One Quarter Earned More Than All of Last Year. That Is the Argument for Downgrading.

Results vs. Consensus

Year-over-Year Comparison

Sequential Comparison

Revenue. Revenue of ₩133.87tn is 69.2% above the year-ago quarter, and the composition is now extreme. DS revenue of ₩81.7tn rose 86% sequentially and contributed ₩53.7tn of operating profit, which is 93.8% of the consolidated total. Samsung is no longer a diversified electronics group with a semiconductor division. For the purposes of the current earnings stream it is a memory company with four attached businesses that collectively contribute about 6% of profit.

Margins. A 42.8% consolidated operating margin, with DS at 65.7%, is the number that changes our rating. Our initiation argued at 14.1% that this could not be a peak, and our Q4 note argued at 21.4% that it probably still was not. At 42.8% the argument no longer holds. Samsung's prior memory-cycle peak produced consolidated operating margins in the mid-20s. The company is now earning roughly double that, and it is doing so on a price series that rose more than 90% in a single quarter.

Earnings. EPS of ₩6,992.86 is up 486.8% year over year. On trailing twelve-month EPS of ₩12,512.79 the shares trade at 17.6x, and on annualised Q1 earnings at 7.9x. The 7.9x figure is the trap: it is the multiple a memory company always displays at the moment its earnings stop being repeatable, and it is precisely why we are not using it to justify a higher target.

Segment Performance

Divisional revenue and operating profit are reported before intersegment eliminations and do not sum to the consolidated totals. Harman's figures were not disclosed this quarter. The Total row is the consolidated income-statement figure.

Memory KPIs

Device Solutions: a 65.7% divisional margin

DS earned ₩53.7tn on ₩81.7tn of revenue. The memory business set a second consecutive quarterly record, and the mechanism was almost entirely realisation: DRAM blended ASP up in the low 90% range and NAND up in the high 80s, against server bit growth in the low teens for DRAM.

"Under these circumstances, in line with growing AI demand trends, we focused on expanding sales for mostly server applications in the first quarter, resulting in server bit shipment growth in the low teens percentage QoQ for DRAM and low 20% Qoq for NAND, as we post a record high quarterly sales in the server segment."
— Jaejune Kim, EVP of Memory Business

System LSI also improved sequentially on expanded flagship SoC sales, and foundry declined on off-peak seasonality.

Assessment: A 65.7% operating margin is not a run rate, it is a moment. It reflects a market where the supplier sets price and the customer accepts it, which is a real condition but a temporary one at this intensity. We are not forecasting collapse. We are declining to extrapolate.

Samsung Display: the internal victim

SDC revenue fell 29% sequentially to ₩6.7tn with operating profit down to ₩0.4tn, a 6.0% margin against 21.1% one quarter earlier. Management was explicit that its customers' memory-driven cost inflation is compressing panel demand and pricing.

"Rising memory prices driven by supply-demand imbalances have increased cost burdens for set manufacturers, which is expected to reduce overall set demand. Accordingly, display demand is likely to weaken and downside ASP pressure is expected to persist."
— Charles Hur, EVP and Head of Corporate Strategy Team, Samsung Display Corporation

Assessment: The clearest evidence available that the memory shortage is now destroying demand elsewhere in electronics, and it happens to be visible inside Samsung's own results. A display division being squeezed by its sister division's pricing is a leading indicator for what is happening to every set maker without a captive memory business.

Mobile eXperience: profitability halved by its own supplier

MX revenue rose 30% sequentially to ₩38.1tn on the Galaxy S26 launch, with operating profit of ₩2.8tn. Management described securing "single-digit profitability" through cost optimisation, a step down from the double-digit profitability delivered for full-year 2025.

"In 1Q 2026, memory prices surged, weakening profitability year-over-year. In 2Q, prices are expected to rise further, adding to cost pressures."
— Seong Cho, EVP and Head of Strategic Marketing Office for Mobile Experience

Assessment: The internal transfer has become large enough to change the division's profitability tier. Management also guided that 2026 handset volume will decline significantly even as value grows slightly, and that profitability will fall year over year. The group nets out far ahead, but the consumer franchise is being structurally weakened while the semiconductor one harvests.

Foundry and Visual Display: the two that improved quietly

Foundry earnings declined on seasonality, but the division established a silicon photonics foundation, will begin mass production for a major optical communications module customer in the second half, and held its Taylor Fab equipment move-in ceremony on April 23 with operations starting in 2026 and mass production in 2027. Visual Display and Appliances swung to a ₩0.2tn operating profit from a ₩0.6tn loss, on premium mix and resource optimisation.

