EMIB Shifts the Packaging Bottleneck to Substrates
emib-substrate-shift · conviction medium · status playing-out · horizon 2027 · as of 2026-07-22
EMIB adoption moves the AI-packaging bottleneck from CoWoS interposers to high-end FC-BGA substrates by 2027, re-rating the substrate cluster (Ibiden, Shinko, Unimicron, AT&S, LG Innotek) while eroding the pure CoWoS-scarcity premium at the margin.
Rests on filed figures, not on modelled shares. 17 premises (5 entity, 4 signal, 3 field, 3 edge, 2 weight); no derived cell is involved, so the undisclosed supply weights that put a range on other pages in this bank cannot move this one.
Exhibits
Exhibit 1Relative performance, indexed to 100How the names in this thesis have traded against SOXX.
Series available as data/emib-substrate-shift.csv
Exhibit 2Who pays ABF substrate and build-up film supply, and who keeps the moneyCapturers average 10.8% operating margin against payers' 37.9% — the owners of the scarce thing earn LESS than the names it constrains.
Green/blue = model marks it as CAPTURING the rent (unbound and supplies the scarce good); faded = PAYING it (bound severe or moderate). Operating margin, live.
Exhibit 3What the conviction is actually made ofEach premise and the number it composes to. A conjunction of plausible premises is far weaker than any of them.
The AI-packaging bottleneck is real and TSMC-concentrated today
† 1 premise marked supporting — shown and arguable, but the conclusion does not depend on them, so they are not multiplied into the composed figure. Citing a filed figure should not cost conviction.
One gating premise, so the conclusion is exactly as strong as it. The figure is an ordering device, not a calibrated probability — see how the numbers are made.
Weakest link: Taiwan Semiconductor Manufacturing Company — 90% supply share of CoWoS (Chip-on-Wafer-on-Substrate) at 0.80 — A 0.9 supplier weight concentrates the scarcity rent at one vendor, which is the condition the variant says gets competed around. The weight is a desk
EMIB is a credible second path, not vaporware
† 1 premise marked supporting — shown and arguable, but the conclusion does not depend on them, so they are not multiplied into the composed figure. Citing a filed figure should not cost conviction.
60% if the 3 gates are independent, 70% if they move together. They are claims about one industry, so the truth is between and nobody can say where. Treat this as an ordering device rather than a calibrated probability — the ranking of premises is the information, not the level.
Weakest link: SK Hynix — signal 2026-07-21 at 0.70 — The dominant HBM maker testing HBM-on-EMIB means the path plugs into the main demand artery. Held at 0.70: SK Hynix discloses no production timeline a
If EMIB scales, the scarce step migrates to FC-BGA substrates
† 1 premise marked supporting — shown and arguable, but the conclusion does not depend on them, so they are not multiplied into the composed figure. Citing a filed figure should not cost conviction.
31% if the 4 gates are independent, 55% if they move together. They are claims about one industry, so the truth is between and nobody can say where. Treat this as an ordering device rather than a calibrated probability — the ranking of premises is the information, not the level.
Weakest link: Elite Material Co., Ltd. (EMC) at 0.55 — EMC is the CCL layer the thesis runs through, and it carries NO financials — every field is null because the entity was built from a channel note. It
The bottleneck migration pulls a PASSIVES layer with it, and that layer is supply-constrained independently
36% if the 3 gates are independent, 60% if they move together. They are claims about one industry, so the truth is between and nobody can say where. Treat this as an ordering device rather than a calibrated probability — the ranking of premises is the information, not the level.
Weakest link: Silicon capacitor (in-package decoupling) at 0.60 — In-package silicon capacitors take the power-integrity role MLCC cannot fill at accelerator current density, with Powerchip named as exclusive supplie
HBM integration gives the migration a demand anchor
52% if the 3 gates are independent, 70% if they move together. They are claims about one industry, so the truth is between and nobody can say where. Treat this as an ordering device rather than a calibrated probability — the ranking of premises is the information, not the level.
