The First Double Miss: A Record ₩79 Trillion Quarter Falls Short on Both Lines as DRAM ASP Growth Halves. Maintaining Hold Through a 52% Drawdown

Results vs. Consensus

Q2 2026 Scorecard

Year-Over-Year and Quarter-Over-Quarter (K-IFRS)

First-half 2026 revenue reached ₩131.895tn, crossing ₩100tn for the first time in company history at the half-year mark, with first-half operating profit of ₩98.153tn. On any absolute measure this remains an extraordinary business at an extraordinary moment.

Quality of the Quarter

Revenue: The 4.9% shortfall is entirely a pricing shortfall. DRAM and NAND bit shipments both landed within guidance (DRAM +high single digit, NAND +mid teens), so volume executed as promised. What undershot was ASP: roughly +30% QoQ in DRAM against the +mid-60% of Q1, and +mid-50% in NAND against +mid-70%. Both remain enormous sequential price increases in any normal context. Both are also roughly half the prior quarter's rate, and in a cycle whose entire earnings expansion has been price, the rate of price change is the variable that matters.

Margins: The 76% operating margin is a new record and, on its face, refutes the argument that Q1's 72% could not be exceeded. The refutation is narrower than it looks. Margin expanded 4pp while ASP growth halved, which means the incremental margin came substantially from the lower-cost inventory flowing through the P&L and from mix rather than from fresh pricing power. That is a lagging benefit. When ASP growth decelerates further, as guidance implies for bits and management concedes for mix timing, the cost tailwind is not large enough to hold 76%.

EPS and net income: The lowest-quality headline we have seen from this company. Net income of ₩93.923tn exceeds operating profit by ₩33.4tn, driven by ₩63.3tn of gains on the sale and revaluation of investment assets and ₩1.1tn of FX. Pre-tax profit was ₩122.7tn, implying roughly ₩28.8tn of tax. We flagged non-operating inflation in Q3 2025 (₩3.3tn) and again in Q1 2026 (₩9.9tn). At ₩63.3tn this is a different order of magnitude, and it is a one-time portfolio event rather than a recurring valuation mark.

Segment Performance

Product-Line Detail — Q2 2026 vs. Q1 2026

DRAM — Volume Delivered, Price Undershot

DRAM bit shipments rose a high single-digit percentage, in line with the guidance given in April, on expanded sales of HBM3E and AI server DRAM under constrained supply. Server LPDDR products including SOCAMM2 grew significantly, and 1c-nm-based shipments began in earnest. ASP rose approximately 30% sequentially.

"In DRAM, amidst limited supply capabilities, we expanded sales centered on HBM3E and AI server DRAM products, achieving high single-digit percent increase in bit shipments in line with our guidance… ASP rose by approximately 30%, driven by the continued price strength in conventional DRAM." — Song Hyeon-jong, President of Corporate Center

The comparison that matters is against the prior quarter's mid-60% ASP gain. Halving the rate of price increase while holding volume is the classic signature of a pricing cycle passing its steepest point. Management's explanation, that some high-value shipments were pushed into the second half and that portfolio mix diluted blended ASP, is credible and was volunteered rather than extracted.

Assessment: Take the mix explanation at face value and the picture still resolves the same way. If high-value volume was deferred rather than lost, H2 blended ASP recovers somewhat, which is what management guides. But a business whose blended ASP depends on the timing of a handful of high-value shipments has less pricing momentum than one where every product line is repricing upward at once, which is what Q1 looked like.

NAND — The AI Storage Story Turns Real

NAND bit shipments rose a mid-teen percentage off the low Q1 base, in line with guidance, with enterprise SSD expansion the driver. ASP rose a mid-50% percentage on strength across all products. The underlying business mix shifted decisively toward AI storage.

"Our enterprise SSD revenue increased twofold from that of the previous quarter, and Solidigm's revenue from high-capacity enterprise SSDs of 30 TB and above also expanded more than threefold from that of last quarter." — Song Hyeon-jong, President of Corporate Center

Doubling eSSD revenue and tripling high-capacity Solidigm revenue in a single quarter is the first hard evidence in this coverage arc that NAND has an AI demand driver rather than merely an AI-adjacent pricing tailwind. Inference workloads, KV-cache offloading and AI data lakes are pulling storage into the memory hierarchy in a way training never did.

