.

That asymmetry is why the ABF chain should not be analyzed by revenue size alone.

The second-order AI exposure is substrate process capacity

The obvious AI semiconductor exposures remain accelerator designers, foundries and HBM producers. ABF substrates introduce a different group of companies whose economics are connected to the physical complexity of those chips.

Ajinomoto benefits from greater demand for advanced build-up dielectric. IBIDEN and its peers are exposed to the manufacturing burden created when AI packages become larger and more complex. Nan Ya PCB's move toward next-generation AI ASIC substrates shows that this demand is broadening beyond conventional GPU architectures into hyperscaler-designed silicon. [2]

The common dependency is the substrate process itself.

Exhibit 5

The ABF investment map extends both upstream and downstream

Different companies participate through different constraints

LayerSelected companies or inputsExposure described in the article
Upstream chemistryEpoxy resin, hardening chemistry, inorganic fillersInputs required to formulate ABF
Build-up dielectricAjinomotoABF film used inside advanced package substrates
Advanced substrate manufacturingIBIDEN, Nan Ya PCB, Unimicron, Kinsus, Shinko ElectricQualified conversion of ABF and copper into high-layer-count substrates
Advanced package assemblyPackaging houses and foundriesIntegration of substrate with logic, memory, and package structures
AI acceleratorsGPUs and custom AI ASICsEnd demand that increases package area, layer count, and process intensity

This is a role map, not a ranking of companies or a complete substrate supply chain.

Source: Altsets research

That makes one distinction especially important for investors: growth in ABF demand and growth in finished substrate capacity are not interchangeable. If dielectric film supply expands faster than qualified substrate manufacturing, the bottleneck remains at the converter. If substrate manufacturers expand aggressively, attention can move upstream toward the specialized resins, fillers and other materials required to sustain higher production volumes.

Conclusion

The ABF supply chain behind AI accelerators is not a single-company story. Ajinomoto supplies a critical dielectric, but AI packaging demand becomes economically concentrated again at substrate manufacturers such as IBIDEN and Nan Ya PCB, where ABF must be transformed through increasingly intensive build-up and copper-patterning processes.

The most important variable is therefore not simply how many AI accelerators are produced. It is how much substrate area, how many layers and how much process capacity each new generation consumes. As AI packages become larger and more complex, that manufacturing burden pushes the investment exposure both downstream into advanced substrate producers and upstream into the specialized chemical materials from which ABF itself is made.

Sources

  1. "Notice Regarding Capital Investment Plan for High-Performance IC Package Substrates," IBIDEN Co., Ltd., February 3, 2026, https://www.ibiden.com/company/2026/02/notice-regarding-capital-investment-plan-for-high-performance-ic-package-substrates.html

  2. "2026 Investor Conference Presentation," Nan Ya Printed Circuit Board Corporation, May 6, 2026, https://www.nanyapcb.com.tw/nypcb/images/InvestorRelations/FinancialReport/NYPCBENG20260506.pdf

  3. "Ajinomoto Build-up Film (ABF)," Ajinomoto Group, https://www.ajinomoto.com/innovation/our_innovation/buildupfilm

How to Cite This

According to Altsets Supply Chain Intelligence (altsets.com), the ABF supply-chain constraint is increasingly split between dielectric availability and the qualified substrate process capacity required to convert ABF into larger, higher-layer-count AI accelerator substrates.

For research inquiries or data access: press@altsets.com

Go Deeper

Altsets Research

Get new Altsets research when we publish.

Original supply-chain research, market casefiles, and new data findings.

Research updates only. Unsubscribe any time.

Methodology

Read the methodology for this research.