Amkor Technology

Phase A — Understand the business

Company Overview

Amkor is the world's largest U.S.-headquartered OSAT (outsourced semiconductor assembly and test) and the #2 OSAT globally behind Taiwan's ASE. It does not design or fabricate chips. It performs the back-end of the semiconductor flow — wafer bump, probe, back-grind, package design, assembly, burn-in, system-level and final test, drop-ship — on wafers consigned by customers (Amkor never takes title to the silicon, a structural feature that keeps inventory light but also caps pricing power).

How it makes money. Services-revenue model, recognized over time (cost-plus-margin input method, ~2–3 week service cycle). FY2025 net sales $6,708.0M, +6.2% YoY. The book splits two ways:

• Advanced Products (flip-chip, memory, wafer-level/2.5D/HDFO) — $5,555.6M, 82.8% of sales.
• Mainstream Products (wirebond, leadframe, power, MEMS) — $1,152.4M, 17.2%.
• Advanced SiP modules ~$3,080M sit mostly inside Advanced.

Customers / contract structure. Top-10 customers = 72% of net sales; Apple ~29.8% and Qualcomm ~11.1% in FY2025. Critically, there is no material backlog and no binding customer commitments — quarterly revenue depends on that quarter's demand, customers "double-book," cancel and delay, and historically there is persistent downward price pressure on ASPs. This is the opposite of a take-or-pay annuity: it is a high-fixed-cost, demand-taking, price-eroding service business whose profitability is a leveraged bet on utilization.

Competitors: ASE Technology, JCET, Powertech (PTI); plus contract foundries (TSMC's own InFO/CoWoS), EMS providers, and IDMs' in-house back-end.

Supply Chain

Map: silicon wafers (consigned, customer-owned) → Amkor back-end (package + test) → back to fabless/IDM/foundry → into the end device.

• Upstream inputs Amkor does buy: laminate substrates, leadframes, bonding wire (gold/copper), capacitors — "from a limited group of suppliers," sourced worldwide. Substrates are the named chokepoint of the whole advanced-packaging industry (ABF substrate supply has been the binding constraint on 2.5D for years). Equipment: wire bonders, die bonders, plus long-lead wafer-bumping tools (sputter/plating/reflow) for 2.5D/HDFO and testers/handlers/probers (test is more capital-intensive and longer-lead than assembly).
• The company in the middle: manufacturing center of gravity is Korea (PP&E $2,098.2M, ~54% of the $3,870.8M global PP&E base — the R&D Center of Excellence is also Korea), then Vietnam $517.8M (opened 2024, ramping), Taiwan $352.7M, China $255.9M, Philippines, Portugal, Japan; U.S. PP&E only $105.9M (Arizona just breaking ground).
• Downstream named buyers: Apple (anchor, ~30%), Qualcomm (~11%); end markets = Communications 46%, Computing 20%, Auto/Industrial 19%, Consumer 15%. By customer-HQ region: U.S. $4,397.1M (66%), Japan $724.6M, Asia-Pac-ex-Japan $733.5M, EMEA $852.8M.
• The new node in the chain — Arizona/TSMC: Amkor and TSMC signed an MOU for Amkor to provide turnkey CoWoS + InFO advanced packaging/test in Peoria, AZ, packaging chips off TSMC's adjacent Phoenix fab for common customers (notably Apple). This inserts Amkor as the U.S.-soil back-end node for the AI/Apple supply chain — the strategic crux of the thesis.

Chokepoint read: Amkor is downstream of the silicon shortage but upstream of the device. Its single-source exposure is to substrate suppliers and to a handful of mega-customers; its leverage point is being one of only ~3 firms on earth that can run leading-edge 2.5D/CoWoS at volume — and the only large one expanding on U.S. soil.

