Amkor Technology, Inc. (NASDAQ: AMKR) — A Cyclical, Capital-Hungry Packager Repriced as an AI-Packaging Pure-Play

Independent Equity Research — Initiation Date & price as noted. · Date: 2026-06-19 · Price (2026-06-18 close): $90.46 · Market cap: ~$22.4B · Net cash: ~$0.5B Sector: Information Technology — Semiconductors (OSAT: outsourced semiconductor assembly & test) · FYE: December 31 · CIK: 0001047127

⚡ Claude’s Take

This is the author’s own independent opinion and general information only — not investment advice and not a recommendation to buy or sell any security. The analysis in sections 1–15 below is deliberately position-free and carries no price target; this block is the single exception.

Verdict: AVOID at $90 for new capital — but NOT a short. Accumulate-on-weakness only toward the low-$40s-to-$30s. Conviction: medium-high on the “too rich here,” medium on the entry zone.

Amkor is a genuinely competent, scaled, technologically-credible #2 OSAT — and the market is pricing it as something it is not: an AI-packaging pure-play with foundry-like economics. The numbers say otherwise. This is a capital-intensive, cyclical, price-taking back-end service business whose returns are dictated by the semiconductor cycle, ~30% single-customer (Apple) concentration, factory utilization, and a structurally deflationary average-selling-price environment. Through-cycle ROIC is ~10–11% and sits at ~7% at the current trough — below its own cost of capital for a beta-2.26 stock. Yet the equity trades at the richest valuation in its entire public history: the AZI own-history percentile is the 99.98th for both price/sales and price/book (composite 97.86th), ~3.1x EV/sales and ~18x EV/EBITDA versus a historical norm nearer 1x sales / 6–9x EBITDA. The denominator is depressed (FY25 diluted EPS $1.51 is half the $3.11 cyclical peak of 2022) and the multiple is at a record — the textbook setup of peak multiple on trough earnings. The June-16 TSMC 10-year Arizona packaging deal is a real, positive datapoint, but it is the halo, not the cash flow: the highest-rent AI-accelerator packaging (CoWoS-class logic-on-logic) is captured by TSMC in-house; Amkor’s AI/advanced revenue is only ~20% of the total and management twice declined to give a clean AI-datacenter number. The whole bull case rests on out-year optionality — a 2028-production Arizona campus that doesn’t reach a ~$1B run-rate until 2029–30 and dilutes operating margin 1–2% in 2027 first — financed by a capex budget tripling to $2.5–3.0B that makes Amkor FCF-negative in 2026.

The framing is a momentum/cyclical-narrative crowding, not a value or quality setup: factor loadings show Momentum +0.55, “Chip Sector Superstars” +0.89, LowVolatility −0.89 (strongly anti-defensive), Value absent — and the stock now trades with semicap equipment names (AMAT, LRCX, FormFactor), riding the AI-capex tape, having risen ~6x off its April-2025 trough. The controlling Kim family (49.4%, single share class) and every named insider are net sellers — a 10-million-share family secondary at $48.75 in February 2026 — with zero open-market buys, and the comp plan has no return-on-capital hurdle while capex triples. The stock even trades above the ~$69 sell-side consensus price forecast. I’d want mid-cycle EPS power (~$2.50–3.50 in a good up-cycle) at a defensible 13–16x — i.e., the low-$40s, scaling harder into the $30s (roughly where it traded as recently as late 2025 / mid-2024) — where the Arizona/AI optionality comes closer to free. It is not a short (near net-cash, real up-cycle operating leverage, a genuine multi-year advanced-packaging option, ~50% idiosyncratic vol and a high short-squeeze beta that has already torched skeptics). Flips bullish if AI-datacenter advanced packaging becomes a disclosed, durable, >30%-of-revenue driver with gross margin sustained in the high-teens through the next mobile down-cycle. Flips more bearish on a mobile/Apple air-pocket or utilization rollover that drops EPS back toward $1.50 while the multiple normalizes. Tag: trough earnings, record multiple — the back-end commodity wearing TSMC’s halo.

📈 Stock Price Action — Five-Year Event Map

Amkor has completed a violent five-year round-trip and then some: from a 2022 cyclical low near $14.55, to a 2024 false-dawn high of $43, down to a $14.69 tariff-crash trough in April 2025, and then a near-vertical ~6x sprint to $90.46 (52-week range ~$20–$90 close; intraday high $96.68 on 2026-06-16, ~6% off that high). The stock now sits at the very top of its own cycle and, on multiples, the top of its entire public history. Price moves below are FACT (AZI daily CSV, split/dividend-adjusted); attributed drivers are INTERPRETATION.

Cycle narrative. (1) Amkor bottomed with the 2022 chip downturn as mobile and consumer demand corrected. (2) A ~3x recovery into mid-2024 rode the first wave of AI/advanced-packaging enthusiasm and a broad semis re-rating. (3) It then gave nearly all of it back: a 2024–25 mobile/automotive inventory correction was capped by the April-2025 tariff shock, which drove the stock to a marginally new multi-year low of $14.69 — a reminder of how violently a 2.26-beta, mobile-concentrated cyclical trades on macro. (4) From that trough the stock tripled into December 2025 on cyclical recovery and two consecutive earnings beats (the print-day pops on 2025-07-29 and 2025-11-03 are visible as ~+16–17% idiosyncratic moves). (5–6) 2026 added another leg as the FY26 guide (computing +20%+, advanced packaging “tripling”) and the Arizona/CHIPS story converted the name into an AI-capex momentum vehicle; the Q1’26 beat (rev +27% YoY) confirmed up-cycle operating leverage. (7) The June-16 TSMC 10-year Arizona deal was the exclamation point, lifting the stock intraday to a $96.68 52-week high before settling at $90.46. The price is now a near-perfect mirror of the narrative; the financials (section 6) are still mobile-led, low-double-digit-gross-margin, and capex-consumed.

1. Executive Summary

Amkor Technology is the world’s largest U.S.-headquartered, and global #2, OSAT — an outsourced provider of semiconductor assembly (packaging) and test services. It takes finished wafers (generally consigned, so it carries little customer-silicon inventory), packages the die into the forms electronics manufacturers use, tests them, and drop-ships. FY2025 revenue was $6.71B with GAAP gross margin of 14.0%, operating margin of 7.0%, net margin of 5.6%, and diluted EPS of $1.51. Advanced products are 82.8% of sales; system-in-package (SiP, largely mobile) alone is ~46% of revenue. Communications (smartphones, ~80% Apple/Qualcomm content) is still the single largest end market at ~46%; computing/datacenter is ~20% and rising only gradually.

