SK Hynix’s 2Q26 Earnings: Record Margins, but the Memory Supply Gap Has Not Closed

SK hynix’s second-quarter results reflect more than another strong quarter for high-bandwidth memory. Pricing power has expanded into server DRAM and enterprise SSDs, while the production capacity intended to relieve the shortage is unlikely to arrive quickly enough to resolve current supply constraints.

What SK hynix Reported

SK hynix reported second-quarter 2026 revenue of KRW 79.3 trillion, operating profit of KRW 60.5 trillion, and net income of KRW 93.9 trillion. Revenue increased 51% from the preceding quarter and 257% from a year earlier. Operating profit rose 61% sequentially and 557% year on year.

KRW trillion	2Q26	1Q26	Sequential change	2Q25	Annual change
Revenue	79.3	52.6	51%	22.2	257%
Operating profit	60.5	37.6	61%	9.2	557%
Operating margin	76%	72%	+4 percentage points	41%	+35 percentage points
Net income	93.9	40.3	133%	7.0	1,242%

Source: SK hynix, 2Q26 Financial Results, July 29, 2026. Figures are preliminary and based on consolidated K-IFRS.

The scale of the earnings acceleration becomes more visible when the first half is compared with the previous full year. First-half 2026 revenue reached approximately KRW 131.9 trillion, about 36% above SK hynix’s full-year 2025 revenue of KRW 97.1 trillion. First-half operating profit of approximately KRW 98.2 trillion was more than twice the KRW 47.2 trillion recorded for all of 2025.

This comparison does not constitute a full-year forecast. It shows how rapidly higher memory prices and a richer product mix have moved through the income statement.

A Record Quarter, but Not an Unqualified Beat

The headline results were exceptionally strong, but market expectations had risen even faster. Revenue and operating profit were below the highest consensus estimates immediately before the announcement. Some higher-value shipments were deferred into the second half, while blended DRAM pricing increased less than the most optimistic expectations.

This distinction matters. A consensus miss in a rapidly rising earnings cycle is different from an operational contraction. SK hynix still produced sequential growth of more than 50% in revenue and more than 60% in operating profit. The issue was not deteriorating profitability. It was the distance between already exceptional reported results and expectations that had been repeatedly revised upward.

The market therefore appears to be evaluating two separate questions. The first is whether SK hynix can continue to increase earnings. The second is whether those increases can continue to exceed an unusually demanding expectations curve.

How Large Is the Memory Semiconductor Shortage?

There is no single shortage percentage for the entire memory market

“Memory shortage” is often treated as one industry-wide figure. In practice, the balance varies by product, customer, specification, qualification status, and contractual arrangement. HBM, server DRAM, mobile DRAM, mature-node DDR4, enterprise SSDs, and consumer NAND cannot be reduced to one interchangeable pool of supply.

An exact total DRAM shortage percentage is not available from a transparent, authoritative public dataset. Public evidence nevertheless supports the conclusion that the market remains undersupplied.

Indicator	Verified evidence	Interpretation
2026 NAND supply balance	TrendForce estimates a global supply deficit of 4–5%.	NAND demand is expected to exceed supply at the total-market level during 2026.
3Q26 conventional DRAM prices	Contract prices forecast to rise 13–18% sequentially.	Suppliers retain pricing power despite weakness in some consumer applications.
3Q26 NAND prices	Contract prices forecast to rise 10–15% sequentially.	The shortage is moderating in some segments but has not disappeared.
Server DRAM	TrendForce describes the market as undersupplied in the third quarter.	AI servers and general-purpose servers continue to compete for constrained capacity.
Peer-company outlook	Micron states that DRAM and NAND demand significantly exceeds industry supply and expects tightness beyond 2027.	The constraint is being observed across major suppliers rather than by SK hynix alone.
SK hynix customer demand	The company states that customer demand exceeds its supply capability and has completed LTAs with around ten customers.	Customers are seeking supply certainty beyond normal quarterly procurement cycles.