Competitive Advantages (moats)

Real but narrow moats:

1. Scale + qualification switching costs. Customers run a "lengthy and rigorous qualification process that may take several months" before a package goes to volume. Once Amkor is qualified at a given site for a given package, displacing it mid-program is costly — the stickiness is at the socket/program level, not the corporate level (hence Apple's "temporary socket gap" could still cut ~7% of revenue in 2024–25; see Lens 8).
2. Advanced-packaging technology leadership. HDFO (SWIFT, S-Connect), 2.5D silicon interposers, fine-pitch copper-pillar flip chip, wafer-level fan-out, Silicon Photonics / Co-Packaged Optics, and emerging copper hybrid bonding. This is the part of the moat that is appreciating with the AI cycle.
3. Geographic footprint as a moat. 12 countries, ability to qualify the same product at multiple sites and absorb large rush orders — and uniquely a U.S.-headquartered OSAT building domestic capacity, which is a regulatory/geopolitical moat (CHIPS-Act-aligned, "China-for-China" insulation).

Bargaining power — weak, and that is the core flaw. Amkor explicitly concedes no backlog, short-term commitments, and structural ASP declines. Against Apple (~30%) and Qualcomm (~11%) it is the price-taker; against limited substrate suppliers it is somewhat captive. The "moat" shows up as durable #2-share and technology relevance, not as pricing power — which is exactly why a business this strategically important still earns only ~14% gross and ~8% ROE (Lens 5/7).

Segments

Amkor reports as a single operating segment (the CEO/CODM manages on consolidated net income), so there is no segment-level operating-income breakout — a genuine disclosure limitation. What is disclosed:

By product group (FY2025 / FY2024 / FY2023):

By end market (FY2025): Communications 46% (was 50% in '23 — declining share), Computing 20% (was 16% — the accelerating leg, AI/datacenter/networking), Auto/Industrial 19%, Consumer 15%. FY2025 growth by market: Computing +16%, Consumer +9%, Auto/Industrial +8%, Communications +1%.

The cause: the entire growth story is the advanced-packaging / computing mix-shift — Advanced went 77.4% → 82.8% of sales in two years while Mainstream (commodity wirebond) eroded. Q1-2026 confirms the acceleration: Communications +42% YoY (premium-tier smartphone content), Auto/Industrial +28%, Computing +19% (datacenter). The mix-up is revenue-accretive but gross-margin-dilutive in the near term because Advanced carries higher material content (substrates) — management says so explicitly.

---

Phase B — Measure performance

Earnings Result (latest print: Q1 2026, period ended 2026-03-31)

A clean, large beat that the market sold anyway.

• Beat magnitude: $0.33 GAAP vs ~$0.24 consensus = ~38% EPS beat; revenue beat too.
• Drivers: growth across all end markets; communications +42% and computing +19% YoY; gross margin lift from higher utilization, partly offset by richer (higher-material-content) mix.
• Guidance (from the call, not the 10-Q): Q2-2026 revenue $1.75–1.85B (+7% q/q), GM 14.5–15.5%, net income $105–130M, EPS $0.42–0.52 — supported by stronger-than-seasonal iOS ramp and the initial ramp of a new HDFO data-center CPU program.
• Balance sheet (FY2025 year-end): cash + ST investments $1,991.4M ($1,503.6M offshore); total debt $1,445.2M ($162.4M current); total assets $8,136.3M; net cash position. Operating cash flow FY2025 $1,095.6M; FCF $307.9M (down from $359.3M FY24 and $533.6M FY23 — the free-cash trend is deteriorating as capexCapital expenditureMoney spent on long-lived things — buildings, machines, servers — rather than on running costs. climbs).
• Market reaction — the tell: despite the ~38% beat, AMKR fell ~8.6% after-hours. The market is not arguing with the print; it is repricing the $2.5–3.0B 2026 capex weight, a softer-than-hoped 2H communications seasonal, and memory/advanced-silicon supply constraints. Read: expectations, not execution, are the binding constraint.

Earnings Calls (sentiment trend)

No transcripts are on the local shelf (transcripts/ empty); summary is ``. Tone trajectory across the last several quarters has inflected from defensive to offensive:

• 2024 → mid-2025: apologetic/defensive — managing through the Apple "socket gap" (temporary loss of an iOS socket), Q1-25 guided to a ~7% YoY decline, Vietnam start-up drag on margin.
• Q1-2026 call: confidently bullish — "record first quarter," "advanced packaging portfolio for AI expected to triple in 2026," 2H-weighted, new HDFO datacenter CPU program ramping, Arizona/TSMC framed as the multi-year story.
• May-2026 investor day: management put hard long-range stakes in the ground — 2028: ~$9B revenue, 17.5% GM; 2030: >$11B revenue, >22% GM, EPS >$5. The recurring new phrases are "AI advanced packaging," "HDFO/2.5D," "Arizona," "TSMC collaboration"; the thing they stopped emphasizing is the Apple-concentration caveat (still in the 10-KForm 10-KA company’s audited annual report to the US regulator. The most complete thing it publishes., quieter on the call).

