Amkor Technology, Inc.
Combined average of Moat (AI Resilience), Growth, and Valuation scores.
Moat Score
High customer-qualification switching costs and a TSMC-anchored advanced-packaging position — but a capital-intensive, thin-margin, cyclical business with no software lock-in.
Amkor's durability is real but narrow — it rests on switching costs and onshoring, not on the wide network or software moats that protect the rest of the semiconductor value chain:
- Customer Qualification Switching Costs: Advanced packages (HD Fan-Out, 2.5D/SWIFT) are co-developed with the customer over multi-quarter qualification cycles tied to a specific package, substrate, and test flow. Once a part ships in volume, re-qualifying a second source is costly and slow — which is the core source of Amkor's stickiness, even though large customers like Apple ultimately retain dual-sourcing leverage.
- TSMC Co-Location & Onshoring: The October 2025 ten-year capacity-reservation agreement with TSMC and the $7B Arizona campus place Amkor physically next to leading-edge US fabs, supported by CHIPS-era onshoring incentives. This creates a contractual and geographic lock-in that a Taiwan- or China-based OSAT cannot easily replicate for US-bound advanced silicon.
- Process IP & Capital Intensity: Decades of accumulated packaging process know-how and ~$2.5–3.0B/year of capital intensity raise the bar for new entrants. But the flip side is the weakness: ~14% gross margins, deep cyclicality, and the fact that IDMs/foundries (TSMC CoWoS, Samsung, Intel) increasingly pull the highest-value advanced packaging in-house, capping how much of the AI value chain Amkor can own.
Moat Verdict
Amkor's moat is narrow and AI-adjacent rather than AI-resilient: it benefits enormously from AI demand for advanced packaging, but its durability rests on customer-qualification switching costs and the TSMC/onshoring lock-in, not on a defensible data, network or software advantage. The persistent risk is that foundries and IDMs pull the highest-value AI packaging in-house, leaving Amkor a thin-margin, cyclical capacity provider.
57.0 resilient · 59.8 vulnerable · 80/20 = 57.6 · = 58
Open a moat to read its note.
Amkor sells packaging and test services, not a software interface; there is no user-facing UI or workflow that creates relearning cost.
Decades of proprietary packaging process IP (HD Fan-Out, 2.5D, SWIFT, flip-chip) and co-developed customer flows are hard to replicate, though leading foundries and ASE possess comparable capability.
the business does not derive any moat from access to or aggregation of a public data source.
Advanced-packaging process engineers are somewhat scarce, but the talent pool is shared across ASE, JCET, foundries and IDMs, so it is not a defensible standalone moat.
Turnkey assembly + test + (increasingly) advanced packaging sold as an integrated service raises switching friction versus splitting the flow across vendors.
Yield, reliability and process data accumulate internally, but they are operational rather than a compounding, non-replicable data asset that competitors cannot match.
The 10-year TSMC capacity-reservation agreement, multi-quarter customer qualification barriers, and CHIPS-era US onshoring incentives create genuine multi-year switching and entry barriers for the Arizona advanced-packaging position.
adding more customers or packages does not make the service more valuable to other customers; there is no Metcalfe-style network dynamic.
Once qualified, Amkor is embedded in a customer's supply chain at the package level — co-located with TSMC's Arizona fabs and designed into multi-year product roadmaps — though large customers retain dual-sourcing leverage.
Amkor is not the authoritative system of record for any critical business function; it is a manufacturing-services supplier.
One of the two largest OSATs: capacity scale matters to win advanced-packaging programmes, but TSMC's in-house CoWoS and ASE match or exceed it.
Buyers are enterprises choosing on switching cost, integration and performance, which the other pillars rate. The name carries reputation, not a price premium it could hold on brand alone.
Combined average of Moat (AI Resilience), Growth, and Valuation scores.
Moat Score
High customer-qualification switching costs and a TSMC-anchored advanced-packaging position — but a capital-intensive, thin-margin, cyclical business with no software lock-in.
Growth Score
Amkor posted record Q2 2026 revenue of $1.90B (+26% YoY) with gross profit of $319M (16.8%) and net income of $174M, and record revenue in both Computing and Automotive & Industrial. The mix is splitting: Computing rose 20% sequentially on data-center demand and is guided up nearly 30% sequentially in Q3 on AI data center and an HDFO CPU ramp, while Android revenue fell 20% on memory supply and pricing and Communications is guided down high single digits in Q3, against normal seasonality. Q3 is guided to $1.95–2.05B — roughly flat against a record $1.99B Q3 2025 — at an 18.5–19.5% gross margin. Arizona Phase 1 is fully committed, and 2026 capex stays at $2.5–3.0B. The long-range plan targets ~$11B revenue by 2030. The catch is still the cyclicality and customer concentration that have historically whipsawed OSAT growth, plus the capex load on free cash flow during the build-out.