Sources: TrendForce, July 21, 2026; TrendForce, July 3, 2026; Micron Fiscal Q3 2026 Earnings Presentation; and SK hynix’s second-quarter earnings release.

The 4–5% NAND estimate is the clearest publicly disclosed measure of the shortage’s scale. It should not be applied mechanically to DRAM or HBM. A small percentage deficit can also produce a disproportionately large pricing effect in a market with low inventories, long qualification cycles, and limited short-term substitution.

Why Supply Is Still Behind Demand

HBM consumes more production resources than conventional DRAM

HBM growth does not simply add another product to a fixed pool of output. Higher-stack products require more DRAM dies, advanced packaging, through-silicon vias, additional process steps, and tighter yield control. Each generation also raises technical and qualification requirements.

Capacity assigned to HBM can therefore constrain the supply of conventional server and client DRAM. This produces a second-order effect: AI demand strengthens HBM pricing directly while also reducing the capacity available for other DRAM categories.

New fabs do not translate immediately into qualified output

SK hynix is accelerating M15X and preparing to expand production after the first Yongin cleanroom begins operating in early 2027. Micron is also expanding capacity, but several of its major new facilities are expected to begin meaningful output from 2027 onward.

The industry is therefore increasing capital expenditure while remaining short of near-term supply. Construction, equipment installation, process qualification, yield improvement, packaging expansion, and customer certification create a lag between investment decisions and commercially usable output.

This lag explains the apparent contradiction between record capital spending and persistent scarcity. Spending is increasing because the market is undersupplied. The same spending may eventually reduce the shortage, but it cannot immediately satisfy orders already being negotiated.

Capacity is being allocated rather than distributed evenly

Suppliers are prioritizing HBM, server DRAM, high-capacity modules, and enterprise SSDs. Consumer applications face a different market. PC and smartphone manufacturers are becoming more sensitive to component costs, but weaker consumer demand does not automatically release compatible supply for AI systems.

Memory products differ by density, speed, endurance, interface, firmware, packaging, and customer qualification. Capacity cannot always move seamlessly between end markets. The result is a fragmented shortage in which selected consumer products can weaken while server-oriented products remain scarce.

Why SK hynix May Still Have Operational Upside

The second half includes both volume and product-mix drivers

The second-quarter result was driven primarily by pricing and mix rather than by a proportional increase in physical shipments. Conference-call metrics summarized in the materials reviewed for this analysis indicate that DRAM bit shipments rose by a high-single-digit percentage sequentially while blended DRAM prices increased by close to 30%. NAND bit shipments rose in the mid-teens, while pricing increased by more than 50%.

This pattern matters because SK hynix expects stronger shipment growth in the second half. HBM4 production is ramping, advanced conventional DRAM products are expanding, and the timing of some high-value shipments moved from the second quarter into later periods.

Further earnings growth therefore does not require another price increase of the same magnitude as the first half. A combination of moderate pricing, higher bit shipments, and a richer mix could continue to lift revenue even as the rate of price inflation slows.

HBM4 was only beginning to contribute

SK hynix began mass shipments of HBM4 during the second quarter and plans a broader ramp in the second half. The company has also supplied HBM4E samples to customers, with volume production targeted for 2027.

The commercial significance is not limited to bandwidth. HBM competitiveness increasingly depends on yield, power efficiency, thermal performance, packaging integration, and the ability to deliver qualified volumes on schedule. Customers face a high system-level cost when an HBM component fails or arrives late.

SK hynix’s established production history in HBM3 and HBM3E provides an advantage that cannot be inferred from laboratory specifications alone. The limitation is that product transitions can still create shipment timing risk, while competitors are investing aggressively to close the gap.

NAND is becoming more relevant to AI economics

AI infrastructure demand is expanding beyond accelerator-attached HBM. Inference workloads require larger context windows, persistent data, retrieval systems, and storage for key-value caches. This broadens demand toward server DRAM and enterprise SSDs.

