SKHY SK hynix

Research status: Primary-source verified through Q2 2026 and the July 2026 Nasdaq ADR listing. Valuation basis: SKHY closed at approximately $165.70 on 21 August 2026; refresh before any capital decision.

SK hynix is the strongest pure-play public exposure to the AI-memory bottleneck. Its July 2026 U.S. prospectus cited 56.4% global HBM revenue share in Q1 2026 and 29.1% DRAM share, while Q2 results showed extraordinary operating leverage: revenue KRW 79.3T, +257% YoY, operating profit KRW 60.5T, +557%, and operating margin 76%. HBM4 entered mass shipment in Q2, HBM4E samples have been delivered to major customers, and the company has finalized multi-year LTAs with around 10 customers. Cash reached KRW 88T and net cash KRW 69.4T at Q2.

The core investment question is not whether SK hynix is currently winning HBM—it is—but how much of today's extreme memory pricing/margin environment is durable. The stock should therefore be valued on normalized through-cycle earnings, not annualized Q2 net income. The ADR also creates a new U.S. valuation/access catalyst, but the July offering issued 17.79M new common shares (~2.5% pre-offering dilution) to back 177.9M ADSs; each ADS represents one-tenth of a common share.

PowerFund 24-month scenario framework: Bear 30% = $95; Base 50% = $225; Bull 20% = $400; probability-weighted value ~$221, implying +33.4% total / ~15.5% annualized from $165.70. 60-month framework: Bear 30% = $85; Base 50% = $325; Bull 20% = $650; probability-weighted value ~$318, implying +91.9% total / ~13.9% annualized. These are PowerFund scenarios, not company guidance or analyst targets.

Investment case

SK hynix is the current technology/share leader in high-bandwidth memory, the memory class most directly tied to accelerator performance. HBM is structurally more difficult than commodity DRAM because stack height, thermals, packaging, yield, interface speed and co-development with accelerator vendors matter simultaneously. The company's 2026 U.S. prospectus cited 56.4% HBM revenue share in Q1, and its multi-year technology partnership with NVIDIA explicitly aligns next-generation memory development with NVIDIA's AI infrastructure roadmap.

The moat is therefore broader than wafer capacity. It includes HBM process/yield know-how, Advanced MR-MUF packaging, customer co-design, qualification history, advanced packaging capacity, and the ability to ramp new generations on schedule. HBM4 began mass shipments in Q2 2026; 12-layer HBM4E samples were shipped in June with up to 16Gbps per pin and improved power efficiency.

The thesis is investable if (1) SK hynix retains clear HBM technology/share leadership through HBM4/HBM4E, (2) long-term agreements and AI-memory mix make the next memory downcycle materially less destructive than historical DRAM cycles, (3) capacity expansion earns attractive returns rather than recreating oversupply, and (4) conventional DRAM/NAND pricing does not overwhelm the superior HBM economics.

Verified operating baseline

• Q2 2026 revenue: KRW 79.319T, +51% QoQ and +257% YoY.
• Q2 operating profit: KRW 60.543T, +61% QoQ and +557% YoY.
• Q2 operating margin: 76% versus 41% a year earlier.
• Q2 cash and cash equivalents: KRW 88T; debt KRW 18.6T; net cash KRW 69.4T.
• Q1 2026 global DRAM revenue share: 29.1%; HBM revenue share: 56.4%, per IDC figures cited in SK hynix's U.S. prospectus.
• HBM4 mass shipments began in Q2 2026 and are scheduled to ramp in H2.
• HBM4E 12-layer samples shipped to major customers in June 2026.
• LTAs finalized with around 10 customers, with additional discussions continuing.
• DRAM represented 77.3% of Q1 2026 sales and NAND 22.0%, highlighting continuing memory-cycle exposure.
• The July Nasdaq offering sold 177.9M ADSs at $149, backed by 17.79M newly issued common shares; each ADS represents 0.1 common share.

Valuation framework

Because current margins are exceptionally above historical memory-cycle norms, PowerFund values SKHY using normalized future ADS earnings/cycle conditions rather than annualizing Q2 net income.

24 months

Probability-weighted working value: ~$221, or +33.4% total / ~15.5% annualized.

60 months

Probability-weighted working value: ~$318, or +91.9% total / ~13.9% annualized.

• HBM4 ramp through H2 2026 with yields/margins consistent with current technology leadership.
• HBM4E customer qualification and conversion from samples to committed 2027 production.
• Multi-year NVIDIA roadmap collaboration translating into durable content and supply commitments.
• Additional LTAs/prepayments/minimum-volume structures that reduce historical memory-cycle volatility.
• Continued high-value AI server DRAM, SOCAMM and enterprise SSD demand.
• U.S. ADR broadening ownership, liquidity and valuation comparability with Micron/U.S. semiconductor peers.
• Capacity additions at M15X, Yongin and Cheongju translating into profitable supply rather than oversupply.
• Capital returns: management's 2025–27 policy targets returning 50% of cumulative FCF subject to balance-sheet goals; the August 2026 announced large buyback/cancellation is an additional per-share catalyst, subject to ongoing verification through company/SEC filings.

