A Seven-Year Record Broken, and the Stock Fell: What the Fade Says About What Is Already Priced

Results vs. Consensus

A 0.3% beat is not a beat in any meaningful sense. It is confirmation that the provisional release on January 8 did its job and that the sell side had three weeks to converge on it. Reading this print as an earnings surprise misunderstands the disclosure structure. The information content of January 29 was the divisional split, the HBM4 status, the 2026 capacity commentary, and the shareholder-return arithmetic.

Year-over-Year Comparison

Sequential Comparison

Full Year 2025

Full-year figures are the sum of the four reported quarters and tie to the company-stated FY2025 totals of ₩333.6tn revenue and ₩43.6tn operating profit.

Revenue. The 9.0% sequential gain understates the internal rotation. DS revenue rose 33% sequentially to ₩44.0tn while DX fell 8% on fading smartphone launch effects and tariff-pressured appliances. Semiconductors are now roughly 43% of pre-elimination divisional revenue and 82% of divisional operating profit. Samsung has become, for practical purposes of valuation, a memory company with a large consumer-electronics attachment.

Margins. Gross margin of 45.1% and operating margin of 21.4% are the numbers that change the investment debate. Our initiation argued that 14.1% was too low to represent a cycle peak. At 21.4% that argument is materially weaker, though still intact: Samsung earned above 25% operating margin at the 2018 peak, and the mix today includes a lossmaking appliance division and a foundry business that has not yet turned. On a like-for-like DS basis, the 37.3% divisional margin is high but not unprecedented.

Earnings. EPS of ₩2,980.93 is up 167.3% year over year and 65.4% sequentially. Full-year EPS of ₩6,711.61 puts the shares on 23.9x trailing at the ₩160,700 close, a multiple that is meaningless in a memory recovery because the denominator contains two depressed quarters. Against annualised Q4 earnings the multiple is 13.5x.

Segment Performance

Divisional revenue and operating profit are reported before intersegment eliminations and do not sum to the consolidated totals. The Total row is the consolidated income-statement figure.

Memory KPIs

Device Solutions: a 37.3% divisional margin

DS delivered ₩16.4tn of operating profit on ₩44.0tn of revenue, with memory setting records in both revenue and operating profit. The mechanism was almost entirely price. Bit shipments matched the guidance given in October rather than beating it, and NAND bits actually declined. What changed was realisation: DRAM ASP up roughly 40% sequentially and NAND ASP up in the mid-20s percent.

"In the fourth quarter, AI-related demand, particularly from hyperscalers, came through even stronger. With the spread of agentic AI, inference workloads expanded significantly, leading to a significant surge in demand, not only for AI servers but for conventional server applications as well."
— Jaejune Kim, EVP of Memory

Assessment: A 37.3% divisional margin driven by price rather than volume is both the most profitable and the most fragile configuration a memory business can be in. It requires no execution to sustain and no execution to lose. The constructive read is that the company shipped into the price increase without chasing volume, which preserves the shortage. The cautious read is that a single quarter of ASP normalisation would remove most of the incremental profit.

Foundry: revenue up, earnings capped again

Foundry revenue grew sequentially on US and Chinese customer demand, first-generation 2nm entered ramp, and shipments of 4nm HBM base-die products began. Earnings improvement was limited by the recognition of provisions. Management set a 2026 target of double-digit revenue growth with improving profitability, and disclosed that 2nm project awards are expected to rise 130% year on year.

"Most notably, we are expecting this year's 2 nm project awards to increase by 130% year-on-year, driven by HPC and AI applications."
— Sukchae Kang, EVP of Foundry

Assessment: The order book is genuinely inflecting and the 4nm HBM base-die shipment is strategically important, because it is the first commercial evidence of the one-stop logic-plus-memory-plus-packaging pitch. But this is now two consecutive quarters where a one-off has absorbed the operating improvement. We continue to assign this business no positive value and continue not to model breakeven before 2027.

