ABF substrates are the one shared bottleneck in the GPU–custom-ASIC race

For AI accelerators—whether NVIDIA GPUs or hyperscaler custom ASIC packages—the package is the unit that has to physically connect the die to the board. In advanced laminate substrates, one insulating film layer is repeatedly central: Ajinomoto Build-up Film (ABF). Ajinomoto describes ABF as an interlayer insulation film used to electrically insulate complex circuit substrates for high-performance CPUs, and as an “essential material” used to form the build-up circuitry between chip-scale terminals and substrate terminals.

On the substrate manufacturing side, Ibiden (a major producer of high-performance IC package substrates that use ABF-based laminate stacks) disclosed a large, explicit expansion plan. In its capital investment notice for high-performance IC package substrates, Ibiden sets a planned period of FY2026–FY2028 with total investment of approximately 500 billion yen, including a first-phase additional investment of approximately 220 billion yen aimed at expanding production capacity, with planned mass-production commencement from fiscal year 2027.

The key causal chain is not “ABF causes AI demand.” It’s more mechanical: when AI designs move into packaging tiers that rely on ABF-based build-up substrates, the constraints become upstream in film availability and downstream in laminate fabrication throughput. That means a build plan can be simultaneously “design-ready” yet still constrained by ABF-driven packaging capacity.

Ajinomoto’s consolidated results show scale and profitability through FY2024–FY2025, giving the company the balance-sheet capacity to fund production scaling in the specialty chemicals layer. This matters to packaging planners and investors because when pricing moves (as it did for ABF in Q3 2026 timing), the ability to sustain and fund capacity additions can determine whether tightness becomes a temporary margin bump or a longer earnings cycle.

ABF is the film ingredient; laminate substrate makers are the throughput layer that converts film into build-up packages. In parallel, substrate expansion announcements (like Ibiden’s FY2026–FY2028 program) indicate that substrate output can increase, but the ramp is multi-year. That timing gap is where build plans get constrained.

In the near term (days to quarters), watch for evidence that substrate makers are prioritizing orders that can be supported by available ABF film and laminate line schedules—pricing and lead times are the first transmission channel. In the medium term (through FY2027 and into FY2028), the market can re-rate the constraint as Ibiden’s stated mass-production timing from fiscal year 2027 approaches, assuming the expansion converts to usable output without unexpected yield and customer qualification delays.

The key investment idea is not that NVIDIA loses and custom ASICs win. It’s that both are gated by the same physical stack when they move into premium advanced packaging: ABF-enabled build-up substrates. Ajinomoto’s Q3 2026 ABF pricing increase is the upstream proof that demand pressure is concentrated in the film layer, while Ibiden’s FY2026–FY2028 substrate expansion schedule is the downstream proof that capacity recovery will take time.

If you want a “supply-chain-only” read on whether AI hardware build-outs accelerate or stall, ABF is close to the source of truth. When upstream film pricing tightens while laminate capex is ramping, the ceiling shifts from silicon to packaging throughput—regardless of whether the workload runs on GPUs or custom silicon.