Assessment: The VD/DA swing is a genuine positive that resolves a bear point we had flagged for two quarters, though a 1.4% margin is recovery rather than health. Foundry's silicon photonics entry and the 4nm HBM base-die traction are the first evidence that the one-stop strategy generates business rather than slideware.

Key Topics & Management Commentary

Overall Management Tone: Confident without triumphalism on memory, and notably candid about the second-order damage the memory cycle is inflicting on the group's own downstream divisions. Where prior calls hedged on HBM competitiveness, this one asserted first-mover status directly. Management was least forthcoming on the labour situation, confirming a strike window and an unrecognised provision without sizing either.

1. First to mass-produce HBM4, on NVIDIA's next platform

The disclosure that resolves two years of debate about whether Samsung had lost high-bandwidth memory permanently: the company shipped the industry's first mass-produced HBM4 and SOCAMM2 for NVIDIA's Vera Rubin platform, beginning February 2026.

"After we became the world's first to commence commercial shipment of HBM4 in February, we are now proceeding with ramp-up as scheduled, with supply volume expected to scale meaningfully in the second half of the year. HBM4 sales are expected to exceed 50% of total HBM sales from the third quarter onwards and also account for roughly half on a full year basis."
— Jaejune Kim, EVP of Memory Business

Management added that the enhanced specifications are "translating to actual premium on pricing," and that production-ready capacity is fully booked and sold out.

Assessment: Complete vindication of the free-option pillar from our initiation. Samsung went from unable to confirm HBM3E qualification in October to first-to-market in the successor generation at the reference customer in six months. This is the strongest competitive news in the coverage period, and it is precisely why the downgrade is a valuation call rather than a business one.

2. Demand fulfilment at a record low, with 2027 already pulling forward

The most striking operational statistic of the quarter was not a margin.

"We also have very tight inventory and available supply is far short of customer demand. In fact, our demand fulfillment rate is now at a record low. Unlike previous years, customers who are concerned about supply shortages are actually bringing forward their demand for 2027 already. Currently, just based on pre-booked demand alone, the supply-demand gap is looking to widen further in 2027 versus this year."
— Jaejune Kim, EVP of Memory Business

Assessment: This is the strongest available argument against our downgrade and we take it seriously. A widening 2027 gap on pre-booked demand alone would justify a longer runway than we are underwriting. The reason it does not change the rating is that it is a supplier's characterisation of its own order book during the most favourable pricing environment in industry history, and customers over-order into shortages. We would rather be paid to wait for confirmation than pay for it in advance.

3. HBM now earns less than ordinary DRAM

An inversion nobody was modelling twelve months ago: conventional DRAM currently carries higher margins than high-bandwidth memory, because of how the two are priced.

"Per industry practice for HBM, we negotiate projected pricing in advance on an annual basis, considering the lead time required to prepare back-end capacity for HBM. Whereas, for conventional DRAM, the negotiations are done on a quarterly basis. As conventional DRAM pricing has continued to rise sharply every quarter, this has resulted in inversion of margins between HBM and conventional DRAM."
— Jaejune Kim, EVP of Memory Business

Management declined to tilt the mix toward the higher-margin product, arguing that starving HBM would constrain the AI build-out that generates all memory demand, and expects the differential to narrow significantly in 2027.

Assessment: Two important implications. First, the annual HBM pricing convention means Samsung has already locked much of its 2026 HBM economics, which caps upside and provides downside protection. Second, declining to arbitrage its own mix for short-term profit is genuinely good long-term stewardship, and it also quietly tells you that reported margins would be even higher if management chose otherwise.

4. Multi-year contracts, now binding

Samsung confirmed it has signed finalised multi-year supply agreements with some customers, a structural change from the industry's historic practice.

"Unlike existing supply contracts, which are based on mutual trust, these multi-year contracts present a higher level of binding commitments compared to the past in today's environment where investment size, timelines, and technical complexity have all increased significantly."
— Jaejune Kim, EVP of Memory Business

Assessment: A partial answer to the question we raised last quarter about whether booked volumes are priced. Binding multi-year contracts reduce volume risk and, by implication, fix or collar price for those volumes. That dampens both tails. It is a reason to lower the multiple on peak earnings, not raise it.