Mobile eXperience: the cost squeeze arrives

MX shipped 60 million smartphones and 6 million tablets, with blended ASP falling to $244 from $304 in Q3 as flagship launch effects faded. Revenue and profit both declined sequentially, though full-year profitability held in double digits. Management stated plainly that mobile memory shortage and price increases began materialising in the quarter.

"As memory demand for AI server has expanded, memory supply shortages for mobile devices and sharp price increases started materializing in the fourth quarter of 2025. Therefore, we expect a challenging business environment in 2026. However, since this is an industry-wide issue, our competitors will also face the same environment."
— Seong Hyuk Cho, EVP for Mobile Experience

Assessment: The second sentence is the important one and it is correct. Samsung's handset margin will compress, but so will every competitor's, and Samsung is the only one of them that collects the offsetting profit upstream. The relative competitive position improves even as the divisional margin falls. The Q1 guide for both shipments and ASP to rise on the S26 launch means the trough in this division is likely behind it.

Samsung Display and the consumer businesses

SDC produced ₩2.0tn of operating profit on ₩9.5tn of revenue, a 21.1% margin, helped by peak-season large-display demand and IT and automotive growth. Visual Display and Appliances posted a ₩0.6tn operating loss, six times the ₩0.1tn loss of the prior quarter, on tariff pressure and continued price competition. Harman contributed ₩0.3tn on ₩4.6tn.

Assessment: The appliance and television loss is now large enough to matter, subtracting roughly 3% from consolidated operating profit. Management's response remains a product roadmap rather than a structural one. Display management's characterisation of 2026 as likely "more challenging than any previous year" deserves more attention than it received on the call, given that panel makers face memory-driven cost inflation with no upstream offset of their own.

Key Topics & Management Commentary

Overall Management Tone: Markedly more assertive than in October, with the CFO opening on a victory lap ("the second half unfolded as we promised") and the memory business volunteering forward disclosures it had withheld a quarter earlier. The confidence is concentrated where the evidence is strongest, in memory and HBM4, while display and mobile used explicitly defensive language about the same year. Management was least convincing on foundry, where a second consecutive quarter of one-off charges was described rather than explained.

1. HBM4 moved from promise to production

In October the company said HBM4 development was finished and samples had shipped. Three months later it is in full-scale production with a February shipment date and a named performance bin.

"We are receiving positive customer feedback on the competitive performance of HBM4, and based on this input, we've already commenced production, and HBM4 is now in stable full-scale production as scheduled, including HBM4 at 11.7 Gbps products at the highest performance bin pursuant to customer requirement, and shipments will start in February as well."
— Jaejune Kim, EVP of Memory

Management added that samples were supplied last year with no redesign required even as customers raised their performance requirements mid-cycle, and that the product is in the final phase of qualification.

Assessment: This is the single most important delivery against our initiation thesis. The "no redesign required" detail is the substantive claim, because it means the pre-built performance headroom described in October was real rather than marketing. HBM4 was an unpriced option at initiation; it is now a dated, quantified revenue event. We move this pillar from option to base case.

2. Capacity is sold out, and 2027 is the negotiation

The disclosure with the longest half-life was not about 2026 volumes but about who is asking for what beyond them.

"Regarding our 2026 HBM sales outlook, all production-ready capacity is currently fully booked with customer POs, and we expect 2026 HBM sales to improve substantially, increasing by more than threefold year-on-year. One thing of note is that despite our efforts to ramp up supply, major customer demand for HBM in 2026 still exceeds available supply from us. So for volumes in 2027 and beyond, major customers are seeking to finalize supply discussions as soon as possible to secure supply."
— Jaejune Kim, EVP of Memory

Assessment: A booked order book converts a cyclical business into something closer to a contracted one, and it is why we can raise the target materially without raising the multiple. The important nuance is that customers negotiating 2027 volumes in January 2026 are behaving as though they expect scarcity to persist, which is a stronger signal than any supplier forecast.