The investment debate is a valuation-versus-quality tension at an extreme. On the quality side, this is a structurally difficult business: capital-intensive (capex ~13.5% of sales; gross PP&E $11.4B), cyclical (no material backlog; high fixed-cost operating leverage that cuts both ways), exposed to a structurally declining ASP environment, dependent on one customer (Apple, ~30%) that multi-sources, and earning a through-cycle ROIC of only ~10–11% (~7% at the current trough) — a return that does not durably exceed the cost of capital for a beta-2.26 equity. There is no durable Greenwald moat: customer captivity runs toward the customer, not Amkor. On the valuation side, the equity is at the richest multiple in its public history — P/S and P/B both at the 99.98th own-history percentile, ~3.1x EV/sales and ~18x EV/EBITDA, ~51x trailing P/E on trough earnings — after a ~6x run off the April-2025 low.

The bull thesis is real but mostly future optionality: AI/advanced packaging (Amkor’s HDFO/2.5D platforms — SWIFT and S-Connect, the CoWoS-R/-L analogues) is guided to “nearly triple” in 2026, a new datacenter-CPU program inflects in H2’26, and the $7B Arizona campus, built adjacent to TSMC’s Arizona fabs with ~$2.8B of government support ($407M CHIPS grant + a 35% investment tax credit) and validated by a June-2026 TSMC 10-year procurement agreement, positions Amkor for a US-onshored advanced-packaging franchise. But Arizona does not produce revenue until 2028, reaches only ~$1B run-rate (~10%+ of revenue) by 2029–30, and dilutes operating margin 1–2% in 2027 before any payoff — financed by a capex budget tripling to $2.5–3.0B that makes Amkor FCF-negative in 2026.

Capital allocation and governance are the weakest pillars. Amkor is a controlled company (Kim family ~49.4%, single share class, family Chairman), returns capital via dividends only (no buyback program), and runs a comp plan with no return-on-capital hurdle while it triples capex into a low-ROIC business at a cycle/narrative peak — a textbook Marathon capital-cycle warning. Every named insider is a net seller, including a 10-million-share family secondary at $48.75 in February 2026 and zero open-market purchases. The market is underwriting a margin and mix transformation that the income statement has not yet delivered, at a price even sell-side consensus (Hold, ~$69 average forecast) sits below.

2. Business Overview

What Amkor does. Amkor sells services, not chips. After a customer’s wafers are fabricated (by a foundry like TSMC or by an IDM’s own fab), Amkor performs the back-end steps of the semiconductor value chain: wafer bumping, wafer probe, back-grind, package design, assembly/packaging, burn-in, system-level and final test, and drop-shipment. Wafers are generally consigned — “title and risk of loss remains with the customer” — so Amkor does not own the silicon and carries relatively little inventory ($438M at YE25 on $6.7B revenue). It offers turnkey solutions (design through test) as well as discrete services. Revenue is recognized as services are performed; there is essentially no material backlog and customers “do not commit to purchase any significant amount,” which can “reduce, cancel or delay” — a structural feature that makes revenue visibility short and the cost base (heavily fixed) the swing variable for margins.

Product mix. Amkor reports a single operating segment but discloses two product groupings:

• Advanced products — $5,556M, 82.8% of FY25 sales (rising from 77.4% in 2023): flip-chip (FC CSP, FCBGA), wafer-level CSP and fan-out (WLFO), high-density fan-out (HDFO: SWIFT/S-Connect), 2.5D silicon-interposer integration, advanced SiP, memory, and related test.
• Mainstream products — $1,152M, 17.2% (falling from 22.6%): wirebond, leadframe, power-device packaging and related test.
• Advanced SiP (system-in-package) alone is ~$3.08B — roughly 46% of total revenue — and is overwhelmingly mobile (RF front-end, connectivity modules for premium smartphones).

End markets (FY2025; Q1-2026 in parentheses):

Despite the AI framing, mobile is still ~44–46% of revenue and is the dominant business. Computing/datacenter — the home of the AI story — is only ~20–21% and has risen only modestly over three years.

Customers. Concentration is extreme: the top-10 customers are 72% of net sales; Apple ≈ 29.8% and Qualcomm ≈ 11.1% (both named in the 10-K), together ~41%. Apple’s share has risen (27.7% → 30.8% → 29.8% over 2023–25). By customer headquarters, US customers are 65.6% of revenue (Apple, Qualcomm, Nvidia, AMD, Broadcom), even though ~95% of the work is performed in Asia.

Geographic footprint. Manufacturing is ~12.6M sq ft across Korea (the dominant hub — net PP&E $2.1B, ~56% of foreign PP&E), Japan, China, Vietnam (Bac Ninh, opened 2024, ramping), Philippines, Taiwan, Portugal, and Malaysia — plus the new Arizona campus under construction. 30,800 employees, ~95% in Asia-Pacific.

Revenue model & recurrence. Revenue is recurring in the sense that Amkor packages successive generations of long-lived sockets (it co-develops packages with lead customers, qualifies them, and then runs volume) — but it is not contractually recurring: no backlog, short visibility, and ASPs that decline over time. The economics are those of a high-fixed-cost services business: profitability is a function of utilization × mix × price, with utilization the largest near-term swing factor.

Verdict: A scaled, technologically broad back-end services provider whose revenue is dominated by mobile (Apple) with a growing datacenter tail. The model is service-based, low-inventory, and high-fixed-cost — recurring in practice but uncommitted in contract, and structurally exposed to price deflation and the chip cycle.

3. Industry Dynamics

Structure of the OSAT industry. The outsourced assembly-and-test market is the back end of the semiconductor value chain. It is fragmented, competitive, and structurally low-margin. The global #1 is ASE Technology (including SPIL); Amkor is #2; JCET (China) #3; with Powertech, Tongfu (TFME), and a growing field of state-supported Chinese OSATs filling out the field. OSATs compete with (a) each other, (b) foundries that increasingly offer their own advanced packaging (TSMC’s CoWoS/InFO, Samsung, Intel Foundry), © EMS/contract manufacturers offering some packaging, and (d) IDM customers’ own in-house back-end, which they continually benchmark against outsourcing.

