The Cycle Got Contracted: Price Floors, Prepayments and a 43% De-Rating Change the Case

Results vs. Consensus

Year-over-Year Comparison

Sequential Comparison

First Half 2026 vs. Full Year 2025

Revenue. Revenue of ₩171.50tn is up 130.0% year over year, and DS revenue of ₩127.5tn grew 56% sequentially while the consumer division fell 9%. The concentration is now total: DS contributed ₩89.2tn of ₩89.49tn consolidated operating profit, or 99.7%. Mobile, displays, televisions, appliances and Harman together are a rounding error on the group's earnings.

Margins. A 52.2% consolidated operating margin, with DS at 70.0%, is roughly double the level at which we downgraded three months ago and roughly quadruple Samsung's own prior-cycle peak. We are not going to argue this is sustainable, and neither is management, which is precisely why the contract structure discussed below matters more than the margin does.

Earnings. EPS of ₩10,849 on the company's basis rose 52% sequentially, corroborated by net income rising 51.6% on an independent basis. Against the ₩207,000 close the shares trade at 5.8x annualised first-half earnings. That multiple is not the reason to own this and we place no weight on it.

Segment Performance

Divisional revenue and operating profit are reported before intersegment eliminations and do not sum to the consolidated totals. Harman and the VD/DA operating result were not quantified this quarter. The Total row is the consolidated income-statement figure.

Device Solutions: volume returned alongside price

DS earned ₩89.2tn on ₩127.5tn of revenue, and the composition improved from the prior quarter. Both DRAM and NAND bit shipments set all-time highs, so this was not another pure-price quarter. Server revenue reached a record share of the sales mix, and foundry contributed higher utilisation with advanced nodes at capacity.

"The DS division added to last quarter's record performance at the shipments of both DRAM and NAND, setting all-time highs driven by robust market demand and the strength of our product."
— Sooncheol Park, CFO

Assessment: A 70.0% divisional margin is not a run rate, but the return of volume growth alongside price makes this quarter higher quality than Q1's. The more durable point is mix: server at a record share of revenue, with server SSD heading above 60% of NAND sales, moves the business away from the consumer-driven demand swings that historically caused memory cycles.

Mobile eXperience: an operating loss

MX and Networks posted a ₩0.7tn operating loss on ₩33.2tn of revenue. Revenue grew year over year on Galaxy S26 and A-series momentum, but component costs overwhelmed it. This is the bear point we have tracked since initiation arriving in full.

"Right now, the massive demand for AI servers is creating a shortage and driving up prices for mobile memory. We already saw this in Q2 as memory prices rose quarter-on-quarter, leading to pressure on our profitability, and we expect that cost burden to continue through the second half of the year."
— Daniel Araujo, EVP and Head of Strategic Planning Group for Mobile Experience

Assessment: A handset business of this scale losing money is a serious development in isolation and an almost irrelevant one in the context of ₩89.2tn of semiconductor profit. The strategic question is whether Samsung is willing to let MX run at a loss indefinitely to maximise DS. The answer implied by the mix policy is yes, for now. Watch for any sign that memory allocation is being redirected internally to rescue the handset margin, which would be value-destructive.

Foundry: close to the turn

Foundry earnings improved significantly before incentive-related provisions, on HBM base-die demand and strong US orders. Advanced nodes at 8nm and below have reached full utilisation, 2nm projects were secured from major cloud and AI customers, and management named Broadcom among parties in discussion.

"From a profitability perspective, we expect significant improvements year-on-year, supported by higher utilization, improved yields, and the effect of pricing adjustments. While it is difficult to predict exactly when we will turn around due to the nature of the Foundry business, which is driven by custom orders, we believe the turnaround is possible in the near term."
— Sukchae Kang, EVP and Head of Sales and Marketing Office for Foundry

Assessment: The closest this division has come to a profitability commitment in the coverage period, and the first time full utilisation on advanced nodes has been claimed. We have carried foundry at zero value since initiation. We are not yet changing that, but a turnaround inside the next few quarters now looks more likely than not, and Taylor Fab 2 breaking ground by year-end signals internal confidence.

Display and the consumer businesses

SDC recovered to ₩0.7tn of operating profit on ₩7.5tn of revenue and began Gen 8.6 IT OLED mass production in July, though management warned that initial ramp fixed costs will weigh on near-term earnings. Visual Display and Appliances slipped back to a slight operating loss as costs rose, and a leadership change at Visual Display accompanied a strategic pivot toward AI TV and a service-platform model.