Assessment: Structurally the best news in the quarter, and a genuine upgrade to the quality of the NAND franchise, which we have consistently weighted as the higher-beta and faster-correcting leg. It does not offset the near-term math: Q3 NAND bit guidance drops to low single digit from mid teens, and a mid-50% ASP gain still sits far above anything sustainable.

HBM — HBM4 Ships, and the Competitive Question Returns

HBM4 entered mass-production shipment during the quarter with a full ramp planned for the second half, at yields management describes as approaching those of mature HBM3E. HBM4E samples were delivered to a major customer in the first half, with volume production targeted for 2027. Beyond that, the company disclosed IHBM, a packaging approach that integrates cooling elements to cut thermal resistance by more than 30% for future generations such as HBM5.

"Building on this foundation, we began mass production of HBM4 for key customers in the second quarter, with the yield and quality today nearing the levels of HBM3E, which is already in the maturity stage." — Song Hyeon-jong, President of Corporate Center

Notably, an analyst opened the HBM discussion by observing that competitors "have made rapid progress in HBM recently," the first time in our coverage that the competitive premise has been put to management this directly. The response leaned on an accumulated bundle of time-to-market, yield, quality and customer trust rather than on any single specification.

Assessment: HBM remains the earnings floor and the reason this is Hold rather than something more bearish. Reaching HBM3E-class yields on a first-generation HBM4 ramp is a real execution win. But the framing of the moat as un-replicable accumulated capability, rather than as a measurable performance lead, is the argument a leader makes when the gap is narrowing. We now track competitive qualification as an active risk rather than a latent one.

Key Topics & Management Commentary

Overall management tone: Composed and unusually well-prepared for scrutiny, with the defensive weight shifted from pricing to demand. Where the Q1 call opened with management rebutting a peak-out question about spot prices, this call opened with management rebutting the proposition that AI infrastructure spending itself is slowing, and closed with a disclosure refusal on shareholder returns. Management was most convincing on supply, LTAs and HBM execution, and least convincing on why blended ASP undershot, where the answer relied on shipment timing that cannot be verified until the H2 results arrive.

1. Another Record, and the First Miss

"Second quarter revenue increased by 51% quarter-on-quarter and 257% year-on-year, reaching KRW 79.3 trillion, marking a record high revenue following the previous quarter." — Song Hyeon-jong, President of Corporate Center

Revenue, operating profit, operating margin and net profit all set company records, and first-half revenue crossed ₩100tn for the first time. Consensus, however, sat at roughly ₩83.4tn of revenue and ₩64tn of operating profit, so the record quarter arrived below the bar.

Assessment: The gap between "best ever" and "below expectations" is the entire story of this print, and it is the condition under which cyclical stocks reprice. Absolute results stopped being the operative variable once the market began underwriting a trajectory rather than a level.

2. The Margin Record Nobody Focused On

"Operating margin also improved by 5 percentage points from that of the previous quarter to record 76%, achieving all-time highs for both operating income and operating margin." — Song Hyeon-jong, President of Corporate Center

The operating margin reached 76%, from 72% in Q1, alongside an 81% EBITDA margin on ₩4.0tn of depreciation. (On the rounded margins the company publishes, the step is 4 percentage points; on the unrounded figures it is 4.8, which management rounded up to 5 on the call. Tables in this note use the 4-point step from the reported margins.) We wrote in April that 72% was a blow-off from which the only direction was compression. One quarter later the margin went up.

Assessment: We were wrong on the timing of margin compression and we should say so plainly. We were not wrong on the mechanism: margin expanded on lagging cost benefit and mix while the pricing engine driving it decelerated by half. A margin that rises on cost while its revenue driver slows is a margin borrowing from the next few quarters.

3. A ₩63 Trillion Investment Gain Rewrites the Net Line

"Net non-operating profit reached KRW 62.2 trillion, including foreign exchange-related net gains of KRW 1.1 trillion due to the rising exchange rate and gains from the sale and valuation of investment assets totaling KRW 63.3 trillion." — Song Hyeon-jong, President of Corporate Center

Net non-operating income of ₩62.2tn lifted pre-tax profit to ₩122.7tn and net profit to ₩93.9tn, a 118% net margin. The investment-asset gains relate substantially to the disposal of the company's Kioxia holding, which an analyst referenced directly in the closing question.

