SK hynix: HBM growth, cyclical valuation and shareholder returns

Business and cycle: how demand reaches the accounts

BUSINESS & CYCLE

Memory-company earnings depend on prices, product mix, shipments and supply timing. AI demand matters, but is not the only variable.

Research on SK hynix should cover both DRAM and NAND, distinguishing high-value products such as HBM, server DRAM and enterprise SSDs. AI infrastructure changes demand, while conventional-memory prices, process transitions and customer purchasing cycles still affect profits.

The July 29 announcement attributes quarterly performance to high-value product sales and rising DRAM and NAND prices. This suggests three research layers: whether end demand persists, whether it converts into shipments, and whether product mix and supply conditions support current margins.

The transmission chain from demand to earnings

AI and server workloads

Training, inference and service expansion create different capacity and bandwidth requirements

Customer procurement and qualification

Long-term agreements, platform design, product qualification and actual delivery

Products and manufacturing

HBM, DRAM and NAND mix; yields, packaging and capacity

Earnings and capital allocation

Prices, costs and capital expenditure jointly determine distributable cash

[1] SK hynix Q2 2026 results · 2026-07-29

HBM

Bandwidth, packaging and platform integration matter; leadership requires sustained customer validation and deliverable capacity.

Server DRAM

Capacity and power requirements affect mix; prices, generational transitions and customer inventories still require attention.

NAND / eSSD

Demand and product upgrades influence revenue mix, while process transitions and supply changes affect costs and prices.

Operating performance: separate operating profit from net income

OPERATING PERFORMANCE

High growth is the starting observation. Earnings quality also requires consistent comparisons and cash evidence.

Revenue and operating profit: three comparison periods

Company disclosure · KRW trillion · 2026-08-22T00:00:00Z

The three comparison periods in the July 29 release are not consecutive quarters and are not plotted as a continuous time series.

[1] SK hynix Q2 2026 results · 2026-07-29

Operating profit increased from approximately KRW 9.213 trillion a year earlier to KRW 60.543 trillion, outpacing revenue. This is consistent with combined effects from prices, product mix and operating leverage. The release does not allocate the entire increase among these drivers, so it should not all be attributed to HBM share or one technical advantage.

The company reports Q2 net income of approximately KRW 93.923 trillion, above revenue. This is not an operating margin. Separate operating profit, net income and operating cash flow; the full components of net income require the relevant financial statements and should not be mechanically extrapolated as recurring operating earnings.

Core financial measures and definitions

The approximately 76% operating margin in the text follows the rounded company disclosure. This table recalculates from underlying amounts: different precision, the same definition.

[1] SK hynix Q2 2026 results · 2026-07-29

Financial capacity: cash, debt and net cash

Company disclosure · KRW trillion · 2026-06-30 · 2026-08-22T00:00:00Z

Net cash equals cash less debt. These balances are related, not three additive asset categories.

[1] SK hynix Q2 2026 results · 2026-07-29

Financial judgment

Improved net cash creates room for expansion and shareholder returns, but a cash balance is not a forecast of future free cash flow. Customer payments, inventories, capital expenditure and buyback timing can all change future distributable cash.

HBM competitiveness: from chips to system integration

HBM & SYSTEMS

Market share is one outcome. Customer qualification, manufacturing yields, packaging and platform integration are processes to monitor continuously.

The July 29 release states that HBM4 mass shipments began in Q2, with production expansion planned for the second half. It also reports long-term agreements with around 10 customers. These technical and commercial developments provide evidence, but the agreement count does not establish contract value or years of secured revenue.

HBM is closely connected to GPUs, packaging and system design. Customers consider not just chip bandwidth but power, packaging, supply reliability and system operation. Competitiveness therefore means more than better specifications: it also requires reliable delivery within customer systems.

Four layers of competitiveness, each requiring evidence

01

Platforms and customers

Examine design wins, qualification and continued adoption; a single qualification does not secure permanent share.

02

Chips and yields

Examine performance, power, process maturity and deliverable quantities; leading specifications do not necessarily imply lower costs.

