After SK hynix’s 76% Operating Margin: How Durable Is HBM Growth?

SK hynix’s preliminary second-quarter 2026 results were exceptionally strong: revenue of KRW79.3187 trillion, operating profit of KRW60.5426 trillion and an operating margin of approximately 76%. The company cited significant sequential price increases in both DRAM and NAND, alongside higher sales of HBM, server DRAM and enterprise SSDs.^[1] Price and mix both matter; the result cannot be assigned entirely to one product.

The next investment question is durability. Multi-year agreements improve demand visibility, HBM4 has reached mass shipment, and new supply still requires construction and manufacturing execution. Those conditions support a positive outlook. High margins also make pricing, yields and investment returns especially consequential.

A 118% net margin cannot be extrapolated as operating profitability

The preliminary release also reported KRW93.9226 trillion of net income, a margin of roughly 118%. Net income above revenue cannot represent the recurring operating margin earned from product sales. Explaining its full difference from operating income requires reconciliation of non-operating and tax items. It should not be used to infer the return on each memory device sold.^[1]

Reported revenue less operating profit is approximately KRW18.7761 trillion. That is the aggregate cost-and-expense difference at this profit level, not manufacturing cost alone or a separation of fixed and variable costs. Strong revenue and earnings establish a profitable quarter without independently identifying contributions from volume, price, mix and cost reduction.

The distinction matters for the cycle. Improvements driven by mix and manufacturing efficiency may prove more durable than those driven predominantly by temporary price increases. The company identifies both types of support. Analysis should preserve both possibilities rather than remove memory cyclicality from an AI narrative.

The company reported quarter-end cash and equivalents of KRW88 trillion, total debt of KRW18.6 trillion and net cash of approximately KRW69.4 trillion.^[1] A stronger balance sheet expands investment and distribution choices. It does not guarantee that new projects earn the same margin. Existing financial resources and prospective investment returns need separate assessment.

What would a 10% price decline do to operating income?

A static sensitivity illustrates pricing exposure. Hold shipment volume and product mix constant, reduce realized prices uniformly by 10%, and temporarily hold costs and expenses fixed. Revenue falls approximately KRW7.9319 trillion and operating profit declines from KRW60.5426 trillion to roughly KRW52.6107 trillion, down about 13.1%. This is not a pricing forecast or an assertion that every product shares one pricing mechanism.

A 20% price decline under the same assumptions would produce operating profit of approximately KRW44.6789 trillion, down about 26.2%. Mix improvement, lower costs or volume growth could offset the effect; inventory charges or idle capacity could worsen it. The calculation establishes the financial sensitivity of the current earnings base.

Manufacturing yield works through a different mechanism. If the same inputs produced 110 usable units instead of 100, unit input cost would fall approximately 9.1%, assuming unchanged total input cost. Advanced memory also requires packaging, testing and qualification, so a wafer-level improvement is not the same as the reduction in finished-product cost. No actual SK hynix yield is assumed.

Price and yield changes also arrive on different schedules. Customer prices may reset at contract renewal, while manufacturing improvements emerge gradually during a ramp. Separating them avoids forecasting annual earnings from one spot-price curve and makes the contribution of technical execution clearer.

What risk do agreements with roughly ten customers remove?

SK hynix said it had finalized long-term agreements with around ten customers and continued multi-year discussions with major buyers.^[1] That provides a clearer customer base for planning. The announcement does not disclose individual prices, quantities, minimum purchases, cancellation rights or compensation. The number of agreements cannot be converted into definite backlog.

An agreement can fix quantities, set a pricing formula, retain adjustment windows or make obligations conditional on qualification. Each structure provides different visibility. If a customer expresses demand without binding payment or procurement, the supplier retains more expansion risk. Minimum purchases and advances may improve financing and production planning. The announcement does not permit a contract-by-contract classification.