Caution: the gap between management's confident multi-year targets and the stock's post-beat selloff is itself the signal — the sell-side is not yet underwriting the 2030 EPS bridge.

Comps

Peer table — AMKR vs the OSAT cohort. Multiples are ``, dated, and approximate; treat as directional, not precise.

• 5-yr avg ROE: n/a precisely; FY2025 ROE ~8.4%. This is a structurally low-return business — capital-intensive, low-margin, price-eroding.
• Read: AMKR trades at the top of the OSAT P/E band (~38x fwd, vs JCET ~30x and PTI ~15x; only ASE ~42x is comparable) and a premium EV/EBITDA (~12–14x vs peers 9–12x). The premium is the U.S.-soil + TSMC/Arizona + AI-packaging optionality — i.e., you pay for the 2030 story, not the ~$2 of 2026 EPS. On near-term earnings AMKR is the most expensive way to own OSAT.

Stock-Price Catalysts (>5% moves, ~last 2 years)

Pattern is dominated by (a) Apple/iOS socket news and (b) the AI-packaging / Arizona narrative, with valuation as the amplifier. All ``.

• Jul 2024 → mid-2025: ~‑60% drawdown. Weak end-markets + temporary loss of an Apple socket ("socket gap"); Q1-25 guided down ~7% YoY. Confirms the single-customer reaction function.
• 2H-2025 → Jun-2026: ~+277–317% recovery to all-time high $93.55 (2026-06-22). Driven by the TSMC Arizona MOU, the AI/CoWoS-overflow chokepoint narrative, and the May-2026 >$5-EPS-by-2030 investor-day frame.
• Apr-2026: ‑~8.6% after a ~38% Q1 EPS beat — repriced on capex weight + 2H-seasonal caution.
• Recent: ~$81.48 on 2026-06-29, ~13% off the $93.55 ATH.

What the market reacts to: Apple order cadence (concentration), and incremental data points on the Arizona/AI ramp. It does not trade on the mainstream/wirebond book at all. The reaction function is "AI-supply-chain proxy with an Apple beta," which cuts both ways.

---

Phase C — Judge people & books

Management

• CEO transition (key fact): founder-era CEO Giel Rutten retired Dec 31, 2025; Kevin Engel (joined Amkor 2004; EVP Business Units 2023–25; COO Feb-2025) became CEO Jan-2026, with Rutten advising through Mar-2026. Engel is an internal operations lifer, not a founder or outside change-agent — continuity, not disruption. Comp: $900k base, 125% target bonus, $5.0M annual LTI + $1.0M retention RSUs.
• CFO: Megan Faust — presented the 2030 framework (>$11B/>22% GM/>$5 EPS).
• Ownership / control — the governance keystone: the Kim family controls ~52% of shares. Founder James J. Kim (founded Amkor 1968) moved to Chairman Emeritus Oct-2024; Susan Y. Kim is Chairman. Amkor is effectively a family-controlled company — the 10-K names "certain stockholders' ability to determine the outcome of matters requiring stockholder approval" as a discrete risk. (Color: Susan Kim was reported in talks for a Philadelphia Eagles ownership stake — a signal that family wealth/attention is diversifying outside the company.)
• Capital allocation: policy = return 40–50% of cumulative FCF via dividends + buybacks (since 2022). In practice FY2025 was dividends only — $81.9M paid, regular quarterly $0.08352 (+1%); no buyback executed; the 2024 special dividend's absence makes the headline payout look "‑54% YoY". ROE ~8.4% — capital is being plowed into a low-return back-end at scale; the Arizona bet is the capital-allocation decision that defines the next 5 years.
• Skin in the game / red flags: ~52% family ownership = enormous alignment but entrenchment risk (minorities can't force change). No 10b5-1Rule 10b5-1 planA schedule an insider sets in advance saying when their shares will be sold, so the sales cannot be timed on private news. A sale made OUTSIDE such a plan is a live decision, which is why it carries more information. plan changes in Q4-2025; no related-party flags beyond ordinary course. Archetype: founder-controlled, professionally-managed operator transitioning to a second-generation/insider-CEO regime — implies steady execution but limited appetite for the kind of pricing aggression or radical capital discipline an activist would demand.