The capital cycle (Marathon lens). OSAT exhibits every feature of a structurally unattractive, capital-cycle-driven industry: high fixed costs, capacity built ahead of demand and “without firm customer commitments,” no backlog, ASP deflation (“prices…have generally declined over time…we expect downward pressure…to continue”), and periodic capacity gluts. The current moment is a capacity-investment surge: Amkor’s 2026 capex tripling to $2.5–3.0B, ASE and the Chinese OSATs adding aggressively (including state-subsidized China-for-China capacity), and government subsidies (US CHIPS Act, and equivalents in China, Japan, and the EU) distorting the cycle by pulling in extra capacity. Heavy capacity additions into a cyclical, price-deflationary service market are a supply-side warning, not a tailwind — they tend to depress through-cycle returns even when end demand grows.

The one genuine structural positive: advanced packaging is the new performance bottleneck. As transistor scaling slows, heterogeneous integration (chiplets, 2.5D/3D, HBM-on-logic, co-packaged optics) has become the principal lever for system-level performance, especially for AI/HPC. This raises the value of advanced packaging and its share of the BOM. The catch for Amkor: the highest-rent advanced packaging — CoWoS-class logic-on-logic for AI accelerators — is dominated by TSMC in-house (and Samsung/Intel), not by merchant OSATs. Amkor participates via HDFO (SWIFT/S-Connect), 2.5D interposers, advanced FCBGA, HBM-adjacent test, and silicon photonics/CPO — real and growing, but generally the less contested, lower-rent tiers, alongside whatever overflow the foundries outsource.

Geopolitics & onshoring. US export controls (BIS) restrict sales of certain advanced packaging to Chinese customers — a headwind that can route business to non-US competitors — while simultaneously the CHIPS Act subsidizes US-located capacity. The Amkor–TSMC Arizona collaboration is the marquee example of supply-chain regionalization: a US “full supply chain from advanced silicon to tested packaged devices.” This is genuine optionality, but it is also capacity built where labor and construction costs are far higher than Asia, requiring subsidies and ITCs to be viable and diluting margins during ramp.

Verdict: structurally unattractive industry with one attractive sub-trend. OSAT is fragmented, capital-intensive, price-deflationary, customer-power-dominated, and now in a subsidized capacity-add phase — a poor industry by Greenwald/Marathon criteria. The advanced-packaging-as-bottleneck trend is real and is the only durable positive, but its richest economics accrue to the foundries, not to the merchant back end.

4. Competitive Position

Does Amkor have a moat? No durable one. Run through Greenwald’s three genuine advantage types:

• Supply / cost advantage — No. Amkor is mid-pack on cost: smaller-scale than ASE, and structurally higher-cost than the labor-cheap, state-supported Chinese OSATs (JCET, Tongfu). It lacks long-term contracts to pass through commodity inputs (gold, substrates, copper) and explicitly expects continued ASP decline. No cost moat.

• Demand-side captivity — Weak, and pointed the wrong way. There is some genuine stickiness: Amkor co-develops proprietary package/process technology with lead customers, and qualification of a package at a specific OSAT creates switching friction within a product generation. But this is overwhelmed by (a) ~30% revenue from one customer (Apple) that deliberately dual/triple-sources (ASE, JCET) and dictates price; (b) no backlog and uncommitted volume; and © the constant IDM in-sourcing benchmark. The captivity in this relationship is Apple’s hold on Amkor, not the reverse.

• Economies of scale + captivity — No. Scale exists ($3.96B net PP&E, 12.6M sq ft) but is not the largest (ASE is bigger), and OSAT scale does not create the local-monopoly captivity Greenwald requires — it is a multi-vendor market where the largest customers structurally refuse to single-source.

The financial proof of “no moat.” A moat must show up as durable, above-cost-of-capital returns and stable share. Amkor fails both tests:

• Returns: ROIC of 11.1% (2020), 17.2% (2021), 16.9% (2022, peak), then 7.3% / 6.6% / 6.8% (2023–25). Through-cycle ~10–11%; trough ~7%, below the cost of capital for a 2.26-beta name. Gross margin is a commodity-services 14%; operating margin ~7% (and even that was flattered in FY25 by a one-time Nanium insolvency cash receipt and an equipment-sale gain).
• Operating leverage cuts both ways: Q1-2025 operating margin 2.4% → Q1-2026 6.0% on a +27% revenue swing. That sensitivity is the signature of a cyclical price-taker, not a franchise with pricing power.

Versus competitors. ASE is larger and more diversified; JCET and the Chinese OSATs are cheaper and increasingly capable in the China-for-China market; TSMC/Samsung/Intel own the leading-edge AI packaging. Amkor’s genuine differentiators are (i) breadth of advanced-packaging technology and qualified capacity, (ii) a global footprint spanning Asia + Europe + (soon) the US that suits supply-chain diversification, and (iii) deep co-development relationships with marquee US fabless/IDM customers. These are real competitive competencies and explain why Amkor is the #2 player — but a competency that must be continually reinvested in (capex “in advance of revenue, without firm commitments”) is not a barrier to entry.

Verdict: a strong #2 in a weak industry, with competencies but no durable moat. Returns are cycle- and utilization-dependent and do not durably clear the cost of capital. The advanced-packaging technology lead is real but generation-resetting and capital-hungry, and the customer relationship confers captivity on the customer, not on Amkor.

5. Growth History and Forward Opportunities

History — flat-to-cyclical, not secular. Revenue: $5.05B (2020) → $6.14B (2021) → $7.09B (2022 peak) → $6.50B (2023) → $6.32B (2024) → $6.71B (2025). FY2025 revenue is still below the 2022 peak. Diluted EPS traced an even sharper cycle: $1.40 → $2.62 → $3.11 (2022 peak) → $1.46 → $1.43 → $1.51. The five-year revenue CAGR (2020–25) is ~5.8% — but that is entirely a 2020–22 up-cycle followed by a 2023–25 plateau; there is no evidence of secular, compounding growth independent of the chip cycle. Growth has been overwhelmingly organic (goodwill is a negligible $18M — Amkor builds, it does not buy).

Q1-2026 — strong cyclical recovery. Revenue $1.68B (+27.5% YoY), gross margin 14.2% (vs 11.9%), operating margin 6.0% (vs 2.4%), diluted EPS $0.33 (vs $0.09). The single biggest driver was Communications +42% YoY (premium iOS content/socket gains), with Auto & Industrial +28% and Computing +19%. This confirms the up-cycle operating leverage — and confirms that the near-term growth engine is still mobile, not AI.