3. Clean rooms, not wafers, are the binding constraint

Management repeatedly located the supply ceiling in physical facility space rather than equipment or yield, and said the constraint extends through 2027.

"Now, due to limited clean room availability, supply growth is expected to be constrained, in 2026 and 2027, and we expect supply shortages to continue."
— Jaejune Kim, EVP of Memory

Correspondingly, the increased 2026 capital budget is directed at securing new fab space and clean rooms preemptively rather than at immediate capacity.

Assessment: This reframes the cycle. Equipment can be ordered and installed inside a year; buildings cannot. A shortage bounded by construction lead times is structurally longer than one bounded by tool delivery, and it explains why an industry earning these margins is not immediately competing them away. It also tells you the 2027 supply response, when it comes, will be large.

4. Capital spending overshot its own guidance

In October the company guided full-year 2025 CapEx to ₩47.4tn. It spent ₩52.7tn, with ₩20.4tn in the fourth quarter alone against ₩9.2tn in the third. DS took ₩47.5tn of the full-year total.

Assessment: A ₩5.3tn overshoot against a guide given thirteen weeks earlier is not a forecasting error, it is a decision. Management accelerated spending into the quarter it became clear demand was structural. We read this as the most credible corroboration available that the supply commentary is sincerely held, and we note that it front-loads the depreciation step-up we flagged at initiation.

5. Shareholder returns quantified, and a policy cliff approaching

Full-year free cash flow was approximately ₩36.5tn, making the 50% shareholder-return basis roughly ₩18.3tn. Against that, regular dividends of ₩9.8tn and an additional ₩1.3tn dividend were declared, with ₩6.6tn of the ₩8.2tn of 2025 share purchases to be cancelled and the cancellation schedule due in Q1 2026.

"In addition, with the 2025 results in the books, the company's free cash flow was approximately KRW 36.5 trillion, and 50% of free cash flow, which is the basis for shareholder returns, is around KRW 18.3 trillion."
— Sooncheol Park, CFO

Assessment: The additional dividend exists because Korea's new separate-taxation scheme for high-dividend companies requires a payout ratio of at least 25% and a 10% year-on-year increase in total dividends. This is policy-driven capital allocation, which is fine but should not be extrapolated as a change in philosophy. The more consequential fact is that 2026 is the final year of the current three-year policy, and with free cash flow about to rise steeply, the successor policy is a genuine catalyst.

6. The mix question nobody expected: server DDR over HBM

Management raised the possibility that conventional server DRAM may be more profitable per wafer than HBM in the near term, and that mix could tilt accordingly.

"Also, when considering that price increases have varied across different products used in AI server applications, we may need to focus our product mix more on server DDR over HBM in the short term from a profitability perspective."
— Jaejune Kim, EVP of Memory

Assessment: A remarkable inversion of the past two years' narrative, in which HBM was the only memory worth owning. If conventional server DDR now competes with HBM for wafer allocation on profitability, the market has tightened further than the HBM-focused commentary suggests, and Samsung's larger conventional franchise is an advantage rather than the liability it was treated as in 2024.

7. Sixteen-high stacking abandoned, hybrid bonding deferred

Management addressed the packaging-technology debate directly, concluding that mass commercialisation of 16-high HBM3E or HBM4 is unnecessary because HBM4E 12-high will reach equivalent density around mid-year. Hybrid copper bonding samples have shipped, with partial commercialisation planned at the HBM4E stage.

Assessment: A disciplined capital-allocation decision that will be misread as a technology retreat. Skipping a generation of packaging complexity to arrive at the same density through a cleaner path is correct if the timing holds. The risk is narrow but real: if a customer demands 16-high before HBM4E samples, Samsung is relying on a mass-production-ready but uncommercialised alternative.

8. The application-processor question, answered

At the October call, management could not confirm whether its own silicon would appear in the next flagship. The Q4 call resolved it: the Galaxy S26 carries a second-generation custom application processor.