Assessment: Monetizing the Kioxia stake near a memory-cycle high is good capital allocation and a genuine addition to the cash pile. It is not earnings power. The disclosure was clear and unprompted, to management's credit, but the resulting headline EPS is a number that should never appear in a valuation multiple.

4. DRAM Pricing: The Mix-and-Timing Explanation

"In the second quarter, shipments of some high value add products were pushed back into the second half, and changes in the product portfolio appeared to have affected our blended ASP." — Song Hyeon-jong, President of Corporate Center

Asked directly why DRAM ASP growth fell short of expectations, management pointed to deferred high-value shipments and portfolio mix, and guided that both factors ease in the second half as HBM4 ramps and 1c-nm conventional DRAM volume increases.

Assessment: This is the most important disclosure on the call and the one carrying the least verification. If it is timing, H2 blended ASP recovers and the cycle extends. If it is the leading edge of price normalization dressed as mix, Q3 will show it. We treat the explanation as provisionally credible and make the H2 blended-ASP path the single highest-priority signpost.

5. Long-Term Agreements Harden Into Contracts With Deposits

"In addition to long-term volume commitments, the agreements include mechanisms such as deposits that can strengthen contract implementation and demand visibility." — Song Hyeon-jong, President of Corporate Center

SK hynix has concluded LTA negotiations with around 10 customers including its key accounts, on terms normally running about five years, with non-uniform pricing mechanisms designed to dampen volatility and financial deposits to enforce commitment. Management described the objective as improving downside resilience while retaining upside flexibility.

Assessment: The most consequential structural development in four quarters of coverage. Customer deposits convert a demand forecast into a financial obligation, which is precisely the mechanism prior memory cycles lacked. This is the thesis pillar we said would need proving through a downturn, and the downturn is now arguably beginning, so the test starts. It raises our estimate of the trough, though it does not tell us where the trough sits.

6. CapEx Steps Up to the High-₩40 Trillion Range

"Due to the schedule acceleration and investment expansion, our 2026 CapEx is expected to reach the high KRW 40 trillion range." — Song Hyeon-jong, President of Corporate Center

The company is pulling forward M15X mass production, opening the Yongin Phase 1 cleanroom in early 2027, and has announced P&T7 advanced packaging, the M17 NAND base and a new domestic semiconductor cluster, all to be executed in phases against demand visibility.

Assessment: The bear mechanism, quantified. Industry-wide capacity commissioned at a margin peak is how every prior memory cycle seeded its own correction. Management's counter, that expansion is phased and matched to contracted demand, is more credible now that LTAs carry deposits. The risk is not this year's spend; it is what arrives in 2028 if AI capex growth moderates while three suppliers build simultaneously.

7. The Nasdaq ADR Lands, and the Multiple Does Not Re-Rate

"On July 10th, we successfully listed our ADRs on the Nasdaq market in the United States. This ADR offering was the largest ever for a foreign company's IPO in the U.S." — Song Hyeon-jong, President of Corporate Center

The ADR listed on July 10 under SKHY. From July 30, ADRs convert freely into ordinary shares, while conversion in the opposite direction is constrained by a regulatory filing process and a cap set at 17,790,000 shares, equal to the offering size, with no decision made on raising it.

Assessment: We called the ADR the most credible non-cyclical re-rating catalyst in the story. It executed flawlessly and delivered no re-rating, because it listed roughly three weeks after the cycle high and into the start of a sector-wide de-rating. The asymmetric fungibility is worth watching: with ADRs convertible into shares but shares constrained from converting into ADRs, the structure permits arbitrage in only one direction.

8. Management Rebuts the AI Capex Slowdown Narrative

"We view these developments not as a sign of AI investment slowdown, but more as a transition towards higher utilization of the AI infrastructure that has already been built at scale and as well as accelerated monetization efforts." — Song Hyeon-jong, President of Corporate Center

The opening question raised data-center leasing by big tech and the emergence of more efficient AI models as evidence that AI infrastructure investment could slow. Management rejected both readings, arguing efficiency broadens adoption rather than reducing infrastructure demand, and that major customers are still requesting more supply than can be delivered.