5. A strike window, and a provision that was not taken

The union held a rally at Pyeongtaek on April 23 and announced a general strike from May 21 to June 7. Separately, the related incentive provision was not booked.

"The incentive related provision you mentioned is under discussion between labor and management. As the specific amount, payment conditions and amounts have not been finalized, it hasn't been reflected in this quarter's results."
— Soon-Cheol Park, CFO

Assessment: Two distinct exposures, neither quantified. A production interruption during the tightest supply conditions on record would be costly in a way it would not be in a normal market, because there is no inventory buffer and the fulfilment rate is already at a record low. Reported Q1 profit also excludes a cost management expects to recognise. Both argue for a wider risk band around near-term estimates.

6. ₩110tn committed, and the buyback actually cancelled

The March corporate value enhancement plan commits over ₩110tn to facilities and R&D across Pyeongtaek, Taylor and the Yongin cluster. Separately, the treasury cancellation promised at the Q4 call was executed: 73.4 million common shares and 13.6 million preferred, representing 1.2% and 1.7% of their respective classes, worth approximately ₩14.6tn.

Assessment: The cancellation is a delivered commitment and we grade it accordingly. The ₩110tn investment figure is the more consequential number, because capital of that magnitude deployed near a cycle peak is how memory companies historically manufacture the next downturn. It is the right strategic decision and simultaneously the mechanism of future margin compression.

7. NAND rejoins the story

Management pointed to NVIDIA's CMX architecture, presented at GTC, which extends AI inference data storage beyond high-bandwidth memory into NAND, driving demand for TLC-based PCIe Gen 6 SSDs. Samsung finished 2Tb QLC development in March and plans a 256TB server SSD lineup.

Assessment: A genuine second leg. NAND has been the neglected half of the memory recovery, and an architectural shift that puts storage inside the inference path broadens the addressable demand rather than merely reallocating it. Samsung's NAND ASP already rose in the high 80% range this quarter.

8. Mature-node discipline in foundry

Alongside advanced-node ambition, foundry management committed to "boldly closing out uncompetitive processes," with 8-inch PMIC, DDI and CIS lines scheduled for phased closeout and CIS/DDI capacity migrating to 7nm.

Assessment: The first genuinely unsentimental capacity decision this division has announced. Closing lines is how a foundry business stops subsidising unprofitable volume, and it matters more for the path to breakeven than any single customer win.

9. Cooling, robotics and the widening perimeter

Management sized the data centre cooling market at $4.7bn in 2024 growing to $16.6bn by 2030, and described plans to expand FläktGroup into North America and Korea. Robotics received its own strategic exposition, including internalising components and considering acquisitions.

Assessment: The cooling logic is defensible because it attaches to the same datacentre demand driving memory. The robotics ambition is the item to watch. Record cash generation plus a stated appetite for strategic acquisitions across HVAC, automotive, medtech and robotics is the classic setup for capital misallocation, and shareholders have no framework yet for judging the returns.

10. Supply-chain and macro exposures acknowledged

Management addressed Middle East conflict risk to processed gases including helium and naphtha, reporting sufficient safety stock and diversified suppliers in the US and Japan, while flagging oil-driven ocean and air freight inflation as an active cost risk.

Assessment: Handled with more specificity than most peers offer. The freight cost point is the live one: it affects the consumer divisions disproportionately, which are already the weakest part of the group.

Guidance & Outlook

Note the shape of this guidance. Every volume line is capped and every price line is open-ended or unquantified. Management has stopped guiding bit growth as an ambition and started guiding it as a constraint, which is the correct description of a sold-out business and also means the earnings model has become a pure price forecast.

The H2 view has firmed. Management had deferred its second-half 2026 outlook at the October call. It now says agentic AI adoption is likely to accelerate faster than initially expected and that conventional server DRAM and SSD demand will rise more sharply than previously anticipated. That is a raise, delivered one quarter earlier than promised.

Analyst Q&A Highlights

Whether the industry is moving to binding long-term contracts

The opening question addressed the structural shift toward multi-year agreements, which would change how a cyclical business is valued. Management confirmed contracts are signed while withholding terms.