Assessment: Resolves the most negative disclosure of the prior quarter in the company's favour, and removes a specific overhang from the System LSI outlook. It does not by itself fix that division, whose earnings declined sequentially, but it restores the vertical-integration argument that the October non-answer had undermined.

9. Acquisitions widening the perimeter

The CFO listed FläktGroup in HVAC, ZF's ADAS business, Zelis in digital healthcare, and Masimo's audio business as strategic acquisitions, alongside stated ambitions in humanoid robotics.

Assessment: Four acquisitions across four unrelated end markets, disclosed in a single sentence during the most profitable quarter in company history, is a pattern worth watching rather than celebrating. Conglomerate capital allocation tends to be at its least disciplined when the core business is generating record cash. None is large enough to matter financially today; the concern is directional.

10. Mobile and PC demand destruction as a stated risk

For the first time, memory management flagged that its own price increases may reduce unit demand downstream, citing higher end-product prices and reduced content per box from bill-of-materials pressure.

Assessment: Intellectually honest and strategically important. It is the first acknowledgment that the pricing environment has a ceiling set by customer economics rather than by supply. We would treat evidence of falling mobile and PC memory content as the earliest available indicator that this cycle is topping, ahead of any move in spot pricing.

Guidance & Outlook

Implied sequential path: with DRAM bits up low single digits, NAND bits up mid single digits, and pricing described as remaining robust, Q1 operating profit should rise again. MX turns from headwind to tailwind on the S26 launch, while foundry is guided seasonally lower. The composition of Q1 growth should therefore be broader than Q4's, which was almost entirely DS.

Guidance style: Samsung has now matched or beaten its own bit-shipment guidance for two consecutive quarters and overshot its capital spending guide by ₩5.3tn. The pattern is a company guiding conservatively on volume and spending aggressively on capacity.

Analyst Q&A Highlights

Decomposing the memory quarter into price and volume

The opening question sought the split between shipment growth and realisation. The answer was unusually precise and made clear that this was a price quarter, not a volume one.

Q: "It seems, yes, memory has achieved very solid performance in the fourth quarter. Could you provide more color on DRAM and NAND bit growth and also the rise in ASP? I'd appreciate more details."
— S.K. Kim, Daiwa

A: "As a result, bit shipments achieved a new record high, consistent with our bit growth guidance from the previous quarter. And, driven by higher overall market pricing and our product mix centered toward high value-added server products, DRAM ASP increased by about 40% quarter on quarter."
— Jaejune Kim, EVP of Memory

Assessment: The answer confirms that volume merely met plan while price did the work. For modelling purposes this is the cleanest disclosure of the quarter, and it sets the bar for Q1: with bits guided up only low single digits, the next quarter's earnings are again a bet on realisation rather than on shipments.

Qualification status and the shape of the 2026 HBM ramp

Where the October call met the qualification question with a confidentiality refusal, this one produced a detailed roadmap covering qualification stage, production status, shipment timing, packaging strategy, and a quantified sales trajectory.

Q: "It seems that we've been hearing quite a lot of good news recently about HBM4 performance from Samsung. Could you provide us an update on the status of your customer qualifications for HBM4 and your development plans for HBM4E, also an update on advanced packaging technology? Also, your outlook for expected HBM sales for 2026 as well."
— Peter Lee, Citigroup

A: "Even as major customers have been raising their performance requirements, we supplied sample shipments last year with no redesign required and have now entered the final phase of qualifications. Everything is proceeding smoothly."
— Jaejune Kim, EVP of Memory

Assessment: The contrast with the prior quarter's non-answer is the signal. A company in the final phase of qualification with production already running and February shipments scheduled is describing a commercial fact, not a hope. The residual risk is that "final phase" is not "passed," and that distinction has cost this company a full product generation before.

Whether the shortage extends beyond a single year

Questioning pressed on whether supply expansion could close the gap and how the portfolio would be managed if it could not. The response extended the shortage horizon across every product category and introduced the multi-year contracting dynamic.