Assessment: The same structure as Q1, one level further up the value chain. Then the first question was whether memory pricing had peaked; now it is whether the demand behind that pricing is peaking. Management's Jevons-style argument is intellectually sound and, so far, empirically supported by customer behaviour. But the debate has migrated from price to demand, and that migration is itself the market's message.

9. Supply Stays Tight, Demand Growth Framed in the Mid-20s

"On the supply side, however, it appears difficult for the supply-demand balance to improve meaningfully in the near term." — Song Hyeon-jong, President of Corporate Center

Management projects DRAM demand growth in the mid-20% range and NAND in the high teens under constrained supply, citing advanced-process complexity for HBM and AI server memory plus new-fab lead times. PC and mobile are experiencing temporary sales adjustments because those customers cannot secure memory, with recovery expected as shortage eases.

Assessment: The supply argument remains the strongest plank in the bull case and we do not dispute it. Worth noting is what it implies: PC and mobile volumes are being suppressed by price and availability, which is latent demand at lower price points. When supply loosens, that demand returns at a lower ASP, which supports volume and not margin.

10. Shareholder Return Deferred Behind the ADR

"While we cannot provide the specifics regarding the format, size, or timing of any additional shareholder return, we do intend to communicate our plans to the market within the year." — Song Hyeon-jong, President of Corporate Center

With cash and short-term investments at ₩88tn and net cash at ₩69.4tn at June 30, after the Kioxia proceeds but before the July ADR raise, management acknowledged strong market interest in incremental returns but declined to specify anything, citing regulatory and procedural restrictions tied to the ADR offering.

Assessment: A real constraint rather than an evasion, but the effect on the stock is the same. In April the company had committed to developing additional return measures within 2026 including buybacks and cancellation; three months and a ₩69.4tn net cash position later, there is still nothing concrete. With the shares 52% off their high, a decisive buyback is the most direct lever management holds and it is currently unavailable to them.

Guidance & Outlook

SK hynix guides bit shipments and frames pricing qualitatively. It gives no revenue or margin guidance. For Q3 2026:

"As HBM4 shipments ramp up in earnest and the 1c nm conventional DRAM shipment increases, we expect bit growth in the second half to be higher than the level in the first half." — Song Hyeon-jong, President of Corporate Center

Implied ramp: Volume growth continues but decelerates in aggregate. DRAM bits accelerate slightly to about 10%, while NAND bits fall from mid teens to low single digit, a deceleration of roughly 12 percentage points on the leg that just doubled its enterprise SSD revenue. Combined bit growth is therefore lower in Q3 than Q2, which places the entire burden of sequential revenue growth on ASP at the exact moment ASP growth has halved.

Guidance style: Consistent with the company's pattern of guiding bits conservatively and delivering in line, which it did again this quarter on both product lines. The absence of any quantified pricing guide is standard practice here and not a new evasion, but it means the H2 recovery in blended ASP that management describes is an assertion rather than a commitment.

Analyst Q&A Highlights

The call was interpreted consecutively from Korean. Questions and answers below are quoted from the interpreted English rendering and attributed to the analyst and to the named executive who spoke.

Is AI Infrastructure Spending Actually Slowing?

The opening question put the bear case on demand directly, citing data-center leasing by major technology customers and the arrival of more efficient AI models. Management reframed both as evidence of maturation rather than retrenchment.

Q: "Recently we see that some big tech companies are considering leasing data centers and more efficient AI models emerging. As a result, there are some concerns that AI infrastructure investment could slow or even decline. Based on the company's talks with customers, how do you see some of the major CSPs AI infrastructure investment evolving?"
— Jay Kwon, JPMorgan

A: "We view these developments not as a sign of AI investment slowdown, but more as a transition towards higher utilization of the AI infrastructure that has already been built at scale and as well as accelerated monetization efforts."
— Song Hyeon-jong, President of Corporate Center

Assessment: That this was the first question, on a call reporting a 557% year-over-year increase in operating profit, tells you where the debate now sits. Management's answer is the strongest available and it rests on customer conversations we cannot audit. The market pre-emptively voted against it in the four weeks before the print.

Why Did DRAM ASP Undershoot?

The most valuable exchange of the call. An analyst put the pricing shortfall to management explicitly, and the answer supplied the mix-and-timing explanation that the entire H2 outlook now rests on.