Weakest link: SK Hynix — signal 2026-07-21 at 0.70 — Same signal as the credibility leg, cited again here for the DEMAND anchor rather than for path viability. NOTE: reusing one signal across two conclus
The variant
Consensus
CoWoS capacity — TSMC ~90% of 2.5D AI packaging — is the single binding constraint on accelerator shipments through at least late 2027; TSMC's expansion (120k-140k wpm in 2026, +60% by 2027, two Arizona packaging fabs in a $265B US buildout) is the read on AI supply, and substrate makers are secondary beneficiaries of a TSMC-led cycle.
Variant
Intel's EMIB scales as a real second path (~90% yields, Google TPU v8e 2H27, Meta CPU 2H28, customers prepaying substrate suppliers, SK Hynix testing HBM-on-EMIB); because EMIB replaces the silicon interposer with bridges in a large FC-BGA substrate, the scarce step migrates to high-end ABF substrate supply — already at full utilization with the film input repricing +30% — and the substrate cluster becomes the levered play while the pure CoWoS-scarcity narrative erodes at the margin.
Differentiator
Consensus watches interposer wafers-per-month; the variant watches substrate qualification, prepayments, and ABF pricing. The tell the market is missing: Intel's customers are prepaying SUBSTRATE suppliers, not bridge capacity — the constrained input in the EMIB path is the substrate, and every EMIB win converts interposer scarcity into substrate scarcity.
Stance Bottleneck migration over interposer scarcity: every EMIB win converts CoWoS interposer demand into FC-BGA substrate demand, so own the substrate incumbents (Ibiden first) rather than the TSMC-scarcity narrative — hedged for the 2027 CoWoS capacity landing and the 2027-28 substrate capex wave.
Recommendations
LONG ibiden — cleanest incumbent expression: top-tier FC-BGA qualification, the richest margins in the cluster (GM ~31.6%, OM ~14%), Intel's historic substrate anchor, and a ~JPY 280bn expansion already approved. Mechanism: every EMIB program routes through high-end FC-BGA where Ibiden is the most-qualified source. Caveat: fwd P/E ~75 already pays for substrate scarcity — size for the re-rating to come from earnings, not multiple.
LONG ats — the least-priced incumbent leg: EMIB-class qualified, the only advanced-substrate footprint outside Asia (supply-diversification kicker), smallest revenue base so EMIB volume moves the needle most. Caveat: capex-heavy balance sheet and it is already Vienna's best 2026 performer; financials unverified in the model (open question).
LG Innotek — highest torque, but PARTLY PRICED: the stock is up ~4x on the substrate story while still at sample stage. Prefer buying the qualification print (Intel/SK Hynix production award) or sizing small pre-print; the 4x move already discounts a successful entry. Mechanism: direct emib -> lg-innotek edge; sample-to-qualified is the whole trade.
RELATIVE: long substrate cluster vs. the CoWoS-overflow scarcity rents (ase-technology, amkor priced as capacity-of-last-resort) — if the bottleneck migrates to substrates, overflow interposer capacity de-scarcifies first. Do NOT express by shorting tsmc: at ~19.5x fwd earnings with packaging a small revenue share, the CoWoS premium erosion is a narrative change, not an earnings event.
AVOID intel as the expression despite EMIB being Intel's technology: ~86.6x fwd P/E on negative operating margin means the packaging win is diluted by node-execution and foundry-loss risk; the substrate cluster is the pure lever on the same flow. EMIB traction is a signal for Intel Foundry credibility, not the trade.
Second order
Substrate incumbents re-rate on a second demand engine: path emib -> fc-bga substrate demand -> ibiden / shinko / unimicron / ats (Intel already partners_with ibiden; 4 Taiwan + 2 Japan suppliers in prepay talks). Cascade shows the cowos node currently touches 179 downstream entities / ~$44T market cap — even marginal re-routing of that flow through substrates is large relative to a ~$80B substrate-cluster cap.
LG Innotek becomes an EMIB-levered entrant: direct graph edge emib -> lg-innotek (weight_assumed 0.5 — true dependency unresearched, flag). Qualification converts the edge from optionality to revenue.