Q: "It seems first for memory, AI demand has been growing faster than the pace of capacity expansions by memory suppliers, and market supply shortages appear to be worsening. So if you could explain more about your business operations and directions for 2026, including your plans for portfolio mix."
— Jay Kwon, JPMorgan

A: "We expect a significant shortage of supply relative to demand to continue across all product categories, whether it is HBMs, conventional DRAM or NAND, with tight undersupply conditions expected for the time being. We've already been receiving requests from large customers, including GPU or ASIC developers and hyperscalers, who are projected to experience a steep rise in demand for multi-year supply contracts."
— Jaejune Kim, EVP of Memory

Assessment: The breadth matters more than the duration. A shortage confined to HBM is an AI story; a shortage spanning HBM, conventional DRAM and NAND simultaneously is an industry-capacity story, and it accrues disproportionately to the supplier with the broadest portfolio. Management's stated intent to accept multi-year commitments only selectively is the right instinct, because locking in volume at today's prices would surrender the upside that makes this cycle valuable.

Where the increased capital budget actually goes

With supply constrained and demand described as unmeetable, questioning turned to whether capital would be deployed to close the gap. The answer clarified that spending targets future optionality rather than immediate output.

Q: "So, like you have said, there is a supply shortage, and you're expecting strong AI-related demand to persist in the long term. So then what is the direction for your CapEx investments for memory this year?"
— Sunwoo Kim, Meritz Securities

A: "The increased CapEx amount budgeted for this year, like I said, will go toward preemptively securing new fab space and clean rooms. This will help strengthen our future supply capacity while allowing us to hedge against market volatility risk."
— Jaejune Kim, EVP of Memory

Assessment: Building shells ahead of demand is the strategy that positioned Samsung for this cycle, and repeating it is rational. It also means investors should not expect the 2026 capital increase to relieve the 2026 shortage. Supply relief arrives in 2027 at the earliest, which is precisely when the depreciation from this spending begins to bite.

Foundry node progress and the customer pipeline

Questioning on advanced-node milestones drew the most concrete competitive disclosure the foundry business has offered, including a quantified order trajectory and confirmation of engagement following a major award.

Q: "Could you update us on the progresses for 2 nm and 1.4 nm nodes in particular? And whether you expect further major customer wins after the Tesla order last summer? And if so, in which segments those would be between mobile and HPC-AI."
— Nicolas Gaudois, UBS

A: "We've been hitting our yield and performance targets and are on track to start 2nm mass production the second half of the year. The 1.4 nm process is also under development, where we're hitting major milestones as planned, with the goal of starting mass production in 2029."
— Sukchae Kang, EVP of Foundry

Assessment: Hitting yield targets is the claim that matters, because yield is what has historically separated this business from its competitor. The 130% growth in 2nm project awards is a real datapoint. None of it changed the fact that provisions again absorbed the operating improvement, and until a quarter arrives where they do not, the division remains a story rather than a contributor.

Treasury share cancellation and the 2025 return arithmetic

Questioning on capital returns extracted the specific free-cash-flow basis, the split between regular and additional dividends, and a commitment on cancellation timing.

Q: "First, when will the company cancel the treasury shares it currently holds? And could you please explain the scale of shareholder return resources generated in 2025?"
— Hu Wenjie, Bank of America

A: "First, with respect to the treasury shares, the board will decide the cancellation schedule and make a public disclosure in the first quarter of 2026. In addition, out of the KRW 8.2 trillion worth of shares acquired in 2025, excluding those reserved for employees, KRW 6.6 trillion worth will be canceled to faithfully implement the shareholder return policy."
— Sooncheol Park, CFO

Assessment: A dated commitment replaces October's vague acknowledgment of "market interest," which is progress. ₩6.6tn of cancellation against a market capitalisation of this size is not transformative, but the direction is right and the Q1 disclosure is a checkable near-term deliverable.