Q: "It appears as if DRAM ASP growth in the second quarter fell below market expectations. What are the reasons, and what is the outlook for the second half of the year?"
— S. K. Kim, Daiwa Capital Markets

A: "We manage the sales mix between HBM and conventional DRAM based on customer demand and some medium to long-term product strategy. In the second quarter, shipments of some high value add products were pushed back into the second half, and changes in the product portfolio appeared to have affected our blended ASP."
— Song Hyeon-jong, President of Corporate Center

Assessment: Management did answer, specifically and without deflection, which is more than many would have done. The answer is also unfalsifiable until Q3 prints. Note what it does not claim: at no point did management assert that conventional DRAM contract pricing itself remained on its prior trajectory. The claim is about which products shipped, not about what they fetched.

The Long-Term Agreement Architecture

A recurring line of questioning sought specifics on the LTA framework now that peers have announced comparable arrangements. Management disclosed contract duration and the enforcement mechanism while withholding pricing detail and coverage percentage.

Q: "The company's memory peers have recently concluded and announced LTAs. While it was briefly addressed in the company's briefing, could you also provide more details on SK hynix's LTA framework such as contract term and pricing structure?"
— Sunwoo Kim, Meritz Securities

A: "The LTAs we are discussing with our customers are designed in various forms to be more specific to each customer and their products. While the contract term normally is around five years, specific conditions may vary depending on the customer and product."
— Song Hyeon-jong, President of Corporate Center

Assessment: Five-year terms with deposit mechanisms across roughly 10 customers is a materially different industry structure from the one that produced the 2018 and 2022 downturns. The refusal to quantify what share of volume is covered is the limitation: without that number, the earnings floor these contracts create cannot be sized, only asserted.

Capacity Expansion and the Oversupply Question

An analyst linked the announced capacity build to the market's oversupply anxiety and asked what demand evidence justifies it. Management grounded the answer in customer partnerships and phased execution.

Q: "The company recently presented plans to significantly expand its capacity over the medium to long term. What is the basis for your long-term memory demand outlook that supports this strategy?… Given the increase in the capacity, there are understandably some concerns in the market about potential oversupply. What is the company's view regarding such concerns?"
— Rokho Kim, Hana Securities

A: "Given that our capacity expansion will be executed flexibly in alignment with confirmed customer demand, we do not believe our medium to long-term investment plans will lead to oversupply right away."
— Song Hyeon-jong, President of Corporate Center

Assessment: "Right away" is carrying weight in that sentence. Phased, demand-matched expansion backed by deposit-bearing contracts is a real improvement on prior cycles. It is still capacity, and capacity commissioned at peak margins has never been absorbed as smoothly as it was underwritten.

Competitive Progress in HBM

For the first time in this coverage, an analyst asserted competitor progress as a premise rather than asking whether it might occur. Management answered on capability breadth rather than on any performance metric.

Q: "Now, some believe that competitors have made rapid progress in HBM recently. What is the competitiveness of our HBM4 and the key differentiators that will enable the company to maintain its leadership in the HBM market?"
— Daehoon Han, SK Securities

A: "SK hynix has consistently demonstrated these capabilities since the HBM2E generation. Our accumulated competitiveness in time to market, product performance, mass production yield, quality, and customer trust are the differentiators that cannot be replicated in a short period of time."
— Song Hyeon-jong, President of Corporate Center

Assessment: The substantive disclosure, that HBM4 yields are already near mature HBM3E levels, is genuinely strong and buried beneath the rhetoric. The rhetoric is nonetheless a shift: leadership framed as an un-replicable bundle rather than a demonstrable gap is what incumbents say when rivals close. We move HBM competition from a latent risk to an active one.

2027 HBM Pricing Negotiations

An analyst asked where 2027 HBM contract pricing stands across generations. Management confirmed negotiations are progressing and, notably, conceded that conventional DRAM prices influence the discussion.

Q: "What is the status for HBM price negotiations for 2027? Could you please explain the whole situation regarding the contract discussions, including your outlook for pricing for both HBM3, HBM4, and also HBM4E forthcoming?"
— Nicolas Gaudois, UBS

A: "With conventional DRAM prices rising sharply in recent months, such market environment may also have some influence on our HBM pricing discussions. Having said that, of course, HBM pricing is not determined solely by conventional DRAM prices."
— Song Hyeon-jong, President of Corporate Center

Assessment: An important and underappreciated admission. If rising conventional DRAM prices lift HBM negotiations, the linkage runs both ways, and HBM pricing is less insulated from the conventional cycle than the structural bull case assumes. The premium is anchored to resource intensity and opportunity cost, both of which compress when conventional pricing does.