Intel packaging/foundry credibility improves: path emib -> intel — prepaying customers are the first external parties underwriting Intel manufacturing capacity with cash.
TSMC's packaging scarcity premium erodes at the margin: path emib -> intel -> tsmc (2-hop competitive edge in graph). Narrative, not earnings: CoWoS stays sold out near-term and TSMC still guides >40% 2026 growth.
ASE/Amkor overflow rents fade: cowos residual capacity (TSMC outsources overflow to ase-technology / amkor) de-scarcifies first if demand re-routes to the EMIB/substrate path.
SK Hynix gains packaging optionality: path sk-hynix -> emib co-development — HBM shipments become less gated on TSMC CoWoS slots.
Third order
Theme feedback: advanced-packaging-bottleneck (status: consensus) turns CONTESTED — the theme's longs (tsmc, ase-technology, amkor) carry the CoWoS-scarcity framing that the variant erodes, while its affects-list substrate names become the new longs. Path: emib -> substrate bottleneck -> theme re-scoring.
AI capex durability: a second packaging path un-gates accelerator supply overall (CoWoS gap 20% -> 10% + EMIB relief). Path: emib -> gpu-accelerator -> nvidia / hyperscalers. More units shipped supports the ai-capex-cycle in volume terms while weakening every scarcity-rent story built on packaging tightness.
Intel Foundry credibility loop: EMIB prepayments are evidence external customers will fund Intel capacity. Path: emib -> intel -> intel-foundry external wins -> us-chips-act policy thesis. Strengthens the case that 18A attracts marquee commitments; supports the ~$673B market cap that currently prices foundry option value.
Korea supply-chain diversification: path emib -> lg-innotek -> sk-hynix builds a non-Taiwan HBM-logic assembly corridor, reducing taiwan-risk concentration in advanced packaging — cross-theme with geographic-derisking (same force as TSMC Arizona and AT&S Europe).
Chokepoint migration upstream to materials: if substrates become the bottleneck, the single point of failure becomes ABF film — Ajinomoto ~95% share, +30% price hike — an entity the ontology does not yet model. A substrate-led thesis creates a new, MORE concentrated chokepoint one layer up.
Indicators
name: TSMC CoWoS capacity and supply-demand gap · entity: tsmc · measure: CoWoS wafers/month (TSMC + OSAT overflow) and unmet-demand gap · current: TSMC ~120k-140k wpm est. 2026, OSAT partners adding 50k-60k wpm (~200k industry); gap ~20% now, seen narrowing to ~10% by end-2026; TSMC plans >60% capacity growth by 2027 · consensus: CoWoS remains THE binding constraint through at least late 2027; expansion still lags AI demand (Goldman) · variant: Gap closes faster at the margin as EMIB absorbs overflow demand; by 2027 CoWoS is no longer the scarce step and the scarcity premium in the packaging narrative erodes · unit: wafers/month; % unmet demand · source: https://www.trendforce.com/news/2026/06/15/news-tsmc-cowos-supply-demand-gap-reportedly-seen-narrowing-from-20-to-10-by-end-2026-as-capacity-expands/
name: EMIB external adoption and substrate prepayments · entity: emib · measure: Named external EMIB customer programs and prepay-backed substrate commitments · current: 2 confirmed external programs (Google TPU v8e 2H27, Meta in-house CPU 2H28); Intel CEO Lip-Bu Tan says EMIB-T customers back substrate prepayments; 4 Taiwan + 2 Japan substrate suppliers seeking upfront commitments; EMIB yields reportedly ~90% · consensus: EMIB is a niche/Intel-internal path; TSMC keeps ~90%+ of AI 2.5D packaging; external programs are late (2H27/2H28) and small vs. NVIDIA's CoWoS pull · variant: 3+ external EMIB programs in volume production by end-2027 with prepay-converted substrate contracts — enough share shift to break the CoWoS-monopoly framing · unit: customer programs; prepayment commitments · source: https://www.trendforce.com/news/2026/05/20/news-intel-says-emib-customers-back-substrate-prepayments-4-taiwan-and-2-japan-suppliers-reportedly-seek-commitments/
