AMKRAmkor Technology Inc.

Investment Thesis

Amkor Technology is the world's largest U.S.-headquartered OSAT — outsourced semiconductor assembly and test — and the leading independent merchant supplier of advanced packaging: flip chip, wafer-level processing, 2.5D integration, high-density fan-out (HDFO), fine-pitch bumping, and advanced system-in-package. As Moore's Law slows and AI accelerators shift to chiplet-plus-HBM architectures, the binding constraint moves from leading-edge wafer fabrication to advanced packaging capacity. Amkor is the merchant alternative to TSMC's captive CoWoS franchise — the only independent player with the scale, qualification record, and CHIPS-gated U.S. footprint to absorb AI-packaging overflow.

The thesis is now backed by filed numbers, not narrative. Q2 2026 revenue was a record $1.898B, up 26% YoY, with gross margin expanding to 16.8% from 12.0% a year earlier, operating income of $200M (10.5% margin), and net income of $174M. Management guided Q3 to $1.95–2.05B with gross margin of 18.5–19.5%. The margin expansion is the clearest evidence of pricing power in the advanced-packaging layer: each +100bp of gross margin is roughly $19M per quarter of incremental gross profit at the current run-rate, and the roughly +6.5pp expansion from Q2 2025 to the Q3 2026 guide midpoint annualizes to approximately $494M of additional gross profit.

The company is spending to lock the position: FY2026 capex guidance of $2.5–3.0B — roughly 37–45% of revenue versus a historical OSAT norm of about 15–18% — is a direct bet that AI/HPC packaging demand outruns supply. With a $16.1B market cap and forward P/E near 25, the market is paying up for growth — but the stock has already re-rated (+187% over one year), so the question is whether the buildout converts into durable margin. Deep-research verdict: STRENGTHENED — scores ts 8 / ec 7 / ttm 8. Research-only at POOL stage.

Physical AI / Value-Chain Relevance

Layer 1 — AI Factory & Cloud Training Infrastructure. Every AI accelerator, HBM stack, and custom ASIC must be physically assembled and packaged before it can run; advanced packaging is the layer where chiplets, interposers, and HBM stacks are joined into a functioning device. Amkor's technology portfolio — HDFO, 2.5D integration, advanced flip chip, fine-pitch bumping, wafer-level processing, and advanced SiP — is not commodity wirebond; it is exactly the layer AI/HPC is pulling on, and 82% of Q2 2026 revenue came from advanced products.

The moat has three pillars. Qualification depth: advanced packaging is qualified per-customer and per-device; re-qualifying a supplier risks a product launch, which is why Apple (29.8% of FY2025 revenue) and Qualcomm (11.1%) are effectively locked into multi-year programs — 40.9% of revenue from two customers who cannot easily switch. U.S. sovereign footprint: the Arizona advanced-packaging facility, backed by up to $407M in direct CHIPS Act funding, is a sovereignty-positioned capacity asset Asian OSAT peers cannot replicate. Scale: 72% of FY2025 revenue from the top-10 customers, concentrated in exactly the layer AI demand is expanding.

Catalysts

• Q2 2026 results (July 27): record revenue of $1.898B (+26% YoY), record Computing and Automotive/Industrial end-markets, gross margin 16.8%, and FY2026 capex raised to $2.5–3.0B.
• Q3 2026 guide: $1.95–2.05B sales with 18.5–19.5% gross margin — a step-up that, if delivered, confirms the mix shift is durable.
• TSMC Arizona advanced-packaging partnership: a 10-year U.S. onshoring ramp that creates merchant capacity demand around TSMC's captive fabs.
• Broadcom/Google TPU V6/V7 CoWoS-style packaging slots: hyperscaler custom-silicon programs that route merchant packaging volume to Amkor.
• 18 active 2.5D engagements: a program pipeline that supports the multi-year revenue build.

Positioning / What the Market May Be Missing

The market has historically priced Amkor as a low-margin, commodity OSAT — structurally inferior to foundry, hostage to smartphone cycles. The Q2 2026 print challenges that framing on two fronts. First, the gross-margin trajectory (12.0% → 14.2% → 16.8% → 18.5–19.5% guide) is a pricing-power proof that looks like a mix shift toward high-value advanced packaging, not a utilization blip. Second, the 'record Computing' end-market line reads through directly to data-center AI/HPC demand — Amkor packages the accelerators that hyperscalers deploy, and it is the only U.S.-headquartered merchant with that exposure.

What is under-appreciated: the CHIPS-gated Arizona facility turns Amkor into a sovereign-capacity story at a time when U.S. policymakers want AI supply chains onshore; the balance sheet already shows the buildout ($629.0M construction-in-progress at June 2026, up from $250.2M in December 2025); and the 82% advanced-product mix means AI/HPC revenue is not an optionality story — it is the current business. The crowding flag is YELLOW; conviction is capped pending graph completion (the alpha pack still lacks layer/technology/customer-class wiring), which keeps this research-only.

The bear case is equally important: if TSMC keeps 2.5D/3D integration largely captive, Amkor captures only the overflow and the $2.5–3B capex bet over-earns capacity; and 40.9% of revenue sitting with two customers is a genuine concentration risk.

Risks and What Invalidates the Thesis

• Captive-foundry risk (the big one): TSMC's in-house CoWoS is the dominant advanced-packaging franchise; Samsung and Intel also package in-house. If AI accelerator packaging stays captive, Amkor's upside is capped at non-captive flows.
• Gross-margin ceiling: if Q3/Q4 2026 gross margin stalls below the 18.5% guide midpoint, the expansion looks like a utilization cycle rather than a durable mix shift.
• Customer concentration: Apple (29.8%) plus Qualcomm (11.1%) equals 40.9% of FY2025 revenue; the 10-K flags customer combinations and order reductions as direct risks.
• Balance-sheet leverage: total debt rose to $2.73B (from roughly $1.46B a year earlier) to fund capex; an overshoot turns the leverage into a headwind.
• Competitive intensity: ASE, JCET, and Powertech compete on price for OSAT work, structurally capping margins below foundry levels.

The cleanest disconfirmers: TSMC publicly confirming 2.5D/3D stays fully captive at scale, or two consecutive quarters of gross margin below the 18.5% guide midpoint.

What to Watch Next

1. Q3 2026 results (late October) versus the $1.95–2.05B / 18.5–19.5% GM guide — the first clean test of margin durability.
2. Disaggregated 2.5D/CoWoS-equivalent capacity disclosures (Amkor does not currently publish them) — the data that would confirm the bottleneck claim quantitatively.
3. Arizona facility milestones and CHIPS disbursement cadence — the sovereign-capacity proof points.
4. The specific AI/HPC customers behind 'record Computing' (22% of Q2 revenue) — naming them would upgrade conviction.
5. Apple/Qualcomm program commentary and any customer-concentration shifts in the next 10-Q.