Capital Allocation After Kioxia and the ADR

The closing question tied the enlarged cash position to the absent shareholder-return announcement. Management declined to provide any specifics, citing ADR-related disclosure restrictions.

Q: "Following the recent sale of the company share in Kioxia, as well as the ADR offering, the company has significantly increased its cash position. Could you discuss your capital allocation strategy? In particular, are there any plans for additional shareholder returns this year?"
— Surim Lee, DS Investment & Securities

A: "Due to regulatory requirements and procedural restrictions related to the ADR offering, please understand that we cannot disclose any new material information that was not included in the offering document."
— Song Hyeon-jong, President of Corporate Center

Assessment: A legitimate legal constraint arriving at the worst possible moment. The company holds ₩69.4tn of net cash, its shares are 52% below their June high, and it is procedurally barred from saying what it will do about it. Resolution of that constraint is a near-term catalyst in its own right.

What They're NOT Saying

1. What share of volume the LTAs actually cover. Management explicitly declined: coverage will be kept at "an appropriate level" without a number. Without it, the earnings floor the contracts create cannot be sized, and the entire downside-resilience argument rests on an undisclosed denominator.
2. Whether conventional DRAM contract pricing itself decelerated. The ASP answer was framed entirely as product mix and shipment timing. At no point did management state that contract prices for conventional DRAM continued rising at the Q1 pace. The omission is conspicuous in an answer that was otherwise specific.
3. Any quantified Q3 or H2 pricing outlook. Consistent with practice, but the assertion that H2 blended ASP improves is the load-bearing claim of the entire outlook and carries no number attached to it.
4. The size, form or timing of incremental shareholder returns. Promised "within the year" in April and again now. The ADR restriction is a valid reason, but investors have waited two quarters for a figure while the cash pile reached ₩88tn.
5. HBM's share of revenue or margin. Never disclosed, and now more consequential: with conventional pricing decelerating and HBM pricing admitted to be partly linked to it, the blended margin path cannot be modelled from outside.
6. Any comment on the share price or the sector de-rating. The stock fell 47% from its June peak into the print and no one asked management about it, nor did management address capital-markets conditions beyond the ADR mechanics.
7. 2027 CapEx. The 2026 figure was raised to the high-₩40tn range and P&T7, M17 and a new cluster were announced, but the multi-year spending envelope those imply was left unquantified.

Market Reaction

All price levels are KRX closing prices verified against end-of-day data; the benchmark is the KOSPI. SK hynix discloses before the Korean market open, so the reaction is the print-day session.

• Cycle peak: ₩2,919,000 close on June 22, 2026. The shares fell 41.0% in the 30 days into the print, from ₩2,628,000 on June 29 to ₩1,550,000 on July 28.
• Pre-print close (July 28): ₩1,550,000, set on a session in which the KOSPI fell 10.8%.
• Reaction (July 29) close: ₩1,401,000, −9.6%, having traded as low as ₩1,246,000 intraday, or −19.6%, before recovering into the bell. The KOSPI fell 6.0% the same session.
• Volume: 12.4M shares against a 6.2M 30-day average, 2.0x, the highest-conviction session of the coverage arc.
• Session after (July 30): ₩1,322,000, a further −5.6%, on 8.7M shares, with the KOSPI down another 1.2%.
• Peak to print: −52.0% from the June 22 close to the reaction close; −54.7% to the July 30 close.
• Setup: still +138.1% year-to-date and +490.5% over the trailing twelve months entering the print, against a 52-week closing range of ₩245,000 to ₩2,919,000.

The reaction has to be read against a sector event rather than as a verdict on the print. The KOSPI tripped market-wide circuit breakers in consecutive sessions, the first time in the exchange's history, and the memory complex globally entered a bear market, with the three largest suppliers among the heaviest losers of market value worldwide. A competing Korean supplier reported without missing in the same window and sold off regardless. What repriced was the multiple the market is willing to pay for peak memory earnings, not SK hynix's execution.