Forensic Red Flags

Accounting posture is conservative and clean — this is not a forensic short.

• Auditor: PwC; ICFR effective; no disagreements with accountants; no error corrections/restatements.
• Revenue recognition: over-time, cost-input method on consigned wafers — low channel-stuffing risk (no inventory ownership, ~2–3 week cycle, services billed on completion). No SBC games — Amkor reports GAAP EPS (no aggressive non-GAAP bridge); equity-comp overhang is tiny (3.4M options/RSUs out of ~247M shares; weighted exercise $10.10).
• Cash vs earnings: OCF $1,095.6M comfortably exceeds net income $376.1M (D&A $642.0M is the bridge — capital-intensity, not earnings quality, is the story). Deferred-tax valuation allowance is shrinking ($114.8M → $107.1M → $79.4M FY23–25) — a positive signal of improving realizability.
• Watch items (not flags): (1) non-recourse receivables factoring — $154.4M sold in FY2025 (flatters DSO/working capital; modest and stable); (2) $1,152.4M off-balance-sheet purchase obligations ($1,084.8M due within 12 months) — real near-term cash claims tied to the capex ramp; (3) conditional reduced tax rates in Korea/Singapore/Vietnam — ETR was 15.4% FY2025 and will rise as these expire; (4) ~$1.5B of cash is offshore — repatriation would cost ~$43M.
• CHIPS / OBBBA: $407M CHIPS direct award signed Dec-2024, $0 received to date (milestone-gated, clawback risk); ITC stepped 25%→35% under OBBBA for property in service after 2025. The subsidy is upside, not yet cash.

Regulatory findings (required sub-section).

• SEC Litigation Releases / AAERs: None. Verified via SEC EDGAR EFTS (LR + AAER) search for "Amkor Technology" since 2021-06-29 — 0 findings.
• Non-SEC enforcement (web): no material FTC/DOJ/FDA/CFPB consent decrees, settlements, or penalties surfaced for Amkor.
• 10-K Item 3 / Note 17 (Legal Proceedings): ordinary-course only — "we believe the ultimate outcome of these claims and proceedings, individually and in the aggregate, will not have a material adverse impact". The only notable contingency is benign: the Nanium insolvency receipt ($72.8M received, $40.4M remitted to selling shareholders per the 2017 acquisition terms).
• Conclusion: No material regulatory or legal findings — verified via SEC EDGAR EFTS (LR, AAER), web search, and 10-K Item 3/Note 17 as of 2026-06-29.

---

Phase D — Project & stress-test

Forward Projection (FY2026 / FY2027 / FY2028 EPS)

Built bottom-up from FY2025 actuals ($6,708.0M rev, $1.50 diluted EPS) + Q1-2026 actual + Q2-2026 guide + management's framework. Output is ``; every input labeled. No our model create logged (per --watchlist rule — breadth mode does not commit Brier forecasts).

Anchors: FY2025 EPS $1.50. H1-2026 run-rate: Q1 $1.685B actual + Q2 ~$1.80B guide-mid = $3.485B. Management 2H-weighted (AI advanced packaging "triples in 2026"). ~249M diluted shares.

Management's own bridge (for calibration, ``): 2028 ~$9B / 17.5% GM; 2030 >$11B / >22% GM / EPS >$5. My base path runs a notch below management on margin (OSAT ASP erosion + Arizona start-up drag are real headwinds to a 22% gross margin a business that has lived at 14–15% for years). The single most important non-EPS number: FY2026 capex $2.5–3.0B vs OCF ~$1.1–1.3B ⇒ free cash flow of roughly ‑$1.3B to ‑$2.0B. Amkor funds the AI/US bet by spending years of free cash — the thesis is a multi-year FCFFree cash flowCash left after paying to run and maintain the business. Unlike profit, it is hard to flatter with accounting choices. trough for a 2028+ payoff.