Valuation Score
At ~$52.24 (September 24, 2026) — down ~46% from the $96.68 all-time high of June 16 after a 24% one-day drop on July 28, when a Q3 revenue guide of $1.95–2.05B came in below the ~$2.12B Street consensus despite a record Q2 — AMKR sits between the reset $37 bear and $68 base, about half of the way from bear to base, yielding a valuation score of 78. The selloff came with earnings going the other way: Q2 EPS was $0.70 (vs. ~$0.47 expected) and the Q3 guide of $0.72–0.82 with 18.5–19.5% gross margin annualises to ~$3.08, putting the stock at ~17× run-rate earnings — back inside the historical 10–20× OSAT band rather than the ~45× forward multiple it carried in June. The margin of safety is real but conditional on the Q3 margin step-up holding through the peak Arizona capex year.
The Qualification & Co-Location Moat
Amkor's durability is real but narrow — it rests on switching costs and onshoring, not on the wide network or software moats that protect the rest of the semiconductor value chain:
- Customer Qualification Switching Costs: Advanced packages (HD Fan-Out, 2.5D/SWIFT) are co-developed with the customer over multi-quarter qualification cycles tied to a specific package, substrate, and test flow. Once a part ships in volume, re-qualifying a second source is costly and slow — which is the core source of Amkor's stickiness, even though large customers like Apple ultimately retain dual-sourcing leverage.
- TSMC Co-Location & Onshoring: The October 2025 ten-year capacity-reservation agreement with TSMC and the $7B Arizona campus place Amkor physically next to leading-edge US fabs, supported by CHIPS-era onshoring incentives. This creates a contractual and geographic lock-in that a Taiwan- or China-based OSAT cannot easily replicate for US-bound advanced silicon.
- Process IP & Capital Intensity: Decades of accumulated packaging process know-how and ~$2.5–3.0B/year of capital intensity raise the bar for new entrants. But the flip side is the weakness: ~14% gross margins, deep cyclicality, and the fact that IDMs/foundries (TSMC CoWoS, Samsung, Intel) increasingly pull the highest-value advanced packaging in-house, capping how much of the AI value chain Amkor can own.
Moat Verdict
Amkor's moat is narrow and AI-adjacent rather than AI-resilient: it benefits enormously from AI demand for advanced packaging, but its durability rests on customer-qualification switching costs and the TSMC/onshoring lock-in, not on a defensible data, network or software advantage. The persistent risk is that foundries and IDMs pull the highest-value AI packaging in-house, leaving Amkor a thin-margin, cyclical capacity provider.
57.0 resilient · 59.8 vulnerable · 80/20 = 57.6 · = 58
Open a moat to read its note.
Amkor sells packaging and test services, not a software interface; there is no user-facing UI or workflow that creates relearning cost.
Decades of proprietary packaging process IP (HD Fan-Out, 2.5D, SWIFT, flip-chip) and co-developed customer flows are hard to replicate, though leading foundries and ASE possess comparable capability.
the business does not derive any moat from access to or aggregation of a public data source.
Advanced-packaging process engineers are somewhat scarce, but the talent pool is shared across ASE, JCET, foundries and IDMs, so it is not a defensible standalone moat.
Turnkey assembly + test + (increasingly) advanced packaging sold as an integrated service raises switching friction versus splitting the flow across vendors.
Yield, reliability and process data accumulate internally, but they are operational rather than a compounding, non-replicable data asset that competitors cannot match.
The 10-year TSMC capacity-reservation agreement, multi-quarter customer qualification barriers, and CHIPS-era US onshoring incentives create genuine multi-year switching and entry barriers for the Arizona advanced-packaging position.
adding more customers or packages does not make the service more valuable to other customers; there is no Metcalfe-style network dynamic.
Once qualified, Amkor is embedded in a customer's supply chain at the package level — co-located with TSMC's Arizona fabs and designed into multi-year product roadmaps — though large customers retain dual-sourcing leverage.
Amkor is not the authoritative system of record for any critical business function; it is a manufacturing-services supplier.
One of the two largest OSATs: capacity scale matters to win advanced-packaging programmes, but TSMC's in-house CoWoS and ASE match or exceed it.
