Amkor at Goldman Sachs Communacopia + Technology Conference: ai demand drives expansion

Amkor raised planned investment in Arizona phase 2 to $12 billion from $7 billion, backed by advanced-packaging demand from AI and compute customers and a $1.5 billion NVIDIA prepayment commitment. Management expects compute revenue to rise 30% sequentially in Q3 and sustain roughly 30% year-over-year growth into 2027, while targeting more than $11 billion of 2030 revenue, 22% gross margin and $5 EPS. The Arizona campus is scheduled to begin production in 2028 and reach full capacity by 2030, but the buildout is expected to dilute 2027 operating margin by 1%-2% and pressure free cash flow during peak 2028-29 capex.

Analysis

AMKR is transitioning from a cyclical OSAT multiple toward an infrastructure-like advanced-packaging asset, but the equity will likely discount the funding burden before it credits 2030 utilization. The key underwriting issue is not demand visibility but whether customer commitments convert into enforceable take-or-pay economics that protect returns on a very large, front-loaded build. A customer prepayment reduces financing risk but does not establish revenue, utilization, or return-on-invested-capital; absent additional deposits or binding volume commitments, AMKR's free-cash-flow profile could deteriorate materially through the 2028-29 capex peak.

Near term (days to 3 months), the sharper compute outlook can support estimate revisions and a rerating versus legacy packaging peers, while smartphone softness and the Vietnam transfer create a cleaner relative setup: advanced-packaging growth can mask weaker communications volumes. The more important 12-18 month catalyst is disclosed customer allocation, pricing, and yield data for high-density fan-out and 2.5D programs. TSM benefits indirectly because nearby packaging reduces cycle time and supply-chain friction for U.S.-made wafers; however, AMKR competes for a portion of the packaging value pool otherwise retained by TSM's CoWoS ecosystem and faces technology pressure from INTC's EMIB/Foveros roadmap.

Consensus may be over-extrapolating AI scarcity into durable returns. Packaging bottlenecks attract capacity quickly, and an eventual easing in substrates, HBM availability, or foundry advanced-packaging capacity can shift bargaining power back to customers before Arizona reaches scale. The thesis is falsified if compute growth decelerates below the stated structural pace for two quarters, if 2027 margin dilution exceeds the indicated range, or if net leverage trends above the stated ceiling without incremental customer-funded capacity commitments.

Key Decisions for Investors

• Initiate a starter long AMKR over the next 1-3 months only on semiconductor-market weakness; target a 12-18 month catalyst path around customer volume commitments and 2027 compute guidance. Size modestly: the upside is estimate/multiple expansion, but the principal risk is a capex-driven FCF reset before Arizona revenue.
• Express the AI packaging theme as long AMKR / short INTC in equal dollar amounts for 6-12 months. AMKR has a nearer-term external-customer packaging ramp, while INTC remains exposed to execution and capital-intensity risk in its foundry build; exit if INTC demonstrates sustained foundry customer wins or AMKR's compute growth falls below 20% year over year.
• Do not underwrite NVDA upside from the supplier relationship alone. Treat additional NVDA exposure as an alert contingent on confirmation that prepaid capacity translates into shipment volumes and that packaging availability, rather than HBM or system integration, remains the binding constraint.
• Set a 2027 risk alert: reduce AMKR if operating-margin pressure exceeds 2 percentage points, capex rises above the communicated deployment cadence, or management funds the expansion with leverage trending above 1.5x. Those signals would imply the project is consuming balance-sheet capacity faster than customer economics are de-risking it.