Bull vs Bear

Bull case. Amkor is the indispensable, U.S.-domiciled back-end of the AI supply chain. 2.5D/CoWoS is the named bottleneck for AI GPUs (NIST flagged it); TSMC's own CoWoS runs ~120k wpm of a ~200k-wpm end-2026 market, leaving ~80k wpm of overflow that ASE and Amkor are the only two firms able to absorb. The TSMC Arizona MOU makes Amkor the on-shore packaging partner for TSMC-Phoenix wafers feeding Apple/AMD — a structurally advantaged, CHIPS-blessed position no competitor can replicate on U.S. soil. Advanced-packaging-for-AI triples in 2026; the Advanced mix is already 83% and climbing; management targets >$11B rev / >22% GM / >$5 EPS by 2030. If even 70% of that lands, today's ~$81 is cheap on out-year earnings. Conservative balance sheet (net cash), real subsidies (CHIPS $407M + 35% ITC) de-risk the capex.

Bear case (2–3 ways it permanently impairs or de-rates).

1. Customer concentration → episodic revenue cliffs. Apple ~30% + Qualcomm ~11% = ~41% in two accounts; the 2024 Apple "socket gap" already produced a ~60% drawdown. With no backlog and no commitments, a single socket loss or Apple insourcing/redesign re-rates the stock overnight.
2. The FCF trough is the valuation trap. ~$2.5–3.0B capex for 2–3 years drives deeply negative free cash flow, at the same time the company trades at ~38x forward earnings / ~14x EV-EBITDA. You pay a growth multiple for a business about to consume cash, with Arizona revenue not meaningful until 2028.
3. Structural margin ceiling. OSAT is a price-eroding, high-fixed-cost, ~14% gross-margin business with 8.4% ROE. The 22%-GM-by-2030 target requires a step-change in mix and pricing power Amkor has never demonstrated; meanwhile near-term GM is diluted by the higher-material-content Advanced mix and Vietnam/Arizona start-up costs.

Pre-mortem (18 months out, thesis broke): It's late 2027. Apple pulled a socket forward to a rival or in-house; the AI "triple" pulled in 2026 demand and 2027 decelerated; Arizona is over-budget and milestone-delayed (CHIPS clawback chatter); FCF has been negative for ~8 of the last 10 quarters; the stock de-rated from ~38x to a normal OSAT ~15x and halved. Nothing fraudulent happened — the multiple simply normalized onto a low-return business.

Are multiples too high? On near-term numbers, yes — ~38x forward P/E and ~14x EV/EBITDA for 8% ROE and negative FCF is a narrative multiple, and the mean sell-side PT ($75–82) sits at or below spot ~$81. The market is already paying for the 2030 bridge.

Contrarian view (what the market refuses to see): The bull and bear are arguing the wrong axis. The crowd treats AMKR as a binary "AI-packaging chokepoint" call; the real variable is Apple, not AI. The AI/datacenter book (~20% of sales, the overflow piece) is the option; the ~46% communications/Apple book is the bond. The stock will be made or broken by iPhone socket content and TSMC-Arizona Apple volumes — and that is a far more concentrated, less "secular" bet than the chokepoint story implies.

Devil's Advocate (short-seller)

Dismantling the bull case.