• Memory-cycle risk: Q2 2026's 76% operating margin is extraordinary and should not be treated as normalized; DRAM/NAND pricing can reverse violently.
• HBM competition: Samsung and Micron can narrow the technology/yield gap, pressure pricing or win larger shares of future NVIDIA/custom-accelerator platforms.
• Customer concentration: the prospectus disclosed meaningful concentration; the two largest Q1 2026 customers represented 14.8% and 12.4% of revenue.
• NVIDIA dependence/factor concentration: deep NVIDIA alignment is strategically valuable but creates customer/platform concentration and overlaps strongly with PowerFund's existing AI-infrastructure factor.
• Capex/oversupply: SK hynix has approved very large new fab and packaging investments. If industry participants build simultaneously, today's shortage economics can self-destruct.
• Geopolitics/export controls: Korean manufacturing, China operations/customers and U.S.-China semiconductor controls create material policy risk.
• FX/ADR: SKHY introduces KRW/USD translation, depositary mechanics, Korean governance/tax considerations and a relatively new U.S. trading history.
• NAND/Solidigm: NAND remains structurally more commodity-like and can dilute HBM economics.
• Technology transition: custom HBM, HBF, CXL or alternative memory architectures can shift value pools; leadership must be renewed every generation.
• Peak-expectations risk: the stock can fall sharply even on excellent results when investors anticipate a pricing peak.

WARNING / FREEZE ADDITIONS

• HBM revenue/share evidence falls materially below the current leadership position or competitor qualification wins accelerate.
• HBM4 yields, power/performance or shipment ramp miss customer requirements.
• Company-wide operating margin falls below ~45% for two quarters while HBM demand is still strong, suggesting mix/price deterioration faster than expected.
• Conventional DRAM/NAND inventories rise materially across the industry while new capacity continues to accelerate.
• Capex rises materially faster than revenue/FCF for multiple quarters without contracted demand support.
• Customer concentration rises further or NVIDIA-related exposure becomes effectively binary.
• ADR trades at a persistent unexplained premium to the KRX economic equivalent after fees/FX.

REDUCE — normally trim 25–50%

• HBM share falls below ~45% with simultaneous ASP or margin pressure.
• HBM4/HBM4E major customer qualification slips by a generation or material share shifts to Samsung/Micron.
• Two consecutive quarters show both falling revenue and sharply compressing operating margin from memory oversupply.
• Management abandons capex discipline and industry capacity growth materially exceeds contracted demand.
• FCF deteriorates despite accounting profit because inventory/capex absorb the cash.
• A major U.S./China policy change impairs key customers, China assets or advanced-memory shipment ability.

EXIT / THESIS INVALIDATED

• SK hynix loses its HBM technology/yield leadership for a full product generation and cannot recover through HBM4E/next generation.
• HBM economics converge toward commodity DRAM with no durable qualification, packaging or customer-co-design moat.
• The company repeats a severe oversupply cycle created by undisciplined capacity additions, destroying through-cycle ROIC despite AI growth.
• A geopolitical/export-control event permanently impairs a material portion of production or customer access.
• Governance/ADR structure materially weakens minority-shareholder economics or access.

The core HBM competitive set is Samsung Electronics and Micron. SK hynix currently leads on market share and has the deepest public evidence of NVIDIA co-development, but the moat is technological rather than permanent: each HBM generation resets part of the competitive race.

Samsung's advantages are semiconductor breadth, enormous balance-sheet/capacity resources and internal foundry/logic capabilities. Micron's advantages include U.S. domicile, improving HBM technology and direct access to U.S. policy support. SK hynix's edge is currently HBM specialization, yield/packaging execution, time-to-volume and customer trust.

The most important structural question is whether LTAs and customized HBM turn memory from a spot/commodity cycle into a partially contracted strategic component. If yes, historical trough multiples/earnings are too pessimistic. If no, Q2 2026 margins are likely closer to a cyclical peak than a new normal.

Portfolio interaction: SKHY is highly aligned with PowerFund's AI Infrastructure theme but would increase the same factor already represented by CLS, VRT, NVT and prospective SNDK/CRDO exposure. It should sit inside the fund's soft memory sleeve rather than be sized like a diversified compounder. Relative to APH, SKHY offers more direct AI-memory upside but much greater cycle/customer/geopolitical risk. Relative to CGNX, it has stronger current AI earnings momentum but much less theme diversification.

1. Build a quarterly HBM share/revenue/ASP tracker versus Samsung and Micron; separate HBM growth from conventional DRAM pricing.
2. Reconcile Q2 net income—which exceeded operating profit—to non-operating/one-time items; do not use reported Q2 net margin as normalized earnings power.
3. Model 2027–2030 wafer and advanced-packaging capacity additions at M15X, Yongin, Cheongju and competitors; compare contracted demand with supply growth.
4. Track HBM4 yield, HBM4E qualification and NVIDIA/custom-accelerator platform allocations.
5. Quantify top-customer/NVIDIA-linked revenue exposure and the economic terms/duration of LTAs where disclosures permit.
6. Build a normalized FCF bridge including capex; test shareholder returns only after funding high-return capacity.
7. Monitor SKHY ADR premium/discount versus 0.1 KRX common share translated at spot KRW/USD, including depositary fees/tax.
8. Compare normalized valuation with MU, Samsung and APH rather than using peak quarterly EPS.
9. Refresh 24/60-month scenarios after Q3 2026, especially if memory ASPs or hyperscaler capex guidance changes materially.

Next review: Q3 2026 results, HBM4E qualification news, or material change in memory pricing/capex plans.