SK hynix reported strong growth in high-capacity enterprise SSDs and is positioning NAND as part of a wider AI memory hierarchy. This matters because NAND has historically generated lower and more volatile returns than premium DRAM products.

A sustained enterprise SSD mix shift could improve the strategic and financial role of NAND. The limitation is that consumer NAND remains exposed to weak PC and smartphone demand, while additional Chinese supply could alter the market balance from 2027 onward.

What Long-Term Agreements Actually Change

SK hynix has completed long-term agreements with around ten customers and is discussing additional contracts. These agreements are intended to provide multi-year supply stability, support joint product development, and improve the efficiency of investment and production planning.

LTAs should not be interpreted as uniform fixed-price contracts. Terms can vary by customer, product, duration, committed volume, deposits, and pricing mechanisms. Management has also indicated that it intends to retain enough flexibility to participate in additional market demand rather than contract its entire output in advance.

The main financial effect is therefore likely to appear in earnings duration rather than in a permanently fixed peak margin. Greater volume visibility can reduce the probability of uncontrolled expansion, improve capacity planning, and support a higher earnings floor during the next downturn.

LTAs also create a two-tier market. Contracted customers gain priority and supply visibility. Uncontracted buyers must compete for the remaining capacity, which may preserve pricing tension in the open market.

The limitation is that long-term contracts do not remove cyclicality. They can contain price bands or adjustment mechanisms, and customers may seek protection if market prices fall sharply. Concentrating more supply among a small number of large AI customers can also increase counterparty and bargaining-power risk.

Earnings Quality: Operating Profit Matters More Than Net Income

SK hynix’s net income exceeded revenue in the second quarter, producing a net margin of 118%. This is not a sustainable operating relationship.

The company recorded substantial non-operating investment gains during the period. Public reporting has connected much of the increase to investment assets associated with Kioxia, although the official earnings release did not provide a complete breakdown of the gain.

For evaluating the underlying business, operating profit of KRW 60.5 trillion is therefore more decision-useful than net income of KRW 93.9 trillion. Operating profit captures the economics of memory pricing, product mix, manufacturing costs, and shipments. The investment gain improves the balance sheet but should not be annualized as recurring earnings.

The balance sheet nevertheless strengthened materially. SK hynix reported cash and cash equivalents of KRW 88.0 trillion, debt of KRW 18.6 trillion, and a net cash position of KRW 69.4 trillion at the end of the quarter.

This financial capacity provides room for capacity investment and shareholder returns. It does not eliminate the need for capital discipline. The company expects 2026 capital expenditure to be in the upper range of approximately KRW 40 trillion, while much larger long-term infrastructure programs extend well beyond the current year.

Capital Allocation Is Becoming Part of the Investment Case

SK hynix’s historical valuation reflected more than memory-price volatility. Investors also discounted the industry because large capital expenditures often followed strong prices, eventually contributing to oversupply.

The company is now attempting to link investment more directly to customer demand, LTAs, and staged capacity deployment. If maintained, this approach could reduce the probability that record profits are immediately converted into excessive undifferentiated supply.

SK hynix implemented approximately KRW 14.3 trillion of shareholder returns based on its 2025 results through additional dividends and treasury-share cancellation. Its 2025–27 shareholder-return framework also allocates half of accumulated free cash flow to potential shareholder returns while maintaining a minimum annual dividend of KRW 1,500 per share.

Management has indicated that additional return measures are under review. The amount, timing, and continuity have not yet been finalized. The relevant question is therefore not simply whether capital returns rise, but whether they remain compatible with the company’s large production and infrastructure commitments.

Competitive Position: From HBM Supplier to Full-Stack Memory Provider

SK hynix’s strongest competitive position remains HBM. Its advantage combines early development, manufacturing yields, customer qualification, advanced packaging, and experience delivering large volumes across several product generations.