Assessment: The consumer complex is now uniformly loss-making or barely profitable, entirely because of input costs generated by its own group. The Visual Display leadership change and the explicit reframing toward advertising and services is the first structural response we have seen after four quarters of asking for one.

Key Topics & Management Commentary

Overall Management Tone: The most substantive call of the coverage period, and the first in which the memory business volunteered contract mechanics rather than deflecting to NDAs. Confidence on supply duration extended a full year further out than the prior call, with the reasoning shown rather than asserted. Management was candid about the consumer divisions' distress and unusually forthcoming on the accounting treatment of the incentive accrual, which it could easily have left for the filings.

1. The contract structure, finally disclosed

Across three consecutive calls we have flagged the pricing terms on Samsung's booked capacity as the single most important undisclosed variable. This quarter it was answered in detail.

"We actually have been signing multi-year supply agreements based on a five-year term, and this is by contract, subject to annual negotiations where the parties can agree to add on an additional year, effectively operating on a rolling basis. We have already finalized agreements with the top five global data center customers, are also in the final stages of talks with five additional major accounts."
— Jaejune Kim, EVP and Head of the Global Sales and Marketing Office for Memory

Management added that advance payments are a contractual requirement maintained across the term, that one-fourth of those payments has already been received, and that the target is 60-70% of total capacity under such agreements.

Assessment: This is the disclosure that changes the rating. A five-year rolling contract, prepaid, covering the majority of capacity with the largest datacentre buyers, is not how a commodity business is sold. It converts a substantial share of revenue from spot-cyclical to contracted, and the prepayments transfer working capital risk to the customer.

2. Minimum price floors

The specific term that matters most was given plainly.

"For mainstream products, we have established minimum price floors at a level sufficient to mitigate against future investment risk arising from market price volatility."
— Jaejune Kim, EVP and Head of the Global Sales and Marketing Office for Memory

Assessment: Price floors are asymmetric protection. They do not cap the upside if spot runs, but they set a contractual base under the downside across the majority of capacity, calibrated by management's own account to cover investment risk. The historic reason memory deserves a low multiple is that the trough is unbounded. A floor under most of the volume is a direct attack on that argument, and it is why we are willing to apply a higher multiple to normalised earnings than we did in May.

3. The shortage call extends to 2028

Management moved its supply-tightness horizon out by a year, and showed the arithmetic rather than asserting the conclusion.

"Moreover, despite industry-wide increases in CapEx, considering the lead time from new fab construction to actual wafer production, which exceeds three years, this means that any meaningful increase in supply capacity will take considerable time. We believe it will be unlikely to see any significant increase in incremental supply through 2028."
— Jaejune Kim, EVP and Head of the Global Sales and Marketing Office for Memory

Management also stated that constraints will be more severe in 2027 than 2026, and that visibility beyond 2029 is limited.

Assessment: The three-year construction lead time is a physical fact, not a forecast, and it is the strongest part of the argument. Every competitor announcing capacity today is announcing 2029 supply. That said, this is still a supplier describing its own market at maximum profitability, and the honest caveat management itself offers about 2029 is the right place for an investor's skepticism to sit.

4. HBM share returning to parity

Samsung expects its high-bandwidth memory share to converge with its overall DRAM share in the second half, which would close the gap opened during the HBM3E generation.

"In the second half, we believe we will be able to achieve HBM market share broadly in line with our overall DRAM market share, resulting in a more balanced business portfolio."
— Jaejune Kim, EVP and Head of the Global Sales and Marketing Office for Memory

HBM4 sales are guided to rise more than threefold sequentially in Q3 and to exceed 60% of the HBM revenue mix in the second half, with 2027 supply agreements already signed and industry-first HBM4E samples shipped.

Assessment: Two years ago Samsung was the memory company that had missed high-bandwidth memory. It is now claiming share parity, first-to-market on two consecutive generations, and pre-signed 2027 volume. This is the single largest competitive reversal in the coverage period and the market is currently paying nothing for it.

5. The incentive accrual, and why Q3 margins will look worse

The CFO volunteered an unusually technical explanation of a cost that will move between periods.