name: ABF film / FC-BGA substrate pricing and supply gap · entity: ibiden · measure: ABF film price changes; substrate supply-demand gap duration; incumbent utilization · current: Ajinomoto (~95% of ABF film) raising prices ~30% from Q3 2026; substrate supply-demand gap extended through end-2027; Taiwanese substrate makers at full capacity · consensus: Gap resolves as announced plants land: Unimicron new Taiwan plant 2H27, Ibiden Japan expansion 2027, Shinko Malaysia fab late 2027, AT&S Austria line 2027-28 · variant: Substrate stays the binding constraint into 2028 — EMIB volume adds demand faster than plants ramp; substrate pricing power outlasts interposer scarcity, proving the bottleneck moved · unit: % price change; months of supply gap · source: https://www.atlaspcb.com/news/news-ajinomoto-abf-substrate-price-increase-2026/
name: LG Innotek EMIB substrate qualification · entity: lg-innotek · measure: Sample-to-production conversion in the Intel EMIB substrate chain · current: EMIB FC-BGA samples shipped to SK Hynix; pushing to enter Intel's EMIB chain; Vietnam substrate capacity expanding; substrate revenue +16%; NOT yet production-qualified · consensus: Sample-stage entrant; the four incumbents keep the roster; high-end FC-BGA qualification takes years and the 4x stock move is ahead of realized revenue · variant: Production qualification with Intel and/or SK Hynix by 2027 — the entrant leg validates and confirms substrate (not interposer) is where new supply is being recruited · unit: qualification milestone (sample / qualified / volume) · source: https://www.digitimes.com/news/a20260604VL211/lg-innotek-fc-bga-intel-emib-supply-chain.html
name: Substrate incumbent capex commitments · entity: unimicron · measure: Announced substrate capex and whether customer prepayments underwrite it · current: Ibiden approved ~JPY 280bn additional substrate investment and sees FY2026 sales +~20%; all four incumbents have announced expansions ramping 2027-28; suppliers demanding upfront commitments before adding EMIB capacity · consensus: Disciplined capex after the 2022-23 ABF overbuild; expansions are sized to the TSMC/CoWoS-era demand curve · variant: Prepay-underwritten capex acceleration — customers de-risking substrate expansion is the market conceding substrate is the scarce step; capex announcements broaden and pull forward through 2027 · unit: capex announcements (JPY/USD bn); prepay-backed share · source: https://www.trendforce.com/news/2026/05/20/news-intel-says-emib-customers-back-substrate-prepayments-4-taiwan-and-2-japan-suppliers-reportedly-seek-commitments/
name: SK Hynix EMIB co-development milestones · entity: sk-hynix · measure: HBM-on-EMIB integration progress from R&D to a qualified production path · current: SK Hynix testing HBM + logic integration on Intel-supplied EMIB substrates; runs a small domestic 2.5D packaging R&D line; reviewing materials/component suppliers for potential mass production; no production timeline disclosed · consensus: A hedge only — TSMC remains SK Hynix's primary 2.5D/HBM-base-die partner; EMIB work never reaches production scale · variant: EMIB path qualified for HBM-logic assembly by 2027, breaking the CoWoS mono-routing of HBM integration and anchoring a Korea packaging corridor · unit: milestone (R&D / supplier review / qualified / production) · source: https://www.trendforce.com/news/2026/05/11/news-sk-hynix-reportedly-tests-intel-emib-2-5d-packaging-with-hbm-amid-tsmc-cowos-tightness/
Tripwires
Ajinomoto announces an ABF film capacity doubling, or a customer qualifies a second-source build-up film — the upstream chokepoint clears and the materials leg of substrate pricing power dies first; cut the scarcity-premium sizing.
NVIDIA qualifies a non-CoWoS package (EMIB, I-Cube, or panel-level) for a volume accelerator — the thesis accelerates beyond the variant; re-size the substrate cluster before the qualification print becomes modeled revenue.