Within that, the print did its own damage. A 9.6% decline on 2.0x volume, after a 41% run down into the event, is not sell-the-news exhaustion of the kind we saw in Q1 when the stock went flat on a record. It is active repricing by holders who had already been selling for four weeks and found in the miss a reason to finish. The intraday low at −19.6% and the partial recovery suggest the first real two-way argument in the stock since we initiated: buyers are now showing up at a price, which is not something that happened on the way up.

Street Perspective

Debate: Is the ASP Undershoot Mix or the Turn?

Bull view: Management gave a specific, checkable reason. High-value shipments deferred to H2 plus portfolio mix diluted blended ASP in a quarter where bit shipments hit guidance on both product lines. HBM4 ramping and 1c-nm volume lift H2 mix, contract pricing remains firm, and Q3 reasserts the trend.

Bear view: Halving the rate of DRAM price increase is what the top of a pricing cycle looks like, and mix is the explanation always offered first. Management conspicuously did not claim conventional contract prices held their trajectory. Q3 puts the entire revenue burden on ASP while NAND bit growth collapses to low single digit.

Our take: Genuinely undecided, and we are comfortable saying so, because the resolution arrives in one quarter. The mix explanation is credible and the deferred volume should reappear. It is also the kind of explanation that is true in the quarter it is given and beside the point two quarters later. We size positions for the possibility that both are true: H2 blended ASP recovers modestly and the cycle still passed its peak rate of change in Q2.

Debate: Do LTAs With Deposits Actually Raise the Trough?

Bull view: Roughly 10 customers on five-year contracts with financial deposits and volatility-dampening price mechanisms is a structural change from every prior cycle. Customers now carry a penalty for walking away. The through-cycle margin floor is permanently higher, which justifies a higher multiple on normalized earnings and makes the coming downturn shallower.

Bear view: Undisclosed coverage percentage, undisclosed pricing mechanics, and no test yet against a falling market. Contracts signed at the top get renegotiated at the bottom regardless of what they say, and deposits sized for a modest correction do not survive a severe one. The claim is unfalsifiable until it fails.

Our take: The bull case here is the strongest structural argument in the story and we have raised our estimate of the trough because of it. But we cannot size what management will not disclose. This is the single most important thing to watch over the next four quarters, because if the LTA architecture holds through a genuine pricing decline, the entire memory sector deserves a higher multiple and we would return to Outperform on that basis alone.

Debate: Is a 52% Drawdown Enough?

Bull view: The shares trade at a high single-digit multiple of trailing operating profit with ₩69.4tn of net cash behind them, HBM4 is ramping into a supply-constrained market, and the company is still setting records. Most of the cyclical air has come out and the sector selloff is indiscriminate.

Bear view: The multiple only looks cheap because the denominator is peak. Trailing operating profit includes two quarters at 72% and 76% margins that will not repeat, and CapEx is rising into it. On any normalized earnings figure the stock is not cheap, which is the peak-earnings low-multiple trap in its purest form.

Our take: The bear framing of the multiple is correct and the bull framing of the drawdown is also correct. At roughly 7x trailing operating profit the stock is not cheap on normalized numbers, but it is no longer priced for the cycle to continue either. That is the definition of balanced, and it is why the rating stays where it is rather than moving in either direction.

Model Update Needed

Valuation framework (KRW). Trailing four-quarter operating profit is ₩128.7tn. At the reaction close of ₩1,401,000 on roughly 732M shares post-ADR, market capitalization is approximately ₩1,026tn and enterprise value approximately ₩956tn net of ₩69.4tn cash, or about 7.4x trailing operating profit. At the July 30 close of ₩1,322,000 that falls to roughly 7.0x. Those multiples are the trap, not the opportunity: they embed two quarters at 72% and 76% margins. Against a normalized annual operating profit in the ₩80tn to ₩100tn range, which sits well above FY25's ₩47.2tn and well below an annualized H1 2026, the same enterprise value implies roughly 9x to 12x, which is a fair rather than a compelling multiple for a memory franchise past its margin peak.

