A ₩107 Trillion Quarter Absorbed a ₩10 Trillion Currency Hit. The Market Sold the Payout Delay Instead.
Key Takeaways
- Operating profit of ₩107.44tn (~$80.3B) set a fourth consecutive record, and it carried a currency headwind of about ₩10tn. Operating margin rose to 55.0% from 52.2% while the dollar fell 12.3% against the won between quarter-ends. On management's own estimate of the currency effect, the quarter would have printed near ₩117tn at second-quarter exchange rates. Revenue of ₩195.25tn missed consensus by 2.7%, the second quarter in a row in which the top line disappointed at the provisional release.
- The contract book doubled in three months. Ten major accounts are now signed on multi-year terms, against the top five at the July call, and ten more are in negotiation. Signed customers are asking for extensions and for extra volume inside years they have already contracted, and 2027 HBM is sold out at agreed target prices. HBM share reached parity with Samsung's overall DRAM share a quarter ahead of plan.
- The weak lines sit below the operating line or are already reversing. A ₩9.5tn net finance expense pushed pre-tax profit below operating profit, so EPS rose 6% while operating profit rose 20%. Bit shipments undershot the July guide for both DRAM and NAND, leaving a DRAM price increase of about 20% to do the work, but the fourth-quarter guide calls for DRAM bits up by the mid-to-high teens.
- The 4.1% fall on the day was about the payout, not the print. The board confirmed ₩55-75tn of 2026 returns still to come, including a substantial buyback and cancellation, but deferred the amount and the successor policy to a late-January 2027 meeting. Full-year capital spending was finally quantified at ₩84.4tn, above our ₩60-70tn estimate.
- Rating: Maintaining Outperform; target trimmed to ₩245,000 from ₩260,000. None of the three triggers we set in July fired, and the contract pillar that drove the upgrade strengthened. The target falls only because the stronger won lowers our normalised earnings in won by about 6%. It implies 6% upside from the ₩232,000 reaction close, before ₩55-75tn of pending 2026 returns.
Results vs. Consensus
| Metric | Actual | Consensus / Benchmark | Beat/Miss | Magnitude |
|---|---|---|---|---|
| Revenue | ₩195.25tn (~$145.9B) | ₩200.64tn; ~₩206.8tn (second estimate set) | Miss | -2.7% / -5.6% |
| Revenue vs. own guide | ₩195.25tn | ₩194-196tn (median ~₩195tn) | In line | +0.1% |
| Operating profit | ₩107.44tn (~$80.3B) | ₩106.64tn | Beat | +0.8% |
| Operating profit vs. own guide | ₩107.44tn | ₩107.40tn (median) | In line | +0.04% |
| DS division operating profit | ₩106.5tn | ₩107.3tn (one estimate set; a second report confirms a miss, without a level) | Miss | -0.7% |
| Gross margin | 69.3% | n/a | n/a | -30bps QoQ |
| Operating margin | 55.0% | n/a | n/a | +280bps QoQ |
| Net profit | ₩75.1tn (~$56.1B) | n/a | n/a | +4.9% QoQ |
| EPS (common, company basis) | ₩11,462 (~$8.57) | n/a | n/a | +5.7% QoQ |
Year-over-Year Comparison
| Metric | Q3 2026 | Q3 2025 | Change |
|---|---|---|---|
| Revenue | ₩195.25tn | ₩86.06tn | +126.9% |
| Gross profit | ₩135.3tn | ₩33.5tn | +303.9% |
| Gross margin | 69.3% | 38.9% | +3,040bps |
| Operating profit | ₩107.44tn | ₩12.17tn | +783% |
| Operating margin | 55.0% | 14.1% | +4,090bps |
| Net profit | ₩75.1tn | ₩12.2tn | +515.6% |
| EPS (common) | ₩11,462 | ₩1,802 | +536.1% |
Sequential Comparison
| Metric | Q3 2026 | Q2 2026 | Change |
|---|---|---|---|
| Revenue | ₩195.25tn | ₩171.50tn | +13.8% |
| Gross profit | ₩135.3tn | ₩119.3tn | +13.4% |
| Gross margin | 69.3% | 69.6% | -30bps |
| SG&A (incl. R&D) | ₩27.8tn | ₩29.8tn | -6.7% |
| R&D | ₩14.4tn | ₩16.0tn | -10.0% |
| Operating profit | ₩107.44tn | ₩89.49tn | +20.1% |
| Operating margin | 55.0% | 52.2% | +280bps |
| Finance income / (expenses) | (₩9.5tn) | ₩4.7tn | Swing of (₩14.2tn) |
| Profit before tax | ₩98.5tn | ₩94.4tn | +4.3% |
| Net profit | ₩75.1tn | ₩71.6tn | +4.9% |
| EPS (common) | ₩11,462 | ₩10,849 | +5.7% |
| CapEx | ₩29.0tn | ₩16.8tn | +₩12.2tn |
Nine Months of 2026 vs. Full Year 2025
| Metric | 9M 2026 (9 months) | FY2025 (12 months) | Ratio |
|---|---|---|---|
| Revenue | ₩500.62tn | ₩333.61tn | 1.50x |
| Operating profit | ₩254.16tn | ₩43.61tn | 5.8x |
- Currency was a headwind of about ₩10tn. Management put the sequential effect of a weaker dollar, euro and other currencies at around ₩10tn of operating profit, mainly in components. That is the opposite of Q2, when currency added ₩3.1tn. The 55.0% margin was earned against the exchange rate, not with it, which makes the operating result higher quality than the 20% sequential growth suggests.