Forward opportunities (the bull’s optionality):

1. AI / advanced datacenter packaging. Amkor’s 2.5D + HDFO (SWIFT ≈ CoWoS-R, S-Connect ≈ CoWoS-L) platforms are guided to “nearly triple” in 2026, with 5+ customers engaged on SWIFT/S-Connect and half-a-dozen+ on 2.5D. A new datacenter-CPU program inflects with “meaningful revenue from Q3’26” and ramps into 2027. This is real and accelerating — but AI/advanced was only ~20% of revenue exiting 2025 (and pure AI-datacenter is a subset of that), and management twice declined to disclose a clean AI-datacenter revenue number.
2. Arizona / US onshoring. A $7B two-phase campus adjacent to TSMC’s Arizona fabs; Phase 1 construction complete in 2027, production in 2028, scaling to ~$1B run-rate (~10%+ of revenue) by 2029–30. Backed by ~$2.8B government support ($407M CHIPS grant + 35% ITC) plus customer prepay/loading agreements, and validated by the June-2026 TSMC 10-year procurement agreement. This is the single largest piece of forward optionality — and the single largest near-term cost (capex + a 1–2% operating-margin dilution beginning 2027).
3. Advanced automotive (ADAS, in-cabin compute) — four consecutive quarters of sequential mainstream recovery, with advanced-auto setting records.
4. Silicon photonics / co-packaged optics — early-stage exposure to AI-cluster optical interconnect scaling.

Verdict: high-cyclicality, modest-secular growth, with genuine but back-end-loaded optionality. The growth that has happened is cyclical and mobile-led; the growth that would justify the multiple (AI/datacenter at scale, Arizona at run-rate) is a 2028–2030 story, gated by customer silicon/memory supply and Amkor’s own ramp, and carries a 2027 margin air-pocket en route. Quality of growth: mixed — real volume leverage on the upswing, but low-margin, capital-consuming, and not yet the high-rent AI mix the price implies.

6. Financial Quality

Margins — commodity-grade and cyclical. Gross margin has ranged 14–20% over six years (FY25: 14.0%, down from 14.8% in FY24); operating margin 7–13% (FY25: 7.0%, near the low end and flattered by one-timers); net margin 5.6%. EBITDA margin 16.5%. These are back-end-services economics, structurally below the foundry (TSMC ~50%+ gross) and fabless-design tiers of the value chain. The FY25 operating-margin increase (to 7.0% from 6.9%) despite a falling gross margin was driven primarily by a one-time Nanium insolvency cash receipt plus an equipment-sale gain — i.e., reported FY25 operating profit slightly overstates run-rate.

Capital intensity — the defining feature. Gross PP&E is $11.4B against $6.7B revenue; depreciation alone is $642M (~10% of revenue). Capex has run $553M–$908M annually (≈11–14% of sales) and is guided to $2.5–3.0B in 2026 (3x the 2025 level) for Arizona Phase 1 + advanced equipment. This is a business where nearly every dollar of operating cash flow is reinvested merely to stay competitive.

Free cash flow — thin and volatile, about to go negative. OCF has been a steady ~$1.0–1.3B, but after capex, FCF was $217M / $342M / $190M / $521M / $345M / $191M (2020–25) — a FY25 FCF margin of just 2.8%. With 2026 capex tripling to $2.5–3.0B against ~$1.1B OCF, Amkor will likely be FCF-negative in 2026 and is pre-funding the gap with debt and a convertible (see section 7).

Returns on capital — at/below cost of capital. ROIC ~7% at the trough, ~17% at the peak, ~10–11% through-cycle (section 4). Reported ROE looks high (15–50%+) but is flattered by leverage and a low book base relative to the depreciated asset base; ROIC is the honest read for a capital-intensive operator, and it does not durably clear the WACC of a high-beta cyclical.

Balance sheet — genuinely strong (today). Cash + ST investments $1.99B against total debt of ~$1.45B → net cash of ~$0.5B at YE25; current ratio 2.27; debt/EBITDA ~1.1x; goodwill negligible ($18M, organically built). This is a real strength and the main reason the name is “not a short.” However, the balance sheet will lever up through the Arizona build (the 2026 capex plan exceeds OCF), and ~$1.5B of cash is held offshore.

Quality of earnings — mostly clean, with several normalizing items:

• FY25: one-time Nanium insolvency receipt + equipment-sale gain inflated operating income; collapsing Singapore tax benefit ($0.13 EPS in 2024 → $0.01 in 2025 as Pillar Two/QDMTT bit) is a structural headwind.
• FY24: a depreciation useful-life extension on test equipment reduced depreciation ~$59M and added ~$0.20 to diluted EPS — inflating the FY24 base.
• SBC is negligible (~$20M, <1% of revenue) — a genuine positive; reported and “adjusted” earnings are close, and there is little SBC dilution.
• Effective tax rate is low (10–18%) given the foreign-domiciled (Korea-heavy) operations, but is structurally rising as global minimum-tax rules phase in.
• Cash conversion is sound (OCF consistently exceeds net income, as expected given heavy D&A).

Verdict: economics do NOT meaningfully improve with scale. Six years and ~$4.5B of cumulative capex have left gross margin roughly where it started (mid-teens) and FCF thin. Earnings quality is reasonable (low SBC, clean cash conversion, strong balance sheet) but several items flatter the recent base, and the business is entering its most cash-consumptive phase. This is a financially sound but low-return, capital-hungry operator — the opposite of a self-funding compounder.

7. Capital Allocation

Governance frame: a controlled company. The Kim family beneficially owns ~49.4% of common stock (122.7M shares; single share class, one vote per share). Susan Y. Kim is Chairman; founder James J. Kim’s family vehicles (915 Investments, Sujochil, KCP) hold the bulk. There is no dual-class structure — control is achieved through the economic stake itself, which is effectively unassailable given turnout. The board serves the controlling family first; minority shareholders have little governance leverage, and a takeover premium is structurally unlikely.

Leadership. An orderly internal succession: Giel Rutten retired as CEO effective 2025-12-31; 20-year insider Kevin Engel (ex-COO) became CEO 2026-01-01. Continuity, not a change-agent — and Engel’s personal stake (~11,645 shares) is de minimis.

The capex decision (the central capital-allocation question). Amkor is tripling capex to $2.5–3.0B to build the $7B Arizona campus and expand advanced-packaging/test capacity, into a business earning ~7% trough ROIC, at a cyclical/AI-narrative peak. The Marathon capital-cycle read is unambiguous: a hot narrative and recovering returns are pulling capital into a structurally low-return, capital-intensive industry — exactly the condition under which returns subsequently mean-revert. Management would argue the spend is strategically necessary to win advanced-packaging share alongside TSMC and to capture US-onshoring demand — but “necessary to defend/win share” is itself the hallmark of a business without durable pricing power. The CHIPS grant ($407M, milestone-conditional, none yet received) and the 35% ITC de-risk the cash cost of the Arizona build; they do not make the underlying economics attractive.