The second-quarter result also suggests that its earnings base is broadening. Server DRAM, SOCAMM-class low-power memory, enterprise SSDs, and high-layer NAND are becoming more important alongside HBM.

This wider portfolio can be strategically valuable in AI systems. Accelerator performance depends not only on HBM bandwidth but also on system memory, storage latency, data movement, power consumption, and total cost of ownership.

The competitive limitation is that leadership is not static. Samsung Electronics and Micron are expanding HBM, advanced DRAM, NAND, and packaging capacity. Customer qualification cycles slow competitive change, but they do not prevent it. SK hynix must maintain yield, performance, cost, and delivery execution as product generations become more complex.

Valuation Sensitivity: The Debate Is About Duration, Not the Peak Quarter

A 76% quarterly operating margin is unlikely to serve as a permanent normalized assumption. The more useful valuation question is how far earnings eventually decline from the current level and how long elevated profitability lasts.

Scenario	Operating assumptions	Financial interpretation	Confirming indicators
Extended shortage	Supply remains constrained through 2027, HBM4 ramps successfully, eSSD demand expands, and LTAs support volume visibility.	Margins normalize gradually, while the cycle-average earnings floor rises above previous memory cycles.	Continued contract-price increases, high HBM utilization, stable LTA commitments, and delayed capacity relief.
Orderly normalization	New capacity enters progressively, price increases moderate, and consumer weakness offsets some AI demand.	Revenue growth slows and margins decline, but earnings remain supported by product mix and contractual visibility.	Stable shipments, narrowing spot premiums, and more balanced DRAM and NAND inventories.
Accelerated downturn	AI capital expenditure slows, customers reduce inventories, Chinese supply expands faster, and new fabs ramp smoothly.	Memory prices fall while depreciation and capital expenditure remain high, compressing free cash flow and valuation multiples.	Contract-price declines, LTA renegotiations, lower HBM utilization, and rising supplier inventories.

The current results strengthen the case that SK hynix’s earnings floor may be structurally higher than in earlier cycles. They do not establish that the present margin is permanent or that security prices must rise in line with accounting earnings.

Editor S’s Interpretation

1. The shortage is broadening from an HBM bottleneck into a memory-system constraint

Evidence: SK hynix’s growth is increasingly supported by HBM, server DRAM, enterprise SSDs, and high-capacity memory modules. Micron and TrendForce also describe supply constraints across both DRAM and NAND.

Mechanism: Agentic AI increases memory capacity, bandwidth, and storage requirements across the system rather than only around the accelerator. Capacity diverted toward advanced products reduces availability elsewhere.

Interpretation: SK hynix’s earnings exposure is becoming broader than a single HBM product cycle. This improves revenue diversity within memory, although it does not diversify the company away from semiconductor cyclicality.

Time horizon and indicators: The thesis is most relevant through 2027. Enterprise SSD growth, server DRAM prices, HBM4 shipments, and general-purpose server demand are the main confirming indicators.

Limitation: Software optimization and memory compression could lower memory required per workload. Consumer demand destruction could also offset part of the AI-driven increase.

2. The strategic significance of LTAs is a higher trough, not a higher peak

Evidence: SK hynix has completed agreements with around ten customers and continues negotiating additional contracts. Peers are adopting similar long-term structures.

Mechanism: Committed demand and financial safeguards improve production planning and reduce incentives for speculative capacity expansion. Customers gain supply security while suppliers obtain better demand visibility.

Interpretation: The agreements may extend earnings duration and reduce the severity of the next downturn. They are less relevant to explaining whether the current quarter represents the absolute peak margin.

Time horizon and indicators: The effect should become clearer over several years. Contract renewals, committed capacity, customer deposits, and capital-spending discipline are the relevant evidence.

Limitation: Contract pricing may adjust with market conditions, and customers can seek renegotiation if economics deteriorate substantially.