- The non-operating line ran the other way this quarter. The usual KRX check is for net income flattered by investment gains. Here net finance income swung to a ₩9.5tn expense from a ₩4.7tn gain, pre-tax profit of ₩98.5tn fell below operating profit, and net profit grew only 4.9%. Neither the deck nor the call broke out the line. The dollar fell 12.3% between quarter-ends and investments on the balance sheet fell ₩7.6tn, so translation and valuation losses are the likely candidates, but that is our inference. We anchor the quarter to the operating line, as we would if the gain were flattering it.
- The deferred incentive cost did not show. We warned in July that the capitalised part of the first-half incentive accrual would enter cost of sales this quarter. Gross margin slipped only 30bps. Selling, general and administrative expense fell ₩2.0tn and R&D fell 10%, which is consistent with the first-half catch-up in Q2 not recurring. Management did not quantify either effect.
- Price, not volume, did the work. DRAM bits rose by a low-to-mid single digit and NAND bits by a mid single digit, both below the July guide. Average selling prices rose about 20% in DRAM and by the high teens in NAND.
- A small one-off. Harman's ₩0.6tn operating profit included non-recurring US tariff refunds.
Revenue. Revenue of ₩195.25tn rose 13.8% sequentially and 126.9% year over year, with Device Solutions revenue of ₩147.1tn up 15% and the memory business alone at ₩140.7tn, up 16%. The consumer division grew 8% on foldable launches. The 2.7% miss against consensus is the second consecutive quarter in which the top line fell short at the provisional. This time supply, not demand, appears to be the constraint: management described managing the product mix "amid capacity constraints".
Margins. A 55.0% consolidated operating margin, with DS at 72.4%, is a level we have repeatedly said we do not regard as sustainable, and nothing this quarter changes that. The point worth making is narrower: margin expanded 280bps through a roughly ₩10tn currency headwind and through the accounting drag we expected from the incentive accrual. Pricing power at the operating line is intact.
Earnings. EPS of ₩11,462 rose 5.7% sequentially, a quarter of the operating growth, because of the finance line. Against the ₩232,000 reaction close the shares trade at 5.1x annualised third-quarter EPS. As in every note since initiation, we place no weight on that multiple; the valuation section works from normalised earnings.
Segment Performance
| Division | Revenue | Operating profit | Operating margin | QoQ revenue | Notable |
|---|---|---|---|---|---|
| Device Solutions (DS) | ₩147.1tn | ₩106.5tn | 72.4% | +15% | 99.1% of consolidated operating profit |
| of which Memory | ₩140.7tn | n/a | n/a | +16% | Record revenue and operating profit |
| MX / Networks | ₩37.7tn | (₩0.6tn) | -1.6% | +14% | Loss narrowed from (₩0.7tn) |
| VD / DA | ₩13.9tn | (₩0.2tn) | -1.4% | -4% | Loss widened from (₩0.01tn) |
| Samsung Display (SDC) | ₩8.0tn | ₩1.1tn | 13.8% | +7% | Up from ₩0.7tn on smartphone OLED |
| Harman | ₩4.3tn | ₩0.6tn | 14.0% | -5% | Includes one-off US tariff refunds |
| Consolidated | ₩195.25tn | ₩107.44tn | 55.0% | +13.8% | Divisional figures include intersegment sales |
Divisional revenue includes intersegment sales and does not sum to the consolidated total. The Total row is the consolidated income-statement figure. The DX division (MX/Networks plus VD/DA) reported ₩52.1tn of revenue and a (₩0.8tn) operating loss.
Memory: Bit Shipments and Pricing
| Product | Q3 bit shipments (QoQ) | July guide for Q3 | Q3 ASP (QoQ) | Q4 bit guide (QoQ) |
|---|---|---|---|---|
| DRAM | +low to mid-single digit % | +mid-single digit % | +~20% | +mid to high teens % |
| NAND | +mid-single digit % | +high single digit % | +high teens % | +mid-teens % |
| HBM | HBM4 sales >3x QoQ; HBM4 at 60% of HBM sales | 2027 volume sold out at agreed target prices; HBM4E mass production H1 2027 | ||
DRAM: price did the work, again
DRAM bits grew by a low-to-mid single digit, short of the mid-single-digit guide, while the average selling price rose about 20%. Management attributed the price strength to server demand and said prices rose across every application. The deceleration from the low-90% price increase of the first quarter is large, but a 20% sequential increase is still a price spike by the standards of any prior cycle.