Incentives — misaligned for a capital-heavy business. The annual bonus pays on Operating Income; long-term PSUs pay on EPS and relative TSR. There is no ROIC, ROCE, or return-on-capital hurdle anywhere in the plan. For a company about to triple its asset base, paying on absolute operating income and EPS — both of which a multi-billion-dollar capacity build mechanically grows over time regardless of incremental returns — rewards exactly the asset-growth behavior the capital cycle punishes. FY25 CEO comp was $12.1M; pay ratio 656:1 (median employee $18,442, reflecting the Asian labor base). Say-on-pay passed at ~97.8% — but the ~49.4% family bloc votes the plan it designed, so this is not an independent endorsement.

Capital return — minimal and dividend-only. Amkor pays a small, growing regular dividend (FY25 $81.9M, ~$0.33/share, ~0.4% yield at today’s price; FY24 included a ~$100M special). There is no buyback program — the only repurchases are tax-withholding shares. With capex tripling, capital return is firmly subordinated to the build.

Financing — pre-funding the wave. Into the capex cycle Amkor has stacked liquidity: a $1B revolver (May-25), a $500M term loan (Jun-25), $500M 5.875% senior notes due 2033 (Sep-25, refinancing the 2027 notes), and — notably — $1.15B of 0.00% convertible senior notes due 2031 (May-26). The zero-coupon convert is cheap cash today but is latent equity dilution if the stock holds (a real consideration at a 99.98th-percentile valuation), and it signals management financing the build with paper while the multiple is rich.

Insider behavior — uniformly distributional. Across the recent Form 4 corpus there are zero open-market purchases (code P) — only grants, option/RSU settlements, tax-withholding, and 10b5-1 sales. Most tellingly, the Kim family vehicle 915 Investments sold 10.0 million shares at $48.75 in a February-2026 underwritten secondary, and a family GRAT gifted 725,000 shares in December 2025. The controlling family is a net seller into the rally; no insider has signaled conviction by buying.

Verdict: the weakest pillar. Competent operations and an orderly succession, but capital allocation is tilted toward asset growth at a cycle peak, incentivized by metrics (no ROIC hurdle) that reward exactly that, with minimal capital return, latent convert dilution, and a controlling family that only sells. This is not how value-accretive capital allocators behave at a record valuation.

8. Changes and Headwinds — Last Two Years

Strategic / structural changes:

• CEO transition (Rutten → Engel, Jan-2026) — orderly, internal.
• Arizona campus scaled up from an initial plan to a $7B two-phase project, relocated adjacent to TSMC’s Arizona fabs, with a 50-acre expansion option (Q3’25).
• TSMC 10-year advanced-packaging agreement (June-16-2026) — TSMC to procure advanced packaging and test from Amkor, expanding Arizona capacity; financial terms undisclosed. The marquee strategic event and the proximate cause of the latest price spike.
• CHIPS Act direct-funding award (up to $407M, Dec-2024) and the OBBBA raising the ITC from 25% to 35% (Jul-2025).
• Vietnam (Bac Ninh) ramp from 2024 — a near-term gross-margin drag now approaching breakeven.
• Japan footprint rationalization — a two-year program targeting ~100bps of corporate gross-margin improvement by exiting 2027.
• Financing build-out — new revolver, term loan, 2033 notes, and the $1.15B 0% convertible (2031).
• February-2026 10M-share family secondary at $48.75.
• May-21-2026 Investor Day — used to defer hard long-term targets; management has been priming investors for a 2027 margin trough (Arizona dilution) before a 2028+ payoff.

Headwinds:

• Cyclical exposure — 2.26 beta, mobile-heavy; the April-2025 tariff crash to $14.69 is a fresh reminder of two-way risk.
• ~30% Apple concentration — pricing leverage, volume cycle, in-sourcing/dual-sourcing risk.
• ASP deflation — structural and management-acknowledged.
• Margin dilution from Arizona (1–2% op margin from 2027) and Vietnam ramp.
• FCF turning negative in 2026 as capex triples; balance sheet levering up.
• Rising effective tax rate as global minimum tax phases in (Singapore benefit already collapsed).
• Geopolitical — Korea/China/Taiwan/Vietnam concentration; BIS export controls limiting China sales; China-for-China competition rising.

Verdict: net thesis-neutral-to-negative for a buyer at this price. The strategic changes (TSMC, Arizona, CHIPS) genuinely strengthen the long-term franchise and optionality — but they also confirm the near-term picture the bulls are looking past: tripling capex, negative FCF, margin dilution, and a 2027 air-pocket, all already in the price.

9. Risk Analysis (Risk Matrix)

Key-person / control risk is structural rather than acute: the Kim family’s ~49.4% stake and chairmanship mean strategy and capital allocation hinge on family alignment. The most probable damage to a buyer here is not bankruptcy (very unlikely — net cash, profitable) but a multiple de-rating combined with a cyclical earnings dip, which on a 2.26-beta name can be severe (the stock has fallen ~65% peak-to-trough twice in the last four years, and ~90% in its lifetime).

10. Valuation Discussion (Embedded Expectations)

No price target; no recommendation. This section frames what the current price implies and the scenario range.

Where the multiple sits. At $90.46 (mkt cap ~$22.4B; EV ~$21.9B):

• EV/TTM sales ~3.1x vs a historical norm near ~1x — the 99.98th percentile of Amkor’s own 10-year price/sales range.
• EV/TTM EBITDA ~18x vs a historical ~6–9x.
• P/E ~51x trailing (on trough EPS of $1.76 TTM); ~45x on a ~$2.00 FY26E; ~29x even on the 2022 cyclical-peak EPS of $3.11.
• P/B ~5.0x vs a historical ~1–2x — also the 99.98th percentile (composite valuation 97.86th).

This is, unambiguously, the richest the stock has ever been on sales and book, on a denominator (earnings/FCF) that is cyclically depressed.

Comp context. Amkor’s ~3.1x EV/sales / ~18x EV/EBITDA is a steep premium to OSAT peers — ASE Technology trades roughly ~1.5–2x sales / high-single-to-low-double-digit EBITDA, and the Chinese OSATs cheaper still. Amkor no longer trades like an OSAT at all: its factor neighbors are semicap equipment names (AMAT, LRCX, FormFactor, Nova, Veeco) and a semis ETF — i.e., the market is valuing it on the AI-capex tape, at equipment-like multiples, despite ~5.6% net margins and ~7% trough ROIC versus those peers’ far higher returns.