"In the second quarter, we recognized a cumulative accrual for the first half, reflecting a special performance incentive equivalent to around 10.5% of cumulative first half operating profit. To help with understanding, it's important to note that there is a timing difference under accounting standards between recording a bonus accrual in the financial statements and recognizing the related expense in the income statement."
— Sooncheol Park, CFO

Part of the accrual sits capitalised in work-in-progress inventory as manufacturing labour cost and will be recognised in cost of sales as that inventory sells, beginning in Q3.

Assessment: Credit for disclosing it unprompted in this much detail. The practical consequence is that Q3 gross margin carries a cost that Q2 economically incurred, so a sequential margin decline next quarter will partly be an accounting artefact rather than a pricing signal. Investors who miss this will misread Q3 as the cycle turning.

6. Free cash flow is about to be distorted by prepayments

The CFO flagged a consequence of the contract structure that directly affects the shareholder-return formula.

"Regarding our current policy of returning 50% of free cash flow, I would like to note that customer prepayments under LTAs for our memory business as well as share repurchases for employee compensation may affect FCF."
— Sooncheol Park, CFO

Assessment: An important warning that cuts both ways. Advance payments inflate reported free cash flow in the years they are received without representing recurring cash generation, and a 50%-of-free-cash-flow policy mechanically translates that into higher returns. Shareholders should want the successor policy to distinguish between operating cash flow and customer financing, or the payout will be pro-cyclical in exactly the wrong way.

7. Capital spending is finally moving, without a full-year number

Second-quarter CapEx was ₩16.8tn, up ₩5.5tn sequentially, with ₩15.4tn to Device Solutions and ₩0.7tn to display, funding the new Pyeongtaek fab, the Taylor ramp and Gen 8.6.

Assessment: Roughly ₩28tn spent across the first half against ₩52.7tn for all of 2025, so the annual figure is heading materially higher. This is a fourth consecutive quarter without a full-year guide, which for a business whose capital intensity determines the next cycle's cost curve is a genuine and persistent disclosure gap.

8. Foundry customers named, and Taylor Fab 2 committed

Foundry secured 2nm projects from major cloud and AI customers, named Broadcom among those in discussion, and expects 2nm project wins to more than double year over year. Taylor Fab 1 remains on track for 2026 operations, and Fab 2 construction begins by the end of this year targeting 2030 mass production.

Assessment: Committing to Fab 2 construction while Fab 1 has not yet started production is an aggressive signal, and one that only makes sense if the order book supports it. The named-customer disclosure is a change in posture for a division that has historically hidden behind confidentiality.

9. NAND becomes a second engine

Server SSD is guided to exceed 60% of the NAND sales mix this year, more than 20 percentage points higher year over year, with QLC bit shipments in the second half set to more than double versus the first, a 256TB server SSD lineup, and V10 V-NAND entering mass production in August.

Assessment: NAND has moved from the neglected half of the recovery to a genuine contributor with its own AI-driven demand vector. A mix shift of 20 percentage points toward server in a single year is a structural change in the profitability of that franchise, not a cyclical one.

10. The perimeter keeps widening

Harman acquired ZF's ADAS business and Sound United. Robotics was consolidated into an RX business office reporting directly to the CEO, with a pilot production line and data factory in Gumi and senior hires including a former Boston Dynamics strategy lead.

Assessment: The robotics build-out is more coherent than the acquisition list suggested last quarter, with an organisational structure, a manufacturing base and named leadership rather than an ambition. It remains capital deployed at the top of a cycle into markets where Samsung has no established position, and shareholders still have no return threshold against which to judge it.

Guidance & Outlook

The shape of guidance has changed in a way worth naming. For three quarters this was a business guiding constrained volume and open-ended price. It is now guiding contracted volume with floored price, over a five-year horizon, for the majority of capacity. That is a different security.

Analyst Q&A Highlights

Contract terms, capacity allocation and prepayments

The most valuable exchange of the coverage period. Questioning pressed for terms rather than existence, and management responded with structure, counterparties, allocation targets, payment mechanics and pricing architecture.

Q: "Across the memory market overall, we're hearing lots of news about multi-year contracts. I believe at the last call, you also said that Samsung Electronics was also pursuing multi-year arrangements. To the extent possible, could you provide us with a status update and some color on the terms?"
— Min-Sook Chae, Korea Investment Securities

A: "To enhance the binding nature of multi-year supply commitments, we've included substantial advance payments as a contractual requirement to be maintained over the multi-year contract period. We've already received a significant portion of these payments, and the overall balance is expected to grow going forward. We've already received one-fourth of these payments."
— Jaejune Kim, EVP and Head of the Global Sales and Marketing Office for Memory

Assessment: Prepayment is the term that makes the contracts real. A customer that has already wired a quarter of a multi-year commitment has bonded itself to the volume in a way no memorandum achieves. Combined with price floors, this transfers both volume and price risk away from Samsung across the majority of its capacity, which is the structural change the equity is not yet paying for.