03

Packaging and systems

Examine stacking, thermal management and integration with compute platforms; technical roadmaps need practical system validation.

04

Capacity and fulfillment

Examine production lines, equipment, materials and delivery schedules; planned capacity is not completed supply.

Industry comparison: conditions as well as outcomes

Use comparable generations, periods and statistical definitions. Verify technical progress, market share and earnings contributions separately.

[1] SK hynix Q2 2026 results · 2026-07-29

Research should not rely exclusively on a leader-stays-ahead narrative. A useful counterquestion is whether current margins can persist if competitors improve customer qualification, supply or product mix. This requires updated evidence, not a permanent technology rating for any manufacturer.

Expansion and returns: two uses of cash

CAPITAL ALLOCATION

Concurrent expansion and buybacks indicate more active capital allocation. Value creation depends on future output and execution prices, not the actions alone.

The quarterly announcement emphasizes accelerating M15X production and investing in future capacity and advanced packaging. Distinguish planning, construction, equipment installation, production and customer qualification: each stage is a different distance from saleable output.

Expansion improves supply capability but can amplify depreciation and price pressure when demand slows. Investment efficiency depends on timely qualified output, customer acceptance, and prices and yields sufficient to cover new depreciation. Cash provides a buffer, not a substitute for testing project returns.

Not an either-or choice: capacity and shareholder returns

Future cash-generation capacity

Invest in operating capability

Production lines and equipment, process transitions, advanced packaging and customer delivery

Return capital to shareholders

Buybacks and cancellation, cash dividends and approved additional distributions

Both uses depend on cash flow, market conditions and board decisions. Authorization is not completed spending, and construction is not operating capacity.

[1] SK hynix Q2 2026 results · 2026-07-29

[2] SK hynix repurchase and cancellation · 2026-08-19

How to assess capital expenditure

Focus on usable capacity, output and unit costs created by investment, not whether the spending amount sets a record.

How to assess capital-return plans

Track actual repurchase prices, cancelled shares and dividend arrangements instead of interpreting an announced amount as a share-price floor.

Valuation: assess P/B and P/E together

VALUATION & ASSUMPTIONS

A cyclical stock's low P/E may reflect high earnings. P/B is not inherently more reliable: returns on assets, asset quality and future earnings still determine the appropriate multiple of book value.

P/B asks what returns book value can generate; P/E asks how long earnings can persist. They are not fully independent, because future profit affects accumulated book value and both can share the same cycle assumptions. Weighting can organize a judgment, but cannot remove forecast error.

The following forward book-value and EPS inputs come from the original report and remain assumptions not checked against the institutional originals. Adjusting multiples illustrates sensitivity; the calculations are neither price targets nor current market quotes.

Original model inputs: identify their evidentiary status

Used only to reproduce and assess the original valuation method. Without the underlying forecast publications, these cannot be described as institutionally confirmed targets or market consensus.

Two-method sensitivity: multiples and conditional values

Conditional calculation, not a price target or return promise.

3542504.36 KRW / ordinary share

Default inputs come from the original report. Multiples are research assumptions and outputs are conditional calculations. P/B and P/E estimates are not independent statistical samples and imply no probabilities or return guarantees.

How the conditional result is calculated

At the default inputs, P/B gives KRW 3,784,536.6 and P/E gives KRW 3,179,456. Weighting them 60%/40% gives KRW 3,542,504.36 per Korean ordinary share. The result changes when forecast earnings or book value change.

A P/B assumption based on a historical cycle peak plus an HBM premium requires an explanation of cycle comparability. Product mix, capital intensity, returns and competition can reduce the relevance of historical multiples. A 20% premium is a judgment, not a fact automatically established by the label leader.

P/E depends on earnings durability. If EPS is near a cycle peak, a lower multiple may compensate for normalization; sustained high returns could support different pricing. Both explanations need testing against subsequent delivery, pricing and cost data rather than selecting only the bullish interpretation.