Embedded-expectations math. To justify ~$22.4B of market cap at, say, a 15x exit P/E, the market needs Amkor to earn roughly $1.5B of net income at maturity — versus $374M in FY25. That requires net margin to roughly double (toward ~12–13% on ~$10–12B of revenue) — i.e., a simultaneous mix shift to high-rent AI packaging, sustained high-teens-plus gross margin, and substantial revenue growth. Possible in a bull case, but it requires the AI/Arizona optionality to convert into both volume and rent while the mobile base holds and the cycle does not roll. The market is underwriting the full success of a transformation that is, today, ~20%-of-revenue and back-end-loaded to 2028–2030.

Scenario analysis (illustrative, 2–3 year horizon):

Spot ($90) sits at the top of even the bull band and far above base — the classic full-price-at-a-cycle-peak asymmetry, where the bull case is needed merely to hold the price and any disappointment re-rates toward base/bear.

What the market is pricing correctly: the structural rise of advanced packaging, Amkor’s #2 position and genuine technology breadth, the strength of the current up-cycle, the TSMC/Arizona US-onshoring optionality, and the strong balance sheet. What it appears to be pricing incorrectly: that a ~5.6%-net-margin, ~7%-trough-ROIC, ASP-deflationary, Apple-concentrated, capital-hungry back-end service business deserves a record, equipment-like multiple on trough earnings — and that the high-rent AI economics (largely captured by TSMC) will accrue to the merchant packager.

11. Variant Perception

Consensus. Sell-side is, tellingly, Hold with an average price forecast of ~$69 — below the $90 spot. The buy-side tape, however, is euphoric: the stock has risen ~6x in 14 months and trades with the AI-capex momentum cohort. So consensus is bifurcated — fundamental analysts are cautious (the stock is above their targets), while price-momentum flows are firmly long. The factor data confirm a momentum/narrative-crowded name, not a value or quality one: Momentum +0.55, “Chip Sector Superstars” +0.89, SmallSize +0.34, Value absent, LowVolatility −0.89 (strongly anti-defensive), beta 2.26, ~50% idiosyncratic vol, related-stocks all semicap equipment. This is a high-beta cyclical being traded as an AI-capex proxy.

Strongest bull case. Advanced packaging is the new performance bottleneck; Amkor is the #2 merchant player and the natural US-onshoring partner for TSMC (validated by a 10-year deal and ~$2.8B of government support). AI/advanced packaging triples in 2026 and compounds for years; the Arizona campus adds a ~$1B, above-corporate-margin franchise by 2029–30; operating leverage drives gross margin to the high-teens; and the multiple is justified because Amkor is re-rating into an equipment-like, AI-levered growth company. Net cash and a high short-squeeze beta protect the downside.

Strongest bear case. This is a capital-intensive, cyclical, price-taking commodity packager at the richest multiple in its history on trough earnings. Through-cycle ROIC ~10–11% (trough ~7%) does not clear the cost of capital; ~30% of revenue is one customer that multi-sources; ASPs deflate structurally; the high-rent AI packaging accrues to TSMC, not Amkor (whose AI/advanced is only ~20% of revenue and undisclosed in detail); capex is tripling into negative FCF with a 2027 margin air-pocket; the comp plan has no ROIC hurdle; and the controlling family and every insider are net sellers (a 10M-share secondary at $48.75). When the mobile cycle or utilization rolls — as it did to −65% twice in four years — both the earnings and the record multiple compress together.

The 3–5 assumptions that matter most:

1. AI/advanced-packaging mix — does it become a disclosed, durable, high-rent, >30%-of-revenue driver, or stay a ~20% lower-rent tail? (Bull needs the former.)
2. Gross margin trajectory — does it sustainably break to the high-teens-plus, or stay mid-teens (commodity economics)?
3. Mobile/Apple cycle — does the ~46% communications base hold, or air-pocket?
4. Arizona returns — does the $7B build earn its cost of capital (above-corporate margin by 2029–30), or become a subsidized, dilutive money-sink?
5. Multiple — does the market sustain an equipment-like ~3x sales / ~18x EBITDA, or normalize toward Amkor’s own ~1x sales history?

Falsification tests. Bull is falsified if: gross margin fails to hold the high-teens through the next mobile down-cycle, or AI/datacenter revenue stalls below ~25% of sales while ASPs keep deflating. Bear is falsified if: Amkor discloses AI-datacenter revenue growing durably >30%/yr to a meaningful share of the mix with gross margin sustained in the high-teens, and Arizona ramps to above-corporate-average margins on real customer commitments — proving a genuine economic re-rate rather than a narrative one.

Where consensus may be offsides. The fundamental sell-side (Hold, target below spot) is likely closer to right than the momentum bid. The variant-perception risk for a short, though, is that this is a crowded, high-idiosyncratic-vol, net-cash momentum name with a live secular option and a controlling family — the kind of stock that squeezes hard on any AI/Arizona headline. The cleaner expression of the bear view is avoidance and patience, not a short.

12. Fact vs. Interpretation Table

13. Open Questions

1. What is Amkor’s actual AI-datacenter revenue (dollars / %)? Management has twice declined to disclose it; the ~20% “AI/advanced” figure conflates datacenter, PC, and advanced mobile.
2. What are the binding terms of the TSMC 10-year agreement (volume commitments, pricing, exclusivity, take-or-pay)? Undisclosed; the value of the “halo” depends entirely on these.
3. What return will Arizona actually earn? Management says “above corporate average” by 2029–30, but with no ROIC disclosure and a subsidized cost base, this is unverified.
4. What is the conversion price / capped-call structure of the $1.15B 2031 convert, and thus the true dilution path at various stock prices?
5. How much of the FY25 operating margin was the one-time Nanium receipt + equipment-sale gain — i.e., what is true run-rate operating profitability?
6. Will the mobile/Apple base hold through 2026–27, or air-pocket as it did in 2024–25?
7. Does Amkor formally claim the Nasdaq “controlled company” exemption, and what minority-protection mechanisms exist?
8. What did the May-21-2026 Investor Day disclose on long-term margin/revenue targets (deferred to repeatedly on calls)?

14. What Must Be True

For the bull case (to justify ≥$90 durably):

• AI/advanced-packaging (datacenter) must become a disclosed, durable, high-rent driver exceeding ~30% of revenue, not a ~20% lower-rent tail.
• Gross margin must break and hold in the high-teens-plus through the next mobile down-cycle (proving real mix re-rate, not just cyclical utilization).
• Arizona must ramp to above-corporate-average margins on real customer commitments and earn its cost of capital — the $7B bet must pay.
• The mobile/Apple base must hold while the AI mix grows.
• Falsification test: gross margin fails to sustain the high-teens through a mobile downturn, or AI-datacenter revenue stalls below ~25% of sales while ASPs keep deflating → the re-rate was narrative, not economic.