The successor shareholder-return policy

With the current three-year framework expiring at the end of 2026 and free cash flow inflecting, questioning turned to what replaces it. Management confirmed a review without committing to timing or shape.

Q: "And 2026 is the final year of your 2024 to 2026 shareholder return policy cycle. So I know this is still, early on in the new year, but have you been examining the new direction for your next cycle?"
— Young-Ho Ryu, NH Investment & Securities

A: "First of all, we are faithfully implementing the current three-year shareholder return policy, and when it comes to a new shareholder policy, management and the board believe enhancing shareholder value is a top priority, and are actively reviewing a new active shareholder return policy based on sustainable growth."
— Sooncheol Park, CFO

Assessment: The word "active" is doing deliberate work and was almost certainly chosen. A policy set in 2023 against ₩36.5tn of free cash flow will look obsolete against the cash generation this business is about to produce. We regard the successor policy as one of the two largest non-operational catalysts of 2026.

What They're NOT Saying

1. Whether HBM4 qualification has actually passed: "final phase" and "proceeding smoothly" are not "qualified." Production has commenced and February shipments are scheduled, which implies confidence, but the company again stopped short of the definitive statement.
2. The 2026 CapEx number: a full quarter after promising a "significant" increase, the guide is still "meaningful" and unquantified, with detailed plans described as not finalised. The company overshot its 2025 guide by ₩5.3tn, so the absence of a figure is now a pattern rather than an oversight.
3. The size of the FX benefit by division: ₩1.6tn was disclosed at the company level and attributed to the component businesses, but not split out. Without it, the true operating improvement in DS cannot be isolated.
4. What the foundry provisions were for: the second consecutive quarter in which one-off costs capped earnings improvement, and the second consecutive quarter with no explanation of what they relate to or whether more are coming.
5. Any HBM revenue or share figure: "more than threefold" growth is given without a base, so the absolute scale of the HBM business and Samsung's share position remain unquantifiable from disclosure.
6. Pricing for 2026 contracted volumes: capacity is "fully booked with customer POs," but nothing was said about whether those orders are priced, and at what. In a rising market, a booked order at a fixed price is very different from a booked order at market.
7. The successor shareholder-return policy: under "active review" with no timing, no framework, and no indication whether the 50%-of-free-cash-flow basis survives into a period when free cash flow could multiply.
8. Any strategic answer on Visual Display and Appliances: the loss widened sixfold sequentially to ₩0.6tn and the response remains a 2026 product lineup. Still no cost action and no structural review.

Market Reaction

• Pre-print setup: shares closed at ₩162,400 (~$113.41) on January 28, up 35.4% year to date and 202.4% over trailing twelve months. The close was itself the top of the 52-week closing range of ₩51,000 to ₩162,400, so the stock entered the print at its highest close of the period.
• Reaction session (January 29): opened at ₩166,200 (+2.3% gap), traded as high as ₩166,600 (+2.6%) and as low as ₩157,100 (-3.3%), and closed at ₩160,700 (~$112.24), down 1.0% or ₩1,700.
• Volume: 36.1 million shares against a 30-day average of 27.3 million, roughly 1.3x normal.
• Relative: the KOSPI closed the same session up 1.0%, so the stock underperformed the index by roughly 2.0 percentage points. The KOSPI was up 22.7% year to date entering the print.

The intraday shape is more informative than the close. A 2.3% opening gap that reversed into a 1.0% decline, with a 5.9 percentage point swing between the session high and low, is a distribution pattern. Buyers arrived on the headline and sellers met them.

We attribute the fade to three things. First, there was no headline surprise available: the provisional release on January 8 had given the operating profit number, and consensus had converged to within 0.3%. Second, the stock had appreciated 35.4% in the four weeks of January alone and entered the print at a record close, so the bar was set by positioning rather than by estimates. Third, the call's genuinely new content was largely about 2027 rather than 2026, and a market paying 13.5x annualised current earnings for a memory company is not yet willing to underwrite a second year of shortage.