TrendForce or TSMC guidance shows the CoWoS gap at <=5% before end-2026 — scarcity is clearing everywhere ahead of schedule; test immediately whether substrate pricing holds without a packaging queue behind it.
LG Innotek prints an Intel or SK Hynix production award — the entrant leg converts from optionality to revenue; execute the buy-the-print plan in recommendations before consensus models it.
Falsifiers
TSMC's >60% CoWoS expansion lands on schedule and the supply-demand gap goes to ~0 WITH substrate pricing rolling over alongside — scarcity disappears everywhere, so there is no bottleneck to migrate and the whole re-rating leg fails.
EMIB external adoption stalls at the two late programs (Google 2H27, Meta 2H28) with no third win by mid-2027 — EMIB is niche, not a shift, and the second demand engine consensus models omit never arrives.
The 2027-28 substrate capacity wave (Unimicron 2H27, Ibiden 2027, Shinko late 2027, AT&S 2027-28, LG Innotek entering) overshoots into a repeat of the 2022-23 ABF bust — the thesis is right on flows and dead wrong on pricing, and the cluster de-rates from 66-75x fwd.
SK Hynix deepens TSMC coupling via HBM4 base-die integration and shelves the EMIB line — HBM integration stays mono-routed through CoWoS and the Korea-corridor and diversification legs die.
Open questions
Prepayment DOLLAR amounts undisclosed — no public figure for the size of EMIB substrate prepay commitments (indicator tracks existence/conversion, not magnitude).
EMIB wafer-equivalent capacity is not published — cannot directly compare EMIB wpm vs. CoWoS 120k-140k wpm; adoption proxied by customer programs instead.
AT&S and LG Innotek current market cap / fwd P/E null in the model — comps check for two of five cluster names is qualitative (both flagged as big 2026 movers).
Which 4 Taiwan suppliers seek EMIB commitments? Only Unimicron is confirmed Taiwanese; AT&S is Austrian — roster reconciliation open (inherited from emib entity).
emib -> lg-innotek and emib -> intel edge weights are assumed (0.5 default) — the cascade/exposure magnitudes for the entrant leg are not calibrated.
Shinko is private (JIC) — no market read-through; its EMIB share must be inferred from Ibiden/Unimicron prints.
Reasoning chain
The AI-packaging bottleneck is real and TSMC-concentrated todayVALID
CoWoS is flagged severe in the model and every packaging thesis in the bank routes through it.
why
Held at 0.85 rather than higher because 'severe' is a desk classification, not a measured utilisation rate, and the gap is reported to be narrowing from ~20% to ~10%.
Taiwan Semiconductor Manufacturing Company — 90% supply share of CoWoS (Chip-on-Wafer-on-Substrate)0.80 strong
A 0.9 supplier weight concentrates the scarcity rent at one vendor, which is the condition the variant says gets competed around.
why
The weight is a desk estimate, and OSAT overflow of 50-60k wpm against TSMC's 120-140k is already visible in the same source.
A severe-constraint supply step held ~90% by one vendor concentrates scarcity rent at TSMC — the condition the variant says gets competed around.
EMIB is a credible second path, not vaporwareVALID
Intel owns the path, which is what makes it a genuine second source rather than a TSMC variant — and also what caps it, since capacity is one company's to allocate.
Two confirmed external programs (Google TPU v8e 2H27, Meta 2H28) are late and small against NVIDIA's CoWoS pull. Held at 0.65 because ~90% yields are reported but unverified, and Clearwater Forest needing TWELVE bridges suggests the cost advantage over an interposer largely evaporates at high die count.
SK Hynix — signal 2026-07-210.70 moderate
The dominant HBM maker testing HBM-on-EMIB means the path plugs into the main demand artery.
why
Held at 0.70: SK Hynix discloses no production timeline and runs this as a small domestic R&D line, which is consistent with a hedge rather than a commitment.