For the bear case (multiple normalizes toward history):

• The current valuation is a momentum/narrative artifact on trough earnings; through-cycle ROIC (~10–11%, trough ~7%) does not clear the cost of capital.
• A cyclical/utilization rollover or mobile air-pocket drops EPS back toward ~$1.50 while the multiple compresses toward Amkor’s ~1x-sales history → price toward $25–40.
• Capex tripling into negative FCF + a 2027 margin air-pocket + convert dilution erode per-share value during the build.
• Falsification test: Amkor sustains high-teens gross margin and discloses AI-datacenter revenue durably >30%/yr to a meaningful mix share and Arizona ramps profitably — proving the business has economically transformed and the multiple is warranted.

The single most important variable: gross margin through the next down-cycle. If it holds in the high-teens, the bull’s “economic transformation” is real; if it reverts to mid-teens, this remains a commodity packager at a record price.

15. Source Appendix

Primary sources: Amkor FY2021–FY2025 Forms 10-K (CIK 1047127), Q1-2026 Form 10-Q (filed 2026-04-28), 2026 DEF 14A (filed 2026-04-02), 8-Ks (CEO transition 2025-10-27; financings 2025-05/06/09 and 2026-05; family secondary 2026-02-13; vote results 2026-05-19), and the Form 4 corpus. Quantitative data: ROIC.ai (statements, ratios, EV), AZI (valuation_index own-history percentiles, daily price CSV, news feed), FactorsToday (factor loadings, leaderboard, related stocks). Catalyst/consensus: Benzinga (TSMC 10-year Arizona agreement, 2026-06-16; consensus/price-forecast, 2026-06-17). Earnings-call transcripts Q3’25/Q4’25/Q1’26 via ROIC.ai. The TSMC relationship is sourced to public press/IR, not the 10-K.

This article (sections 1–15) is deliberately position-free and contains no price target. The sole exception is the labeled “Claude’s Take” block at the top, which is the author’s own independent opinion and general information, not investment advice.

APPENDIX A — Standard Diligence Questionnaire

Supplemental to the research memo. Fact / Interpretation labeled where it matters. As of 2026-06-19; price $90.46.

General

What thoughtful questions have other investors asked about this company?

• Is the AI/advanced-packaging opportunity large enough — and high-margin enough — to justify a record multiple, or is the high-rent CoWoS work captured by TSMC in-house? (The central debate.)
• How much of revenue is really AI-datacenter (management won’t disclose; ~20% “AI/advanced” conflates datacenter/PC/advanced-mobile)?
• Does the $7B Arizona build earn its cost of capital, or is it a subsidized, dilutive money-sink with a 2027 margin air-pocket?
• How exposed is Amkor to a mobile/Apple cycle, given ~30% single-customer concentration?
• What does it mean to own a minority stake in a ~49.4% Kim-family-controlled company whose insiders only sell?

Cyclicality & Earnings Nature

Are earnings at a cyclical high or low? Low. FY25 diluted EPS $1.51 is roughly half the 2022 cyclical peak of $3.11. Revenue ($6.71B) remains below the 2022 peak ($7.09B). The trailing P/E of ~51x reflects a depressed denominator, not a structurally high one. (Fact + Interpretation.)

Driven by external environment or internal actions? Overwhelmingly external — the semiconductor cycle, mobile demand, utilization. Internal levers (Vietnam ramp, Japan rationalization, advanced-packaging investment) matter at the margin but do not override the cycle. (Interpretation.)

How stable are revenues? Unstable/cyclical. No material backlog; customers “do not commit to purchase any significant amount” and may “reduce, cancel or delay.” Revenue swings with mobile seasonality and the broader chip cycle; the stock fell ~65% peak-to-trough twice in four years. (Fact.)

Outlook for products/services? Advanced packaging (2.5D/HDFO/SiP/CPO) is the structural growth area; mainstream (wirebond/leadframe) is shrinking as a share. Advanced packaging’s value is rising with heterogeneous integration. (Fact + Interpretation.)

How big is the market, growing or shrinking? The OSAT market grows roughly with semiconductors (low-to-mid single digits through-cycle), with advanced packaging growing faster. International by nature (~95% of work in Asia; US customers 65.6% of revenue). (Interpretation.)

Business Quality & Competitive Moat

Is the industry getting more or less competitive? More — China-for-China OSAT capacity (JCET, Tongfu) is rising with state support, foundries (TSMC/Samsung/Intel) own leading-edge packaging, and a global capacity-add surge is underway. (Fact.)

How profitable is the business (ROIC, ROE)? ROIC ~7% trough / ~17% peak / ~10–11% through-cycle — at or below cost of capital for a 2.26-beta name. ROE looks high (15–50%) but is leverage/low-book-flattered; ROIC is the honest read. (Fact + Interpretation.)

How profitable is the industry — competitors, barriers? Low-margin, fragmented; #1 ASE, #2 Amkor, #3 JCET, plus Powertech/Tongfu/foundries. Low barriers to entry in commodity packaging; advanced packaging requires capital and qualification but is not a durable barrier. (Fact + Interpretation.)

Can the business be easily understood? Yes — a back-end semiconductor service business (package + test). (Fact.)

Can it be undermined by low-cost labor? Already is — its own cost base is Asian; Chinese OSATs are cheaper and state-supported. (Fact.)

Do brands matter? No consumer brand; what matters is qualified capacity, technology breadth, footprint, and co-development relationships. (Interpretation.)

Nature of competition / switching costs? Competition on technology, scale, footprint, and price (which “has generally declined”). Switching costs exist within a qualified product generation but customers deliberately multi-source. Captivity favors the customer. (Fact + Interpretation.)

Financial Condition & Balance Sheet

Assets not fully recognized? The qualified-capacity/co-development relationships and CHIPS grant/ITC receivables are economic assets not fully on the balance sheet; offset by the fact that the $11.4B gross PP&E is heavily depreciated ($7.4B accumulated). (Interpretation.)

Off-balance-sheet liabilities? Capex purchase obligations ($1,152M at YE25, mostly due within 12 months); operating leases; the $1.15B 0% convertible (on balance sheet but latent equity dilution). (Fact.)