Institutional differences: assumptions, not just targets

EXPECTATIONS & DISPERSION

Analysts may use different forecast years, earnings definitions, reference cycles and multiples for the same company. A list of targets cannot replace an explanation of those differences.

Differences often concern earnings duration, normalized profitability and capital returns, not simply the multiple. Align forecast years and security definitions before attributing gaps to inputs or methods. The six institutional figures collected in the original report lack complete original-source verification and remain an unverified appendix, not current market consensus.

Four questions for understanding differences

Original institutional summary · Original publications not verified

Retained for comparison with the supplied original report. This does not verify current institutional views, publication dates or permission status. It is not used to calculate potential returns or claim that all institutions remain bullish.

Research conclusion

A better question than who has the highest target is which operating change would lower valuations under several methods together. If they share an assumption of sustained high margins, margin normalization should be a central counter-scenario.

Buybacks and cancellation: execution, not a price floor

REPURCHASE & RETIREMENT

Board authorization, actual purchases and final cancellation are separate events. Buyback intent does not establish fixed demand or guarantee price support.

The August 19 plan totals KRW 40 trillion, estimated at approximately 24.07 million shares or 3.3% of then-outstanding shares using the prior day's reference price. Actual purchasable shares depend on execution prices and implementation; this is not the number already cancelled.

The announcement says repurchases start August 20, are expected to take approximately three months, and are intended to be cancelled after acquisition. The return framework proposes expanding from up to 50% to over 50% of three-year cumulative FCF. Arrangements remain subject to cash flow, markets, distributable profits and board approval, rather than an unconditional fixed payout commitment.

Execution chain: from authorization to per-share effects

Board authorization

KRW 40 trillion plan; share estimate at a reference price

Actual purchases

Exchange disclosures of quantities, prices and cumulative spending

Completed cancellation

Verify cancellation announcements and changes in shares outstanding

Per-share effect

Calculate using consistent profit and share-count definitions

[2] SK hynix repurchase and cancellation · 2026-08-19

Conditional arithmetic: how fewer shares affect EPS

1 / (1 − 0.033) − 1 = 3.412616%

Assumes unchanged profit, a 3.3% share reduction and completed cancellation. Excludes funding costs, changes in cash income and other equity changes. This is not realized EPS growth.

[2] SK hynix repurchase and cancellation · 2026-08-19

Mechanically assuming a 3.3% reduction in shares and unchanged profit produces an EPS increase of approximately 3.4%, not a 3.3% share-price return. The company also spends cash; interest income, investment capacity and other equity changes affect per-share value.

Market impact depends on execution timing and prices. Dividing authorization by planned trading days gives a mechanical average, not proof of fixed daily buying. Track actual purchases, cancellations and share-count changes instead of assuming a price floor.

Risks, catalysts and cross-market boundaries

RISKS & NEXT EVIDENCE

Operating research and the ADS premium are separate analytical questions. Improvement in the former does not guarantee convergence in any particular direction in the latter.

Earnings-cycle normalization

Changes in prices, supply or product mix reduce margins

Update prices, shipments, mix and costs; recalculate EPS and BVPS assumptions.

Competition and customer concentration

Competition intensifies within a generation, or qualification and delivery timing change

Use verifiable qualification, fulfillment and customer information rather than a single market-share estimate.

Slower AI capital spending

Customer investment and actual utilization fall short

Distinguish announced spending, actual construction and ultimate demand.

Capital-allocation outcomes

Expansion returns weaken or buybacks and cancellation differ from expectations

Verify capacity efficiency, cash spending and actual cancellation; authorization is not an outcome.

Follow-up events: update judgments with evidence

Event dates follow official announcements. This is a research checklist, not an enabled automatic reminder.

[1] SK hynix Q2 2026 results · 2026-07-29

[2] SK hynix repurchase and cancellation · 2026-08-19

Relationship to the homepage premium tracker

This report retains the August 22 operating-research window rather than backfilling later quotes into a historical judgment. Cross-market gaps should be assessed using the separate premium tracker and its market timestamps. Conditional valuation does not replace current prices, ADS conversion or trading-feasibility checks.