Duration of the shortage and mid-term demand

Questioning opened on whether the imbalance persists into next year. The answer went further than asked, extending the call to 2028 and describing a new class of buyer approaching Samsung directly.

Q: "The supply shortage in the memory market has continued, do you expect the supply shortage to continue into next year? If possible, could you share your mid to long-term outlook for memory demand?"
— Dongwon Kim, KB Securities

A: "The supply constraints are expected to become even more severe in 2027 than 2026, reinforcing our view that the supply shortage will persist through 2028. Beyond 2029, it is hard to say because of more limited visibility."
— Jaejune Kim, EVP and Head of the Global Sales and Marketing Office for Memory

Assessment: The willingness to name 2029 as the limit of visibility is what makes the 2028 claim credible. A management team simply talking its book does not volunteer where its own foresight ends. The disclosure that AI frontier model developers are approaching the supplier directly, bypassing hyperscalers, is a genuinely new demand channel.

Product mix between high-bandwidth and conventional memory

Questioning tested whether the balanced-mix commitment survived another quarter of margin inversion. Management reaffirmed it and paired the answer with the share-parity claim.

Q: "I believe, at the last earnings call, you explained that between HBM and conventional DRAM, you wanted to maintain a balanced product mix. Is that stance still valid today? Also, I heard some good news about your HBM4 performance. Could you provide us an update, a little bit more detail on the status of your HBM business?"
— Analyst, HSBC Securities

A: "Consequently, expect HBM4 sales to increase by more than three-fold Q on Q in the third quarter. Based on our outlook for the second half, HBM4 is expected to account for well over 60% of our total HBM revenue mix."
— Jaejune Kim, EVP and Head of the Global Sales and Marketing Office for Memory

Assessment: Consistency across quarters on a policy that costs reported profit is the strongest available evidence of genuine long-term orientation. The HBM4 ramp numbers are specific enough to be checked next quarter, which is how forward statements should be made.

The incentive accrual and its accounting treatment

Questioning returned to a cost deliberately excluded from the prior quarter. The response quantified it against operating profit and explained why the income-statement effect lags the accrual.

Q: "I remember that the bonus accrual was not recognized in the first quarter. Could you comment on the amount of the second quarter bonus accrual and your plan going forward?"
— Ise Choi, Citigroup

A: "A portion of the second quarter bonus accrual was included in the cost of work in progress inventory as manufacturing labor costs in accordance with accounting standards. Thus, the recognized expense amount in the two key income statement is lower than the total bonus accrual. The remaining capitalized amount will be recognized as cost of sales when the related inventory is sold to customers."
— Sooncheol Park, CFO

Assessment: A clean, unprompted explanation of a genuine earnings-quality issue. The practical takeaway for modelling is that Q2 operating profit is somewhat overstated relative to economic cost and Q3 will be understated for the same reason. Analysts who annualise Q2 without adjusting will be wrong twice.

Foundry utilisation, yields and the path to profit

Questioning sought the one metric this division has never given: when it stops losing money. Management gave utilisation and yield detail and came closer to a commitment than at any prior point in the coverage.

Q: "On the Foundry side, what is the current status of utilization rates and also yields? It seems that margins overall are improving, so to what extent are margins expanding, and when do you expect a turnaround to profit for Foundry?"
— Analyst, Bank of America Securities

A: "Regarding utilization rates have improved year-over-year across all nodes. Advanced nodes, 8-nano and below, actually have reached full utilization levels as we maximize sales of high growth products with strong underlying demand."
— Sukchae Kang, EVP and Head of Sales and Marketing Office for Foundry

Assessment: Full utilisation on advanced nodes is the precondition for profitability in a fab business, and this is the first quarter it has been claimed. "Possible in the near term" is still not a date, but the gap between rhetoric and evidence has narrowed substantially. We continue to carry the division at zero and regard that as increasingly conservative.

Whether a US listing is under consideration

Media reports of a possible American depositary receipt prompted direct questioning. Management denied an active review while leaving the option open.