How conservative is the accounting? Reasonable — low SBC (~$20M), clean cash conversion. But note FY25 operating margin was flattered by a one-time Nanium receipt + equipment-sale gain, and FY24 EPS by a +$0.20 depreciation useful-life extension. Normalize these. (Fact + Interpretation.)

How CapEx-hungry? Extremely. Capex ~11–14% of sales historically; tripling to $2.5–3.0B in 2026. D&A ~10% of revenue. This is the defining financial feature. (Fact.)

Capital Allocation & Management

How much FCF, and how used? Thin and volatile FCF ($191M FY25, 2.8% margin); used for a small dividend (no buyback). With capex tripling, FCF likely negative in 2026, funded by debt/convert. (Fact + Interpretation.)

Significant acquisitions? No — goodwill is a negligible $18M; Amkor grows organically. (Fact.)

Buying back shares? No buyback program (only tax-withholding shares). (Fact.)

Issuing shares to insiders? SBC is small (~$20M). But the $1.15B 0% convert is latent dilution, and the controlling family sold 10M shares at $48.75 (Feb-2026). (Fact.)

Compensation of directors/management? CEO FY25 $12.1M; pay ratio 656:1. Bonus on Operating Income; PSUs on EPS + relative TSR; no ROIC hurdle. Say-on-pay ~97.8% (family bloc votes it). (Fact + Interpretation.)

Motivations of management? Career operator CEO (Engel, 20-yr insider, de minimis stake) under a controlling family Chairman (Susan Y. Kim). Incentives reward asset/EPS growth, not returns on capital — aligned with a capacity build, not with per-share value at a peak. (Interpretation.)

Valuation & Market Data

ADR, MLP, or K-1 issuer? No — US-domiciled C-corp common stock; standard 1099 dividend reporting. (Fact.)

Dividend policy? Small, growing regular quarterly dividend (~$0.33/yr, ~0.4% yield); a ~$100M special was paid in 2024. Dividend-only capital return, subordinate to capex. (Fact.)

How profitable? Low-margin (5.6% net FY25), low-ROIC (~7% trough). (Fact.)

Net income diverging from cash from operations? OCF consistently exceeds net income (heavy D&A) — normal and healthy; the divergence that matters is OCF vs. FCF (capex consumes most OCF). (Fact.)

Risks & Downside

What would cause the stock to decline? Multiple de-rating from record levels; a mobile/Apple air-pocket or utilization rollover; AI optionality disappointing; Arizona overruns; convert dilution; a cyclical/tariff shock (as in April-2025). (Interpretation.)

Risk of catastrophic loss? Low — near net-cash, profitable, scaled #2 player. The realistic downside is a ~50%+ de-rate in a cyclical dip (which has happened), not insolvency. (Interpretation.)

Chance of total loss? Very low absent a multi-year structural demand collapse. (Interpretation.)

Recent News & Events

Has the business environment changed recently? Yes — the TSMC 10-year Arizona advanced-packaging agreement (Jun-16-2026), the $7B Arizona scale-up, CHIPS funding + 35% ITC, the CEO transition (Engel, Jan-2026), and a strong up-cycle (Q1’26 rev +27%). The stock has run ~6x off the April-2025 trough. (Fact.)

Significant acquisitions? None. (Fact.)

Change in accounting policies? FY24 depreciation useful-life extension (+$0.20 EPS); collapsing Singapore tax benefit. (Fact.)

Recent changes — new markets/facilities/management? Arizona campus (new US market/facility), Vietnam ramp, Japan rationalization, new CEO, multiple financings ($1.15B 0% convert, 2033 notes, term loan, revolver). (Fact.)

APPENDIX B — Source Appendix

Primary sources first. Accessed 2026-06-19 unless noted. Quantitative figures reconciled to SEC filings; third-party aggregated data (ROIC.ai, AZI, FactorsToday) cross-checked against filings.

Primary — SEC Filings (CIK 0001047127; SEC EDGAR, CIK 0001047127)

Primary — Earnings Call Transcripts (via ROIC.ai)

Quantitative Data Sources (third-party; reconciled to filings)

• ROIC.ai MCP — income statement, balance sheet, cash flow (FY2020–FY2025), profitability ratios (ROIC/ROE/margins), enterprise value, earnings-call list/transcripts.
• AZI (azitrading.com) — valuation_index own-history percentiles (P/E 93.6th, P/B 99.98th, P/S 99.98th, composite 97.86th; 2026-06-18); daily price CSV (split/dividend-adjusted full history, EMAs, beta); news feed.
• FactorsToday — factor loadings (Momentum +0.55, Chip Sector Superstars +0.89, LowVolatility −0.89, Value absent), leaderboard (beta 2.26, lifetime max drawdown −90.4%, y1 return +348%), specific volatility (~50% annual), related stocks (FORM/NVMI/VECO/UCTT/AMAT/LRCX/PSI).

Secondary — News & Catalyst

Key Quantitative Figures Used (FY2025 unless noted)

• Revenue $6.71B; gross margin 14.0%; operating margin 7.0%; net margin 5.6%; diluted EPS $1.51. (ROIC; 10-K)
• Revenue/EPS history: $5.05B/$1.40 (20), $6.14B/$2.62 (21), $7.09B/$3.11 (22 peak), $6.50B/$1.46 (23), $6.32B/$1.43 (24), $6.71B/$1.51 (25). (ROIC)
• ROIC: 11.1% (20), 17.2% (21), 16.9% (22), 7.3% (23), 6.6% (24), 6.8% (25). (ROIC)
• CapEx: $553M/$780M/$908M/$749M/$744M/$905M (20–25); FCF: $217M/$342M/$190M/$521M/$345M/$191M. 2026 capex guide $2.5–3.0B. (ROIC; transcripts)
• Balance sheet YE25: cash+ST inv $1.99B; total debt ~$1.45B; net cash ~$0.5B; gross PP&E $11.4B; goodwill $18M; shares 247.3M. (ROIC; 10-K)
• Valuation at $90.46: mkt cap ~$22.4B; EV ~$21.9B; ~3.1x EV/TTM sales; ~18x EV/TTM EBITDA; ~51x trailing P/E; ~5.0x P/B. (ROIC EV at spot; AZI)
• Price arc: 2022 low $14.55; 2024 high $43.02; April-2025 low $14.69; Dec-2025 high $46.90; 2026-06-18 close $90.46; intraday 52-wk high $96.68. (AZI CSV)

Note: the TSMC relationship and the June-2026 10-year agreement are sourced to public press/IR (Benzinga; company release), not the SEC 10-K, which does not name TSMC.