Q: "There's a lot of talk in the sector about multi-year contracts and LTA. Are you also seeking this kind of multi-year contracting for your memory products?"
— Seicheol Lee, Citigroup

A: "Based on these requests, we have been pursuing multi-year supply agreements, of course, within our available supply capacity, and have already signed finalized contracts with some customers."
— Jaejune Kim, EVP of Memory Business

Assessment: Confirmation without terms. Binding volume commitments cut both tails: they protect the downside in a correction and they cap participation if spot pricing keeps climbing. For a business being valued on peak quarterly earnings, the second effect is the more relevant one and it argues for a lower multiple, not a higher one.

Decomposing a quarter that nearly tripled operating profit

Questioning sought the drivers behind a result that exceeded the prior full year. The response separated volume from price with unusual precision and revealed that NAND beat its own guidance.

Q: "During the first quarter, I think, your solid performance was in large part driven by the memory business. If you could provide more color, more details on the memory performance."
— Woo Dong-je, Bank of America

A: "Consequently, we met our bit growth guidance for DRAM while exceeding guidance for NAND, with NAND bit growth increasing by a high single-digit percentage QoQ. Amid a sharp rise in market pricing and the effect of our improved product portfolio following expanded server application sales, our blended ASP rose by low 90% range QoQ for DRAM, high 80% QoQ for NAND."
— Jaejune Kim, EVP of Memory Business

Assessment: The cleanest disclosure of the quarter and the basis for the downgrade. When blended ASP rises more than 90% sequentially, the resulting margin tells you about the market's condition rather than the company's earning power. A price series that moves that far in one quarter is not a base to capitalise.

Whether the shortage extends into 2027

A recurring line of questioning tested the durability of the up-cycle. The answer introduced the record-low fulfilment rate and the pull-forward of 2027 demand.

Q: "The memory sector is expected to see a continued upcycle, what is your business outlook for the second quarter? If you could provide bit growth guidance for Q2."
— Dong-won Kim, KB Securities

A: "With available capacity under such constrained conditions, we plan to maintain our product mix focus on servers in the second quarter, and we're expecting DRAM bit growth to increase by single-digit percentage QoQ. While for NAND, considering limited available capacity, from a reduced inventory level in the first quarter, bit growth is expected to be constrained at a low single-digit level QoQ."
— Jaejune Kim, EVP of Memory Business

Assessment: Single-digit and low-single-digit bit growth against a business earning 65.7% divisional margins means Q2 earnings are almost entirely a price bet. The constraint is capacity rather than demand, which is bullish for pricing and simultaneously removes the volume lever that would let Samsung defend earnings if price turned.

Product mix between high-bandwidth and conventional memory

Questioning pushed on whether the company should chase the higher-margin product, given the pricing inversion. Management declined and explained the reasoning in terms of the ecosystem rather than the quarter.

Q: "Recently there's been a rapid rise in conventional DRAM prices, and some have been suggesting that focusing on conventional DRAM sales over HBM may be better in terms of margins. What are your plans for product mix between HBM versus conventional DRAM?"
— Chae Min-Suk, Korea Investment & Securities

A: "If we were to focus our product portfolio on conventional DRAM, looking to achieve a short-term performance only, this could potentially pose constraints on the build-out of the underlying AI infrastructure itself. Which is why we believe a balance in supply between HBM and conventional DRAM is necessary in order to continue to generate AI-driven demand."
— Jaejune Kim, EVP of Memory Business

Assessment: The best answer on the call and a real signal about management quality. Declining to maximise a reported quarter in order to protect the demand ecosystem is the opposite of how this industry historically behaves. It also implies that current margins are a floor on what the company could be reporting, which is a curious thing to find reassuring.

The labour dispute and potential production disruption

Questioning turned to a risk sitting outside the financials entirely. Management confirmed the strike window and described a mitigation posture without sizing the exposure.

Q: "As the CFO mentioned, just now, there does seem to be a lot of labor union issues. According to the media reports, a general strike has been announced for May. Do you anticipate any disruptions to production or otherwise, any negative impact on performance, any other management issue?"
— Woo Dong-je, Bank of America

A: "The labor union on April 23rd held a rally at the Pyeongtaek site and has announced a general strike from May 21st to June 7th. While it is difficult to comment further at this stage, even in the event of a strike, the company plans to respond through dedicated teams and response system within the legal framework to minimize potential production disruptions."
— Soon-Cheol Park, CFO

Assessment: A dated, specific operational risk that the reported numbers do not reflect and that the market appears to be discounting entirely. In a quarter where the fulfilment rate is at a record low, lost output cannot be recovered from inventory. This is the most concrete near-term downside catalyst in the coverage.