Q: "Following a competitive U.S. listing, there has been growing market interest in the possibility of a Samsung ADR. Also, there have been recent media reports that the company is in the early stages of reviewing a U.S. listing. Could you comment on the possibility of an ADR listing?"
— Ise Choi, Citigroup

A: "Since the company provides stable cash generation based on its diversified business portfolio, the need for an ADR as a means of raising new capital is not high. While recent media reports have suggested the possibility of a U.S. listing, we're not currently reviewing an ADR issuance."
— Sooncheol Park, CFO

Assessment: A clear denial of an active process, with the door left open on a mid-term view. The reasoning offered, that a partial float in a separate market can create a value gap against domestic shares, is sound. For non-Korean investors the practical consequence is that access remains via the local line or the unsponsored over-the-counter alias.

Balancing employee compensation against buybacks

With treasury shares committed to both employee incentives and shareholder returns, questioning probed whether both can be funded simultaneously. The answer signalled a phased approach and a new regulatory constraint.

Q: "You may be required to do a sizable share buyback, but if you had to do both at the same time, employee compensation and share buyback, do you believe that you can do that all at once? Would that be feasible in the management's view?"
— Analyst, Bank of America Securities

A: "With respect to the timing of share buybacks, rather than one large-scale execution, we will determine the optimal method after considering both shareholder value and the effectiveness of employee compensation."
— Sooncheol Park, CFO

Assessment: A pre-emptive dampening of expectations for a single large buyback, and a reminder that treasury use for employee compensation now requires shareholder approval under the revised Commercial Act. Anyone underwriting a dramatic capital return announcement should read this answer as a caution.

What They're NOT Saying

1. The size of the advance payments: one-fourth received, balance growing, absolute figure withheld under confidentiality. This determines how much of the reported free cash flow is customer financing rather than operations.
2. Where the price floors actually sit: floors exist and are calibrated to cover investment risk, but no level or relationship to current spot was given. A floor 20% below spot and one 70% below spot are entirely different securities.
3. The exact incentive accrual amount: given only as a percentage of first-half operating profit, with the split between expensed and capitalised undisclosed, so the Q3 cost-of-sales drag cannot be sized precisely.
4. Full-year 2026 CapEx: a fourth consecutive quarter without a figure, now with half the year already spent.
5. The successor shareholder-return policy: a fourth consecutive quarter of process language. Still no framework, no date, and now an acknowledged complication about how prepayments interact with the free-cash-flow basis.
6. Harman's and VD/DA's actual numbers: Harman was described as improved and VD/DA as a "slight operating loss" with neither quantified, having been given in prior quarters.
7. A foundry breakeven date: "possible in the near term" after a quarter of full advanced-node utilisation. Still no quarter named.
8. Why revenue missed: the top line came in below expectations at the provisional release and cost the shares 12.7% in two sessions, and the full call offered no explanation of the shortfall against the Street's number.

Market Reaction

• Pre-print setup: shares closed at ₩208,500 (~$146.75) on July 29, up 73.9% year to date and 187.2% over trailing twelve months, but down 37.6% in the trailing thirty days from ₩334,000 on June 30. The 52-week closing range was ₩64,700 to ₩362,500, so the stock entered the print 42.5% below its high.
• Reaction session (July 30): opened at ₩214,000 (+2.6% gap), traded a wide ₩202,000 to ₩226,000 range, and closed at ₩207,000 (~$145.70), down 0.7% or ₩1,500.
• Volume: 46.7 million shares against a 30-day average of 32.0 million, roughly 1.5x normal.
• Relative: the KOSPI fell 1.2% the same session, so the stock modestly outperformed the index by about 0.5 percentage points. The KOSPI was up 34.4% year to date entering the print.
• Provisional release (July 7): shares fell 6.9% that session and a further 6.3% on July 8, a two-session decline of 12.7% from ₩318,000 to ₩277,500, on a revenue figure below the Street's expectation.
• Next-day session (July 31): shares rose 26.8% to ₩262,500. This was a market-wide move rather than a stock-specific one: the KOSPI rose 17.9% the same session and the closest domestic memory comparable rose 30.0%.

Three separate reactions, and only one of them is about this company. The July 7 provisional decline was Samsung-specific and was about revenue rather than profit. The July 30 print-day move was a 0.7% decline that slightly beat a falling index, which for a quarter of this magnitude means the information was already absorbed. The July 31 advance was an index-level event that swept the whole Korean market, and reading it as a verdict on these results would be a mistake.