A ramping Intel-owned packaging path with an HBM co-development signal from the dominant HBM maker means the alternative is being engineered, not just discussed. QUALIFIED 2026-08-04 BY INTEL'S OWN PRODUCT. Clearwater Forest uses TWELVE EMIB bridges embedded in organic substrate to join three Foveros base dies, for 29 total silicon pieces in one package. Each bridge is process steps and yield loss, so at that count the cost advantage EMIB is supposed to hold over a silicon interposer largely evaporates — a point made in the same source that documents the design. THE THESIS ASSUMES EMIB IS THE CHEAPER PATH. If Intel needs twelve bridges to make 18A economic, EMIB is credible as ENGINEERING and unproven as ECONOMICS, and those are different claims. What would settle it: a disclosed cost-per-package comparison against CoWoS at equivalent die count, or Intel winning external EMIB business on price rather than on capacity access.
If EMIB scales, the scarce step migrates to FC-BGA substratesVALID
EMIB is constrained but less so than CoWoS, which is what makes it able to absorb overflow demand.
why
If it were equally constrained there would be no migration to argue about.
LG Innotek supplies IBIDEN Co., Ltd.1.00 strong
3.72%
A small competing supplier set is where a queue re-forms.
why
Held at 0.60 because LG Innotek is sample-stage and NOT production-qualified, so the entrant leg is optionality rather than supply — and its 3.72% operating margin is the thinnest in the cluster.
Advanced Packaging Bottleneck — signal 2026-07-210.75 strong
The bottleneck is a live desk theme with its own evidence trail, so this premise rests on an argued position rather than an assertion made here.
Emib Substrate Shift — signal 2026-08-030.75 strong
Channel evidence upgrades the migration from FORECAST to OBSERVATION: packages reported waiting on substrate and a named TPU program slipped one to two quarters on substrate yield.
why
A delay attributable to substrate rather than logic or interposer is the cleanest available evidence the scarce step has already moved. Capped at 0.75 by two details in the SAME source: the announced capacity is REPURPOSABLE, and CCL price increases are largest in the LOWER grades — a capacity-reallocation signal, not a demand signal.
Elite Material Co., Ltd. (EMC)0.55 moderate
EMC is the CCL layer the thesis runs through, and it carries NO financials — every field is null because the entity was built from a channel note.
why
It cannot be scored by recommend.py or carry a pricing-power verdict, so it supports the mechanism and cannot price it.
EMIB replaces the interposer but consumes high-end substrate; a small competing supplier set (incumbents + one entrant) is where the queue re-forms — scarcity moves to the slowest-scaling input. The 2026-08-03 channel evidence upgrades this from FORECAST to OBSERVATION: accelerator packages are reported waiting on substrate and a named TPU program slipped one to two quarters specifically on substrate yield, and a delay attributable to substrate rather than to logic or interposer is the cleanest available evidence the scarce step has already moved. Held back from a larger re-rating by two details in the SAME source: the announced capacity is REPURPOSABLE (optional, not committed), and the CCL price increases are largest in the LOWER grades — a capacity-reallocation signal, not a demand signal.
The bottleneck migration pulls a PASSIVES layer with it, and that layer is supply-constrained independentlyVALID
premises
Multilayer ceramic capacitor (high-capacitance, AI server)0.70 moderate
10x
AI servers use >10x the high-capacitance MLCC content of standard servers, with Murata high-cap lead times at ~30 weeks and price increases pushed through.
why
Held at 0.70 because the content multiple and the lead time are both unattributed channel figures, though lead-time extension plus price increases is a signature this desk verifies elsewhere.
Murata Manufacturing Co., Ltd.0.85 strong
15.39%11.5 bn
The constrained supplier is in the model with primary-sourced financials: 15.39% operating margin on $11.5bn revenue, FY to 2026-03-31 from JPX filings via J-Quants.
In-package silicon capacitors take the power-integrity role MLCC cannot fill at accelerator current density, with Powerchip named as exclusive supplier for EMIB-T.
why
Powerchip monthly revenue NT$6.47bn, +68.8% YoY from TPEx. Held at 0.60 because the exclusivity claim is unattributed and single-source status in a ramping category is as much a risk as a moat.