Fair-value range: base ₩1.20M to ₩1.40M, bracketing the current price; bull case ~₩2.0M if H2 blended ASP recovers as guided and the LTA architecture demonstrably holds contract pricing through a soft patch; bear case ~₩800,000 if pricing normalizes toward FY25-like earnings power while 2027 capacity arrives. With the base case sitting on top of the market price and roughly symmetric tails, the risk/reward supports the rating rather than an action.

Thesis Scorecard: 2026_Q1 Signposts Revisited

Our Q1 2026 downgrade set seven signposts for this quarter. They graded almost exactly as the bear framing predicted on pricing and earnings quality, and better than expected on structure and the balance sheet.

Scorecard summary: Four of seven graded as the downgrade anticipated. The pricing deceleration we called the primary watch item arrived in the very next quarter and produced the first operating-profit miss of our coverage. Net-income quality deteriorated to its worst level. The ADR executed exactly as flagged and delivered none of the re-rating we hoped for, because it listed roughly three weeks after the cycle high. We were wrong that margin would compress from 72%; it expanded to 76%, and we should own that. Against those, the LTA architecture hardening into five-year deposit-bearing contracts across roughly 10 customers is a materially better structural outcome than we expected, and it is the reason our estimate of the trough moves up rather than down.

Overall: Thesis broadly confirmed on cycle position, improved on structure. The standing view that this is a flawless franchise at a cyclical peak survives the quarter intact, with the peak now dated to Q2 2026 on margin and to Q1 2026 on the rate of price change.

Action: Hold. No trim into this print given the drawdown already realized, and no add until the H2 ASP path resolves or the price reaches our stated upgrade zone.

Bottom Line: Maintaining Hold

Rating decision: We maintain Hold. We initiated at Outperform in July 2025 at ₩268,500, maintained through the Q3 and Q4 2025 records, and downgraded to Hold in April 2026 at ₩1,223,000 on cycle position and price.

An honest accounting of that downgrade. It was early and it looked wrong for two months. From our April call the stock rose a further 138.7% to ₩2,919,000 on June 22 before round-tripping. Anyone who acted on it forfeited that move. Measured to the reaction close, the shares are +14.6% since the downgrade while the KOSPI has fallen 11.8%, so the call has since worked relative to the local benchmark by a wide margin, and the reasoning has been vindicated by this quarter's pricing deceleration. Both statements are true. A rating that requires a 139% adverse move before it pays is not a good rating, and the lesson we carry forward is that identifying a cyclical peak is not the same as timing one.

The case for staying at Hold rather than acting now is that the two forces are close to balanced. Against the stock: the first miss on both revenue and operating profit in our coverage, DRAM ASP growth halved, NAND bit growth guided down to low single digit, CapEx raised into a decelerating price environment, net income almost entirely non-operating, and a shareholder-return announcement deferred. For the stock: a 52% decline already realized, a record 76% operating margin, ₩69.4tn of net cash, HBM4 in mass production at near-mature yields, enterprise SSD revenue doubling, and an LTA architecture with deposits that genuinely raises the trough relative to prior cycles.

Why not upgrade. Our published trigger was a pullback toward ₩850,000 to ₩950,000. At ₩1,322,000 the stock still sits about 39% above the top of that range, and would have to fall a further 28% to reach it. Discipline means honoring a stated level rather than reaching for a falling stock because the drawdown has been dramatic. The base case sits on top of the market price, which is the definition of a Hold.

Why not downgrade. Downgrading after a 52% decline, into a net-cash balance sheet, a ramping HBM4 and contracts that carry customer deposits, would be chasing the tape. The margin has not actually rolled yet, Q3 volume guidance is positive on both product lines, and the sector's de-rating has already done the work that an Underperform call would be predicting.

What would move us to Outperform: the shares reaching ₩850,000 to ₩950,000 with the franchise intact; Q3 blended ASP confirming that the Q2 undershoot was mix and timing rather than the turn; or hard evidence that the deposit-bearing LTAs hold contract pricing through a genuine spot decline, which would raise the through-cycle multiple for the whole sector.

What would move us to Underperform: operating margin rolling decisively below 60% with no offsetting volume; evidence that LTA coverage is thin or that customers are renegotiating; a competitor HBM4 qualification at a major accelerator customer, now an active rather than latent risk; or 2027 CapEx guidance that extends the build after demand growth has visibly moderated.

Signposts for Q3 2026 earnings (late October 2026):