The more useful observation is the setup rather than the sessions. A stock that has fallen 37.6% in thirty days and 42.5% from its high, while reporting first-half operating profit at 3.4x the prior full year, has undergone a violent de-rating against rising earnings. That is the market pricing the durability question rather than the earnings question, and durability is precisely what this call addressed.

We would also flag the trading environment itself as a risk. Daily index moves of the magnitude seen in late July are not a normal market, and position sizing should reflect that regardless of the fundamental view.

Street Perspective

Debate: Do multi-year contracts actually de-cyclicalise memory?

Bull view: Five-year rolling terms, prepayments already a quarter collected, price floors on mainstream products, and 60-70% of capacity committed to the largest datacentre buyers. This is a structural change in how the product is sold, and it deserves a structurally higher multiple.

Bear view: Contracts signed at the top of a cycle get renegotiated at the bottom, whatever they say. Take-or-pay arrangements in commodities have a long history of proving flexible when the buyer's economics deteriorate, and the floors have not been disclosed at a level anyone can assess.

Our take: The bears are right that contracts are only as good as counterparty willingness, but the prepayment mechanism is what distinguishes this from prior attempts. Cash already received is not renegotiable. We think this genuinely reduces trough severity without eliminating cyclicality, which justifies a higher multiple on normalised earnings rather than a higher estimate of peak earnings.

Debate: Is the July de-rating an opportunity or the start of the downturn?

Bull view: Shares fell 42.5% from the high while operating profit rose 56% sequentially and the shortage call extended a year further out. The stock is discounting a downturn that management says physically cannot arrive before 2029 given construction lead times.

Bear view: The market moved first in every prior memory cycle, and the revenue miss at the provisional release is the first crack. Mobile is losing money, mobile and PC demand is moderating, and the Q3 cost-of-sales drag from the deferred incentive accrual will make the next print look like deceleration.

Our take: We side with the bulls, with the specific caveat that Q3 will optically disappoint for accounting reasons. The bear case rests on demand destruction in consumer end markets, which is real but is roughly 6% of group profit; the bull case rests on datacentre demand, which is 99.7% of it.

Debate: Should the consumer divisions be restructured or sold?

Bull view: They are now immaterial to profit, provide a hedge for the eventual downturn, and Visual Display's new leadership is pivoting toward higher-margin advertising and services. The cost of keeping them is small relative to the option they represent.

Bear view: A handset business losing ₩0.7tn a quarter with guidance that cost pressure continues is not a hedge, it is a subsidy. Management attention and capital are being spent on divisions that will never again move the group's earnings.

Our take: The hedge argument is sound and we would not want these businesses sold at the top of a semiconductor cycle. But the case for a structural review of appliances and televisions specifically has strengthened for four consecutive quarters, and the Visual Display leadership change suggests management may finally agree.

Model Update Needed

Valuation. At the ₩207,000 close the shares trade at 5.8x annualised first-half earnings of ₩35,944 and roughly 4.7x to 5.4x our FY2026 estimate. Against normalised mid-cycle EPS of ₩18,000 to ₩22,000 the multiple is 9.4x to 11.5x.

Price target: ₩260,000, raised from ₩230,000, implying roughly 26% upside. As in the prior two notes, the target is anchored to normalised earnings rather than to what the company is currently earning, applying approximately 13x to the ₩20,000 midpoint of our normalised range. Two things changed: the normalised range moves up because price floors and contracted volume raise the trough, and the shares have de-rated 42.5% from their high, so the same normalised framework now produces meaningful upside rather than none.

Thesis Scorecard Post-Earnings

Scored against the standing thesis and the commitments management made on the April call.

Management commitments from the April call, graded

Overall: Six of nine commitments delivered, including the one we had identified as most important across three consecutive notes. The two persistent failures are both disclosure rather than operational, and both concern how capital gets allocated rather than how the business performs.

Action: Upgrade to Outperform, target ₩260,000. This reverses our May downgrade on new information rather than a change of view about peak margins, which we still regard as unsustainable. The downgrade rested on an inability to underwrite 2027; management has answered with contracted, prepaid, floor-priced volume covering the majority of capacity, and the share price has fallen 42.5% from its high in the interim. We would move back to Hold if the disclosed floors prove to sit far below mid-cycle economics, or if 2027 pre-bookings show evidence of double-ordering. We would move to Underperform on contract renegotiation by any of the signed counterparties.