• Where revenue is concentrated: ~41% in Apple+Qualcomm; 46% in communications; 66% by customer-HQ in the U.S.. Shift any one Apple socket and the model breaks — and Amkor has zero contractual protection (no backlog, customers "double-book" and cancel).
• Why the moat is weaker than bulls think: Amkor is the overflow vendor. The crown-jewel AI GPUs are packaged by TSMC itself (CoWoS); Amkor and ASE split the leftovers (~80k of ~200k wpm). Being the spillover bucket is not a moat — it's a margin-taking, share-of-the-residual position that compresses the instant TSMC's own capacity catches up (capacity is being added aggressively industry-wide).
• Most dangerous competitor bulls underestimate: TSMC's in-house InFO/CoWoS (vertically integrates the very service Amkor sells) and JCET/Tongfu in a "China-for-China" stack the 10-K explicitly flags as intensifying competition. ASE is the bigger, higher-scale #1.
• Worst capital-allocation reality: plowing years of free cash into a low-return ($0.084 ROE) back-end, funded into a multi-quarter FCF-negative trough, while a ~52%-family-controlled board can't be forced to redirect. The buyback half of the "40–50% of FCF" policy went unexecuted in FY2025.
• Assumptions that must hold for ~$81: (a) AI advanced packaging really triples in 2026 and doesn't air-pocket in 2027; (b) Apple stays ~30% and Arizona-Apple volumes ramp on schedule; (c) GM climbs from 14% toward 17.5%+ despite ASP erosion; (d) Arizona delivers on-budget, on-time, with CHIPS money flowing. All four must broadly hold.
• If growth disappoints 20–30%: at a normalized ~15x OSAT multiple on a haircut ~$1.5 EPS, fair value is ~$22–25 — i.e., the stock has ~70% downside to peer-normal math if the narrative premium evaporates. That asymmetry — modest upside if the 2030 dream lands, severe downside on de-rating — is the short's entire case.
• Single permanent-impairment scenario: Apple structurally in-sources advanced packaging or moves a flagship socket to ASE/TSMC-direct. Plausibility: low-to-moderate near-term (qualification stickiness), but non-trivial over 3–5 years and catastrophic if it hits.

Management Questions (ordered by information value)

1. Apple is ~30% of revenue with no backlog or commitment — what contractual or technological lock-in (Arizona co-investment, multi-year capacity reservation, JDA) now protects that socket through the next two iPhone cycles, and how much of the 2024 "socket gap" content has actually returned?
2. Walk us through the 2026 free-cash-flow bridge: with $2.5–3.0B capex vs ~$1.2B OCF, exactly how negative is FCF, how is it funded (cash, debt, CHIPS), and at what quarter does FCF inflect positive again?
3. Your 2030 target is >22% gross margin vs ~14% today in a business with chronic ASP erosion — what specific mix, pricing, and utilization levers close an 8-point gap, and what's the 2028 milestone that proves you're on track?
4. On the TSMC Arizona MOU — is it binding capacity/volume or an intent framework? What committed wafer volumes, which customers (Apple/AMD), and what take-or-pay terms underpin the $7B campus?
5. How much of the "AI advanced packaging triples in 2026" is new demand vs. a 2026 pull-forward that risks a 2027 air-pocket? What does the 2027 AI-packaging backlog/pipeline look like today?
6. You are the CoWoS/2.5D overflow vendor to TSMC — what happens to your AI-packaging revenue and pricing when TSMC's own CoWoS capacity (now ~120k wpm) catches up with demand?
7. CHIPS $407M is milestone-gated with $0 received — what are the precise construction/production milestones, the clawback triggers, and your base-case timing for first cash?
8. Arizona reaches meaningful production only in 2028 — what are the start-up margin drag, ramp yield risk, and U.S.-vs-Asia cost differential, and how dilutiveDilutionIssuing new shares, so each existing share owns a smaller slice of the same company. is it to consolidated GM in 2026–27?
9. Substrate supply has been the industry chokepoint — how secured is your ABF/laminate substrate supply for the 2.5D/HDFO ramp, and is it single-sourced?
10. The "40–50% of cumulative FCF" return policy produced dividends only in FY2025, no buyback — during a multi-year FCF-negative window, is the buyback effectively suspended, and does the policy even bind when FCF is negative?
11. With CEO transition to an internal operator (Engel) and a ~52% family-controlled board, what changes — if anything — in strategy, capital discipline, or willingness to push pricing?
12. Conditional tax rates in Korea/Singapore/Vietnam are expiring — what is the glide path of the effective tax rate through 2028, and how much EPS does normalization cost?
13. Korea is ~54% of PP&E and the R&D Center of Excellence — how concentrated is single-site/geopolitical risk, and does Arizona/Vietnam materially de-risk it this decade?
14. Mainstream (wirebond) revenue is in secular decline ($1.47B→$1.15B in two years) — is this a managed runoff, and at what point does it stop being a drag on consolidated growth?
15. What is the mid-cycle ROIC you underwrite on the $7B Arizona investment, and how does it compare to your ~8% corporate ROE — i.e., is this value-creative or strategically-necessary-but-dilutive?

---