The thesis has been argued at the substrate layer alone. Packaging change pulls a component layer with it, and that layer has its own constraint: board-level bulk capacitance at 30-week lead times AND a new in-package category ramping at +68.8% YoY. CRITICALLY THIS IS EXPANSION, NOT SUBSTITUTION — silicon capacitors take the package role while MLCC keeps the board role, so advanced packaging grows the total capacitor opportunity rather than moving it, and neither name is disrupting the other. That makes Murata a supply-constrained incumbent rather than a contested one, which is a BETTER position than the substrate cluster this thesis is built on occupies. WHAT WOULD CHANGE IT: the 800V distribution shift in the 3DHI roadmap moves conversion closer to the die and moves capacitance into the package with it. That is the mechanism by which complementary becomes substitutive, and it is the thing to watch rather than current share.
HBM integration gives the migration a demand anchorVALID
premises
SK Hynix — 62% supply share of High-Bandwidth Memory (HBM3E / HBM4)0.75 strong
A 0.62 share makes SK Hynix the marginal decider on HBM integration routing.
why
The weight is a desk estimate rather than a disclosed share.
SK Hynix supplies NVIDIA Corporation1.00 strong
The #1 HBM supplier selling into the #1 accelerator maker is the main demand artery, and this edge is one the spillover study validates for co-movement.
SK Hynix — signal 2026-07-210.70 moderate
Same signal as the credibility leg, cited again here for the DEMAND anchor rather than for path viability.
why
NOTE: reusing one signal across two conclusions means those conclusions are not independent, so the two composed numbers should not be multiplied together downstream.
The #1 HBM supplier to the #1 accelerator maker validating HBM-on-EMIB means the alternative path plugs into the main demand artery rather than a niche.
Sources
TSMC CoWoS capacity 120k-140k wpm 2026, OSAT +50k-60k wpm, gap narrowing 20% to ~10% by end-2026, >60% capacity growth by 2027 — linkacc 2026-07-22
Intel CEO: EMIB-T customers back substrate prepayments; 4 Taiwan + 2 Japan suppliers seek commitments; Google TPU v8e 2H27, Meta CPU 2H28; Ibiden ~JPY 280bn substrate investment, FY2026 sales +~20% — linkacc 2026-07-22
SK Hynix testing HBM integration on Intel-supplied EMIB substrates amid CoWoS tightness; EMIB yields ~90%; small domestic 2.5D R&D line — linkacc 2026-07-22
Ajinomoto raising ABF film prices ~30% Q3 2026; substrate supply-demand gap through end-2027; incumbent expansion timelines (Unimicron 2H27, Ibiden 2027, Shinko late 2027, AT&S 2027-28) — linkacc 2026-07-22
Consensus framing: CoWoS capacity, not wafers, is the binding constraint through at least late 2027; TSMC $265B US commitment incl. two packaging fabs; Goldman sees capacity lagging AI demand; TSMC Q2 2026 revenue $40.2B +36%, FY26 growth guided >40% — linkacc 2026-07-22
LG Innotek targets Intel EMIB substrate chain with SK Hynix samples; Vietnam expansion — linkacc 2026-07-22
LG Innotek substrate revenue +16%; ECTC pitch after 4x stock surge — linkacc 2026-07-22
Intel Foundry design-win momentum; EMIB yields reported up to ~98% — linkacc 2026-07-22
Intel entry-level advanced packaging draws interest from Google and Amazon — linkacc 2026-07-22
JPMorgan upgrades Ibiden to Overweight on AI substrate growth potential — linkacc 2026-07-22
LG Innotek targets $700M FC-BGA business by 2030 (Dream Factory); ~$410M government-backed expansion; 1T-won operating-profit target in substrates within five years — linkacc 2026-07-22
Ajinomoto ABF price hike ~30%; supply-demand imbalance to persist through 2027 — linkacc 2026-07-22