Customers also value supply certainty. Missing a critical memory component can delay an expensive accelerator cluster, giving reliable delivery economic value. Product performance and supply consistency can strengthen negotiating power. They do not ensure permanently rising prices: customers still seek alternatives and lower system costs.

Subsequent shipments, receivables, inventory and cash should show what the agreements accomplish. Reliable conversion and reasonable inventory would strengthen the visibility argument. Changed customer schedules against inflexible expansion could reveal less protection than the headline suggests. Risk allocation matters more than the multi-year label.

HBM4 must deliver consistent supply and competitive system economics

The company said HBM4 mass shipments began in the second quarter and would ramp in the second half, while HBM4E sample shipments were completed during the first half.^[1] These describe different generations and stages. Sampling, qualification, mass shipment and revenue recognition are not interchangeable, and next-generation samples do not establish current sales.

HBM provides bandwidth within space and power constraints. A higher-priced memory component does not necessarily raise the cost of useful system output: improved bandwidth can reduce accelerator waiting and increase completed work. That depends on whether the actual workload is memory-constrained. A specification improvement cannot establish an equivalent economic benefit by itself.

More stacking, complex packaging and demanding qualification also increase manufacturing challenges. Selling prices and production costs can rise together. Stable yields, timely delivery and acceptable customer pricing support earnings; a generation name does not. Company statements about performance, power and cost remain company disclosures rather than independent comparisons across every platform.

The quarterly release supports the reported sampling and production progress, but does not disclose a complete customer allocation, memory content per accelerator or supply share.^[1] Participation in a new platform still requires customer qualification and reliable delivery. Earnings should be judged through actual shipments and retained profit.

Expansion spends first; returns wait for equipment and yields

The company is accelerating M15X production and plans to expand capacity after the Yongin phase-one cleanroom opens in early 2027. P&T7 advanced packaging, the M17 NAND base and later clusters are to proceed in phases according to demand and investment efficiency.^[1] A cleanroom opening is an infrastructure milestone, followed by tool installation, process commissioning and output ramp.

Timing matters for returns. Assume a project has received KRW1 trillion of investment and requires a 10% annual pre-tax cash return. A stable full year would need KRW100 billion of corresponding cash contribution. Half a year of operation would provide only KRW50 billion at the same rate. This simplified timing example excludes construction-period cash, financing and taxes and is not company guidance.

The type of expansion matters as well. HBM wafers, advanced packaging and testing need not increase conventional DRAM supply at the same pace, and production lines cannot be switched without cost. Conversely, attractive returns draw investment. Competitors releasing capacity together could reduce scarcity. Industry supply discipline must be judged from investment and deliverable output across producers.

Constraints in tools, electricity, land and staffing may extend favorable pricing while also increasing the supplier’s own costs. Treating bottlenecks only as a pricing benefit misses that burden. Investors need to know whether supply arrives when customers require it and whether the incremental assets still earn adequate returns after prices normalize.

Long-term value rests on competitiveness after peak margins

The positive case is that customers continue reserving supply, HBM4 scales successfully, high-value sales expand and investment remains phased. Earnings visibility could then extend beyond a short inventory recovery, with financial resources supporting the next generation of technology and manufacturing. The current quarter supports that interpretation.

The countercase is that pricing already accounts for substantial earnings and that demand timing and industry supply can change. Customers can defer systems, qualification can take longer and competitors can narrow technical gaps. Expansion planned around peak margins may earn much less when the new capacity arrives.

Valuation should therefore avoid treating a 76% operating margin as permanent, and should not extrapolate a 118% net margin. Without a complete set of cycle pricing, investment and equity assumptions, this article does not derive a price target. The relevant long-term assets are product position, cost capability and customer relationships that can support normalized cash generation.

Three disclosures would most change the assessment: stable HBM4 shipments and earnings contribution; realization of long-term arrangements; and investment growth relative to deliverable supply. Progress in all three would strengthen the long-term AI-memory case. Continued price increases without better delivery and costs would leave cyclical downside more important.