3. Higher capital expenditure can prolong the shortage before resolving it

Evidence: SK hynix and its peers are expanding capacity, but meaningful new fab output is concentrated from 2027 onward.

Mechanism: Fab construction, equipment installation, process development, packaging capacity, yield improvement, and customer qualification create long lead times.

Interpretation: Rising capital expenditure is not evidence that immediate oversupply is inevitable. In the near term, it confirms that current facilities are insufficient for expected demand.

Time horizon and indicators: M15X, Yongin, Micron’s new facilities, packaging output, and wafer-start growth should be monitored through 2027 and 2028.

Limitation: Several projects reaching commercial output simultaneously could eventually produce a faster supply response than customers require.

4. The earnings announcement supports operational upside, but not an automatic security-price conclusion

Evidence: Reported revenue and operating profit reached records but did not exceed the highest market expectations. Net income was also materially affected by non-operating gains.

Mechanism: Share prices reflect future earnings expectations, valuation multiples, capital allocation, and positioning rather than reported profit alone.

Interpretation: The second-quarter result supports the view that SK hynix can continue to grow operational earnings. Whether that growth creates additional equity value depends on how much duration and profitability are already embedded in market expectations.

Limitation: A detailed reverse-valuation analysis requires an up-to-date share price, consensus earnings estimates, and assumptions that can change materially after publication.

What to Watch

• DRAM and NAND contract prices: The direction of prices matters more than whether quarterly increases slow from unusually high levels.
• HBM4 shipment execution: Volume, yield stability, customer qualification, and 2027 pricing negotiations will determine whether technology leadership converts into durable earnings.
• Enterprise SSD growth: Continued expansion would confirm that AI memory demand is broadening beyond HBM.
• LTA structure: Contract duration, committed volumes, financial safeguards, and the proportion of capacity reserved under long-term arrangements are more relevant than the number of agreements alone.
• M15X and Yongin timing: Faster ramps would support shipments but could also bring the industry closer to supply normalization.
• AI infrastructure spending: Hyperscaler capital expenditure, AI-service revenue, server shipments, and GPU utilization are key demand indicators.
• Capital allocation: The balance between large infrastructure commitments, balance-sheet liquidity, and additional shareholder returns remains unresolved.

Risks and Thesis Breakers

The strongest challenge to the current interpretation would be a simultaneous weakening of AI infrastructure demand and acceleration of memory supply. Either event alone could moderate pricing. Together, they could create a conventional inventory-led downturn.

Persistently high memory prices can also reduce demand. PC and smartphone vendors may lower specifications, delay purchases, or raise retail prices. AI customers may redesign systems, compress models, reduce memory per workload, or shift expenditure between infrastructure layers.

Competitive pressure remains another risk. Samsung Electronics and Micron are expanding advanced memory and packaging capabilities. Chinese producers are increasing NAND and mature DRAM capacity, although their ability to enter the most demanding HBM applications remains limited in the near term.

Execution risk rises with technical complexity. HBM4 and HBM4E require high yields, advanced packaging, power efficiency, thermal control, and close integration with customer systems. Qualification delays can shift revenue between periods even when final demand remains intact.

Finally, the company’s planned investment scale is large. If supply arrives after demand expectations have been revised downward, depreciation and fixed costs could amplify the next earnings contraction.

Sources and Methodology

This analysis prioritizes SK hynix earnings releases and public company disclosures. Micron’s earnings materials were used to compare industry supply conditions across major suppliers. TrendForce’s public memory-market updates were used for independent pricing and supply-balance estimates.

User-supplied brokerage research was reviewed as secondary background to identify disputed assumptions, earnings-call details, and issues requiring public verification. No proprietary charts, target prices, report screenshots, or report language have been reproduced.

The first-half comparisons were calculated independently from SK hynix’s reported quarterly and annual results. Forecasts and scenarios are labeled separately from reported facts.

About the Author

Editor S writes independent analysis for Sector Foundry, focusing on companies, industries, technologies, and global value chains.