Buyers are enterprises choosing on switching cost, integration and performance, which the other pillars rate. The name carries reputation, not a price premium it could hold on brand alone.
Growth Analysis
Growth Drivers
Key Risk
OSAT is a thin-margin (high-teens gross), capital-intensive, cyclical business with meaningful customer concentration (a single large mobile customer is a substantial share of revenue). The 2026 capex of $2.5–3.0B exceeds net income and pressures free cash flow during the Arizona build-out, and foundries/IDMs (TSMC CoWoS, Samsung, Intel) are pulling the highest-value AI advanced packaging in-house. Falsifiable test: if FY2026 free cash flow turns materially negative or the AI advanced-packaging portfolio fails to roughly triple as guided by the Q4 FY2026 print (Feb 2027), the growth-and-margin thesis weakens and the cyclical multiple compresses.
Score Derivation
76.4 base + 1.3 trajectory + 4 margin − 10 risk = 72
Base 76.4 (10–15% CAGR, midpoint 12.5%, baseFromCagr) + 1.3 trajectory (Computing and Automotive & Industrial accelerating, Communications now decelerating: (2 − 1) / 3 × 4) + 4 expanding margin (gross margin 16.8% in Q2 and guided 18.5–19.5% for Q3, against 14.3% in Q3 2025) − 10 high keyRisk. Severity stays high: in the AI-hardware cohort capex-cycle risk is graded by exposure, and a thin-margin OSAT with a large single mobile customer and $2.5–3.0B of 2026 capex sits with SMCI, DELL and MU, not with the wide-margin names. The Communications decline is an observed fact and is charged in its driver's trend, not in severity. The old author string added +3 for TAM expansion; primaryType no longer scores, and the formula computed 73 on the old inputs = 72
Price Scenarios (12–24 Months)
Valuation Multiples
| P/E on Q3 guide run-rate | ~17× |
| Forward P/E (consensus) | ~18× |
| FY2026 EPS to date | ~$1.80 |
| P/E on 2030 Target | ~12× |
| Consensus Target | ~$76 |
AMKR has round-tripped most of its June AI-packaging re-rating even as earnings accelerated: the Q3 guide implies ~$3.08 of annualised EPS, so the stock is back to ~17× — the top of its historical cyclical range rather than a premium to it. The market is pricing the Communications-segment softness, the Q3 revenue miss versus consensus and the $2.5–3.0B capex year; the upside case needs the 18.5–19.5% gross-margin step-up to prove durable and the TSMC/NVIDIA agreements to convert into revenue.
Approximate figures as of September 24, 2026 (price ~$52.24).
Where We Are vs Targets
Loading live price…
Communications (mobile) weakness deepens, the Q3 margin step-up proves temporary, and peak Arizona capex turns free cash flow negative — the stock de-rates to a trough OSAT multiple.
- ~12× the ~$3.08 Q3-guide run-rate EPS (≈$37), the low end of Amkor's historical 10–20× cyclical range
- Communications-segment softness that drove the July guide miss persists into 2027 as a major mobile customer trims orders
- Gross margin slips back from the guided 18.5–19.5% toward the ~14–15% band as utilisation falls
- 2026 capex of $2.5–3.0B pushes free cash flow materially negative and caps shareholder returns
Q3 lands within the $1.95–2.05B / $0.72–0.82 guide, Computing grows ~30% sequentially as guided, and the market credits a modest premium to the historical OSAT multiple for the AI-packaging mix.
- ~22× the ~$3.08 Q3-guide run-rate EPS (≈$68), a modest premium to the 10–20× historical band and below the ~$76 consensus target
- Computing revenue grows nearly 30% sequentially in Q3 on AI data-centre demand and the high-density fan-out CPU ramp
- Gross margin holds in the high teens as advanced-packaging mix rises
- The 10-year TSMC advanced-packaging agreement and multi-year NVIDIA partnership keep the Arizona ramp on schedule
AI/HPC packaging demand outruns capacity, the Arizona campus fills early, and the market re-underwrites the 2030 ~$11B / ~$4.50 EPS plan — returning toward the June highs.
- ~25× ~$3.80 EPS (≈$95) — the ~$3.08 run-rate grown ~25% as Arizona ramps and mix improves; still below the $96.68 June all-time high
- Gross margin moves above 20% as advanced-packaging pricing tightens and utilisation stays full
- NVIDIA and TSMC agreements broaden into additional 2.5D/CoWoS-class volume, taking OSAT share from ASE and JCET
- The 2030 ~$11B revenue / ~$4.50 EPS roadmap is re-credited, supporting a sustained premium multiple