Foundry order momentum and the Taylor timeline

Questioning on advanced-node commercial traction produced specifics on customer discussions, a new business line, and a firm construction milestone.

Q: "Driven by a recent increase in orders, there are growing expectations for increased investments. Could you provide an update on the expansion status of the U.S. Taylor Fab and whether the company's considering the construction of new fabs too?"
— Jongwook Lee, Samsung Securities

A: "For Taylor Fab, as of last week, April 23 to be exact, we had a successful ceremony with the local community for Fab 1, commemorating the move-in of the equipment and lines. As scheduled, we plan to start operations in 2026 and commence mass production in 2027 and gradually expand to nano capacity. Fab 2 is in the early review phase."
— Sukchae Kang, EVP and Head of Sales and Marketing Office for Foundry

Assessment: On schedule, which is itself notable for a project with this history. Note the sequencing: operations in 2026, mass production in 2027. Taylor is a 2026 cost and a 2027 revenue contributor, which is the wrong way round relative to where we are in the memory cycle.

The successor shareholder-return policy, still undefined

With the current framework expiring this year and cash generation transformed, questioning returned to what replaces it. The answer was again a process update rather than a direction.

Q: "This year marks the final year of the three-year shareholder return policy. Will the company execute the previously announced policy as committed? Could you share the direction of the next shareholder return policy?"
— Sun-woo Kim, Meritz Securities

A: "Management and the board are currently gathering various views and engaging in deep discussions on the next shareholder return policy. Centered on the board, we will continue to carefully review and develop the optimal policy to enhance shareholder value, and we will share details once the direction is decided."
— Soon-Cheol Park, CFO

Assessment: A second consecutive quarter of process language with no framework and no date. Against free cash flow that has been transformed and a ₩110tn investment commitment competing for the same money, the absence of direction is now a genuine gap. This remains the largest non-operational catalyst available, and it is being deferred.

What They're NOT Saying

1. The size of the labour incentive provision: confirmed to exist, confirmed to be excluded from Q1, not sized and not dated. A known cost sits outside the reported margin.
2. Any quantification of strike impact: a specific three-week strike window was confirmed with no estimate of potential lost output, in a quarter where demand fulfilment is at a record low.
3. Pricing terms on multi-year contracts: contracts are signed and "binding," but whether volumes are fixed-price, collared or market-linked is undisclosed. This determines how much of 2027 upside shareholders actually own.
4. 2026 CapEx, still: a third consecutive quarter without a figure. The ₩110tn multi-year plan was disclosed instead, which is a longer horizon and a less useful number for modelling this year.
5. Harman's results: revenue and operating profit were disclosed for every division except Harman, which was described only as declining. It was quantified in the prior quarter.
6. The successor shareholder-return policy: no framework, no date, no indication whether the 50%-of-free-cash-flow basis survives a period when free cash flow could multiply several times.
7. What happens to HBM pricing when the annual negotiation resets: management explained that the margin inversion exists because HBM is priced annually in advance, but said nothing about where the next annual reset lands.
8. Any framework for judging the acquisition programme: HVAC, ADAS, digital healthcare, audio and now robotics, with more "strategic investments or acquisitions" flagged, and no stated return threshold or capital cap.

Market Reaction

• Pre-print setup: shares closed at ₩226,000 (~$153.40) on April 29, up 88.5% year to date and 307.2% over trailing twelve months, having gained 35.2% in the trailing thirty days. The close was the top of the 52-week closing range of ₩53,900 to ₩226,000.
• Reaction session (April 30): opened at ₩229,000 (+1.3% gap), traded ₩220,500 to ₩230,000, and closed at ₩220,500 (~$149.66), down 2.4% or ₩5,500. The close was the session low.
• Volume: 23.6 million shares against a 30-day average of 26.5 million, roughly 0.9x normal and below average.
• Relative: the KOSPI fell 1.4% the same session, so the stock underperformed by roughly 1.0 percentage point. The KOSPI was up 58.8% year to date entering the print.
• Provisional-release reaction (April 7-8): the earlier provisional disclosure drove a 9.0% two-session advance, from ₩193,100 on April 6 to ₩210,500 on April 8, with 37.8 million shares traded on the second session.

The two reactions together are the story. The provisional release, which carried a 35% to 41% beat against the then-prevailing consensus, produced a 9.0% advance on heavy volume. The full release, which carried a 3.5% beat against a reset consensus plus the entire strategic disclosure set, produced a 2.4% decline on below-average volume.

