Table of Contents
Buyers need a qualified delivery date, not a substrate expansion headline. A new factory does not add the same amount of usable supply to every AI processor program; package dimensions, build-up layers, qualified processes, yield and customer reservations decide which products can use the expansion—and when.
The ABF substrate question is not simply whether capacity is increasing. It is whether a specific expansion can deliver the qualified package substrate needed by a named device within its production schedule. Ajinomoto describes ABF as an insulating film used in build-up substrates. Film supply and finished-substrate output are separate stages.
What do the October announcements actually establish?
They establish planned manufacturing expansion and customer commitments, with different scopes and dates. They do not establish an industry-wide reduction in substrate lead time or availability for every customer.
NVIDIA announced on October 21, 2026 that agreements with IBIDEN, Samsung Electro-Mechanics and Unimicron would support an aggregate 2.5-fold increase in substrate capacity for its products over three years. The three ABF substrate manufacturers named in the announcement are tied to NVIDIA’s product programs, not to a general market allocation. The customer-specific scope matters: that is not a claim that all merchant substrate capacity will increase by the same factor.
In its October 30 first-half FY2026 results presentation, IBIDEN links higher production load to larger, more multilayered IC package substrates. It identifies sequential operation starts from FY2027 for Ono Cell8 and FY2029 for Gama Cell6. These are the company’s fiscal-year references and should not be silently converted into calendar-year delivery dates.
The announcements come from different corporate issuers but concern a connected supply chain. They are corroborating signals of investment and complexity, not two independent measurements of a market shortage. SupplyICs editorial analysis is that buyers should examine qualified output and allocation more closely as package complexity grows.
Why can substrate demand grow faster than chip unit demand?
A larger or more complex package can consume more manufacturing resources per delivered chip. Unit shipments alone therefore understate the load on substrate fabrication when package area, layer count or process complexity increases.
A simple area comparison illustrates the issue. Increasing a square substrate from an assumed 80 mm side to 100 mm increases its area from 6,400 to 10,000 mm², or 56.25%. That calculation does not include panel utilization, additional layers, process steps, edge losses or yield.
If a supplier reports capacity in panels per month and a customer forecasts devices per month, neither number can be compared directly until the product mix and usable pieces per panel are known. A higher layer count may also change bottlenecks within the factory even when the final package footprint is unchanged.
Ask which unit is being quoted: substrate pieces, processed area, panel starts, a particular process-stage load, or qualified good units. Retain the original unit and date in the procurement record.
Translate capacity into qualified supply
Build the delivery estimate from the limiting qualified process, then apply the applicable yield and customer allocation. Do not treat a building opening, equipment installation or pilot start as equivalent to an accepted production lot.
| Reported milestone | What it supports | What remains to be established |
|---|---|---|
| Capital expenditure approved | Funding and investment intent | Equipment availability, installation and qualification |
| Factory or line begins operation | A production stage has started | Product-specific yield, reliability and accepted output |
| Substrate design qualified | A defined design/process combination is accepted | Sustained volume and delivery allocation |
| Capacity reserved for a customer | A contractual allocation framework exists | Exact product, dates, conditions and downstream readiness |
| Finished package delivered | An actual output milestone | Repeatability at the required future rate |
Take a line with 10,000 substrate starts and 90% substrate yield. It produces 9,000 good substrates. If assembly yield is 95% and final-test yield is 98%, the modeled output is 10,000 × 0.90 × 0.95 × 0.98 = 8,379 good packages. The assumptions make the arithmetic visible; they are not a forecast.
Increasing substrate starts by 10% would raise that modeled result to about 9,217 only if assembly, test and allocation can absorb the extra units. If a downstream stage is already fully loaded, the substrate expansion may instead increase work in process. Use stage-specific capacities and avoid applying a yield twice when a supplier’s capacity number already represents good output.
How should a component buyer use this information?
A buyer of finished ICs should ask the device supplier which package or assembly constraints affect the exact ordering code, and whether acknowledged deliveries already include those constraints. The buyer usually cannot reserve a different substrate and make the finished IC interchangeable.
Capture the full part number, package variant, required quantity, delivery periods and the supplier’s commitment status. If the response cites “packaging capacity,” ask which stage is limiting and what milestone would change the promised date. Preserve uncertainty when the supplier cannot disclose its manufacturing detail.
The existing CoWoS-to-CoPoS roadmap covers packaging-platform development. This substrate analysis adds a different question: whether the supporting package material and process can supply a particular released design. Neither a platform roadmap nor a substrate expansion authorizes a package substitution.
For a BOM sourcing review, separate the constrained processor from supporting memory, power and interface parts. Buying every support component early can create surplus if the processor delivery remains uncertain. Align those commitments to an acknowledged build schedule and the quantities that can form complete kits.
What evidence would change the 2027 assessment?
The assessment should change when a named program moves from planned capacity to qualified, repeatable deliveries, or when demand and package mix materially change. Broad investment figures are less decisive than those program-level milestones.
Track qualification completion, good-unit ramp, customer allocation, assembly/test readiness and actual delivery performance. A slower AI deployment cycle or improved manufacturing yield could reduce pressure; larger packages, additional process steps or concentrated reservations could absorb expansion. These are conditional scenarios, not price forecasts.
File the latest supplier acknowledgment separately from the original announcement. Procurement can then distinguish an improving industrial backdrop from a delivery commitment that has actually become more reliable.
Frequently Asked Questions (FAQ)
Is ABF film capacity the same as finished package-substrate capacity?
No. Film is an input material. Finished substrate output also depends on substrate fabrication, design rules, process load, yield and qualification.
Can a finished-IC buyer specify a different substrate supplier directly?
Usually that change belongs to the device and package manufacturer. A different substrate may require design and qualification work; the finished-IC buyer should seek the manufacturer's approved supply commitment.
Do IBIDEN fiscal-year ramp dates state a calendar delivery month?
No. A fiscal-year operation-start statement does not establish a specific delivery month or customer allocation. Preserve the company's original date basis and obtain program-level confirmation.