What did not cause the fade is equally worth stating. The HBM4 production disclosure, the threefold sales guide, and the fully-booked capacity commentary were all incrementally positive. A stock that absorbs disclosures of that quality and finishes red is telling you the good news is in the price, not that the news was bad.

Street Perspective

Debate: Does breaking the 2018 record mean this is the peak?

Bull view: The 2018 record was set with essentially no HBM business, no AI datacenter demand, and a foundry operation that has since added advanced nodes. The comparison flatters the past. With HBM4 shipping in February and 2026 HBM revenue guided to more than triple, the earnings base is structurally higher than it was in the prior cycle.

Bear view: Every memory peak in history has been announced with an explanation of why this time the cycle is structural. Operating margin has gone from 6.3% to 21.4% in two quarters almost entirely on price, and prices mean-revert.

Our take: The bull case is stronger but the margin of safety has narrowed sharply since our initiation. At 14.1% operating margin the "not a peak" claim was nearly self-evident; at 21.4% it depends on the 2027 supply picture being as tight as management says. We remain constructive, with materially less cushion than three months ago.

Debate: Is the fully-booked order book de-risking or a ceiling?

Bull view: Capacity sold out against purchase orders converts a spot-priced commodity business into something closer to contracted revenue, which justifies a higher multiple on the same earnings.

Bear view: Sold-out capacity means volume upside is gone for the year. All remaining earnings growth must come from price, in a market where management has itself flagged demand destruction in mobile and PC.

Our take: Both are true, and the resolution is in the pricing terms of those orders, which the company did not disclose. This is the single most important unanswered question from the call. If the booked volumes are priced at market, the bull case holds; if fixed, 2026 upside is substantially capped.

Debate: Should the foundry business be valued at all?

Bull view: 2nm awards up 130%, a Tesla anchor customer, 4nm HBM base-die shipments started, and a one-stop logic-memory-packaging offering no competitor can match. A 2026 target of double-digit revenue growth with better profitability.

Bear view: Two consecutive quarters where one-off charges consumed the operating improvement, no disclosed path to breakeven, and Taylor adding fixed cost during 2026 before it adds meaningful revenue.

Our take: We continue to carry this at zero, which is now arguably too conservative given the order trajectory. We would rather be late to credit a turnaround in a business with this history than early. The 4nm HBM base-die shipment is the datapoint we will be watching, because it is the one that makes foundry strategically inseparable from the memory franchise.

Model Update Needed

Valuation. At the ₩160,700 close the shares trade on 23.9x FY2025 EPS of ₩6,711.61, a multiple distorted by two weak quarters in the base, and on 13.5x annualised Q4 earnings. Against our revised FY2026 range of ₩13,000-16,000 the shares are on 10x to 12x.

Price target: ₩190,000, raised from ₩130,000, implying roughly 18% upside from ₩160,700. The target applies approximately 13x to the midpoint of our FY2026 EPS range. Note what has and has not changed: the earnings estimate has roughly doubled while the applied multiple has fallen from 16x to 13x. That is deliberate. As a memory business moves through an up-cycle, the appropriate multiple on its earnings should decline, because each successive year of peak-like earnings is less likely to repeat.

Thesis Scorecard Post-Earnings

Scored against the standing thesis established at initiation in October, and against the specific commitments management made on that call.

Management commitments from the October call, graded

Overall: Thesis strengthened on the operating evidence and weakened on the risk/reward. Management delivered or beat on six of eight commitments. The uncomfortable truth is that the thesis worked so quickly that the stock now discounts much of it, and the marginal buyer is being asked to underwrite 2027 rather than 2026.

Action: Maintain Outperform, target raised to ₩190,000. We would not add to a position at these levels with the same conviction we held at ₩104,100. The specific things that would change this rating: pricing terms on the booked 2026 order book proving to be fixed rather than market, evidence of memory content per device declining in mobile and PC, or operating margin reaching the mid-20s without the 2027 order book firming.