Closing at the session low on light volume after gapping up is a market that has finished buying. The absence of volume matters as much as the direction: there was no capitulation and no rotation, simply an absence of marginal demand at ₩226,000 after a 307% twelve-month move. That the HBM4 Vera Rubin confirmation, the record-low fulfilment rate, and the widening 2027 gap could not lift the shares is the most useful information the session produced.

We read this as the market beginning to apply the peak-earnings discount rather than as a verdict on the quarter. That is also our reading, which is why the rating changes here rather than after the price does.

Street Perspective

Debate: Is 42.8% operating margin a new baseline or a moment?

Bull view: The mix is structurally different from prior cycles. HBM4 barely contributed in Q1 and scales to over half of HBM sales from Q3, multi-year contracts underwrite volume, and the supply constraint is physical construction rather than equipment. This margin is defensible for longer than any prior peak.

Bear view: DRAM ASP rose more than 90% in one quarter. No pricing series that moves that far is a baseline. Every memory peak in history has been accompanied by a structural explanation, and the margin is now roughly double the prior cycle high.

Our take: The bears have the better of it now, which reverses our position from the prior two quarters. We accept the mix is genuinely better than in 2018. We do not accept that it is twice as good, and that is what 42.8% against a mid-20s prior peak requires you to believe.

Debate: Does the sold-out order book de-risk 2027 or borrow from it?

Bull view: Customers pulling 2027 demand forward and signing binding multi-year contracts is the strongest possible evidence that scarcity persists. On pre-booked demand alone the gap widens next year.

Bear view: Pull-forward is precisely what customers do at the top, and double-ordering into a shortage is the oldest pattern in semiconductors. A pre-booked order book assembled during panic buying is not a forecast, and Samsung's ₩110tn investment programme plus every competitor's response arrives in 2027.

Our take: Genuinely unresolved, and the honest position is that we cannot distinguish real 2027 demand from insurance ordering with the disclosure available. That uncertainty is itself the argument for Hold rather than Outperform: we are being asked to underwrite a year we cannot yet see, at a price that assumes we can.

Debate: Is the group structure now a liability?

Bull view: The consumer divisions cost roughly 6% of profit and give Samsung a captive demand signal, supply-chain leverage, and a hedge for the eventual downturn that pure-play competitors lack.

Bear view: Display margin fell from 21.1% to 6.0% and handset profitability dropped a tier, both directly because of the memory cycle. Investors wanting memory exposure can buy it undiluted elsewhere, and the record cash is now funding acquisitions across five unrelated industries.

Our take: The dilution is real but small and the hedge is real and undervalued at this point in the cycle. Our concern is not the existing divisions, it is the acquisition programme. Peak-cycle cash plus a stated appetite for inorganic growth across HVAC, automotive, medtech and robotics is where memory windfalls have historically gone to be destroyed.

Model Update Needed

Valuation. At ₩220,500 the shares trade at 17.6x trailing twelve-month EPS of ₩12,512.79 and 7.9x annualised Q1 earnings. Against our FY2026 estimate of ₩26,000 to ₩32,000 the multiple is roughly 7x to 8x.

Price target: ₩230,000, raised from ₩190,000, implying about 4% upside. The target is deliberately not derived from FY2026 earnings. It applies roughly 14x to a normalised mid-cycle EPS estimate of ₩14,000 to ₩18,000, a level well above the pre-cycle earnings base and well below what the company is currently earning. Anchoring to peak EPS at any respectable multiple produces a target multiples higher than the current price, which is exactly the error this sector punishes.

Thesis Scorecard Post-Earnings

Scored against the standing thesis carried since initiation, and against the commitments management made on the January call.

Management commitments from the January call, graded

Overall: The thesis is not weakened, it is finished. Five of nine commitments were delivered outright and the two most important, HBM4 and the buyback cancellation, were delivered in full. The problem is that a thesis built on the gap between 14.1% margins and mid-20s peak margins has no remaining gap to close at 42.8%.

Action: Downgrade to Hold, target ₩230,000. We would revisit upward on disclosure that multi-year contract volumes are market-priced rather than fixed, or on a successor shareholder-return policy that commits a materially higher share of a much larger free cash flow. We would move to Underperform on evidence that mobile and PC memory content per device is falling, on a strike materially disrupting output, or on any sign that 2027 pre-bookings contain double-ordering.