"Therefore, compared to the previous quarter, our DRAM ASP rose by 20% level, and our NAND ASP also increased by the high teens percentages."
— Jaejune Kim, EVP, Memory Global Sales and Marketing
Assessment: The bit shortfall is the more informative number. A supplier running short of its own volume guide in a market where customers are asking for more is capacity-constrained, not demand-constrained, and the fourth-quarter guide of mid-to-high-teens bit growth says the constraint is easing. That combination, volume returning just as price gains slow, is the pattern we would expect from a supplier entering the contracted phase of the cycle rather than the spot-driven one.
NAND: the agentic-AI storage case gets specific
NAND bits rose by a mid single digit, below the high-single-digit guide, with prices up by the high teens. The more important disclosure was demand composition: management described large incremental QLC orders tied to cloud-based personal AI agents and rising demand for high-capacity TLC drives to hold KV cache for agentic inference, and guided fourth-quarter NAND bits up by the mid-teens.
"On top of that, on the NAND side, we have received additional QLC demand in a large volume, in conjunction with a surge in storage demand triggered by the expansion of cloud-based personal AI agent services."
— Jaejune Kim, EVP, Memory Global Sales and Marketing
Assessment: This is the first time the NAND demand case has been tied to a mechanism specific enough to test: per-user virtual machines in the cloud and KV-cache offload. It moves NAND further from the consumer-device cycle that has historically set its price. One soft spot: the V10 node, guided in July for August mass production, was described this time as "production ready", which is not the same claim.
HBM: parity a quarter early, 2027 sold out
HBM4 sales more than tripled sequentially, as guided in July, and reached 60% of HBM revenue against July's guide of "well over 60%" for the second half as a whole. HBM share reached parity with Samsung's overall DRAM share in the third quarter, against a second-half target, and 2027 HBM volume is sold out at agreed target prices with key customers still asking for more. HBM4E, sampled first in the industry in May, enters mass production in the first half of 2027 with a volume ramp in the second half.
"Thanks to these solid results in HBM4, we have actually achieved our goal of bringing HBM market share in line with our overall DRAM market share already in the third quarter. This is ahead of our original target, which was the second half of the year."
— Jaejune Kim, EVP, Memory Global Sales and Marketing
Assessment: The competitive reversal we described in July is now complete on the company's own measure. "Sold out at an agreed target price" for 2027 matters more than the share claim, because HBM is priced annually in advance: next year's HBM revenue is substantially fixed today. The unresolved question is whether a lead customer moving some platforms to shorter stacks, as reported in late October and not addressed on the call, compresses the premium from 2028.
Foundry and System LSI: earnings improving, demand outrunning capacity
Foundry revenue and earnings improved significantly on 4nm HBM base-die sales and advanced-node demand, with 2nm orders from hyperscale and AI customers continuing. Management guided double-digit fourth-quarter revenue growth both sequentially and year over year, expects 2nm project orders to more than double in 2026, and expects every node below 8nm to run above 100% utilisation in 2027. Taylor Fab 1 has started clean-room operations and wafer input for line setup, although the release now places the start of its operations in the 2027 outlook. System LSI grew revenue on new-model SoCs and 200-megapixel sensors but expects the mobile slowdown to continue into 2027.
"Furthermore, as demand across all nodes below 8nm is projected to substantially exceed production capacity, we expect them to be running at over 100% utilization."
— Sukchae Kang, EVP, Foundry
Assessment: A foundry projecting over-100% utilisation on its advanced nodes next year has a capacity problem, not a customer problem, and that is the transition this division has needed for three years. We still carry foundry at zero in our target. Management again did not say whether the division is profitable, so we keep it there, but the case for doing so is getting harder to defend.
Mobile eXperience and Networks: the loss narrows
MX and Networks lost ₩0.6tn on ₩37.7tn of revenue, a slightly smaller loss than in the second quarter, as foldable launches lifted the flagship mix and average prices. Networks earnings fell on lower North American and software sales. Management guided fourth-quarter MX revenue and profit lower and expects memory costs to stay high into 2027.
"Although profit declined year-on-year, it improved quarter-on-quarter, supported by higher ASPs from an improved flagship sales mix."
— Daniel Araujo, VP, Mobile eXperience
Assessment: The handset business is now a priced-in cost of owning the memory business. The narrowing is real but small, the fourth-quarter guide reverses it, and the cost driver is Samsung's own memory pricing. Nothing here moves the investment case in either direction.
Display, VD/DA and Harman
Samsung Display improved to ₩1.1tn of operating profit on ₩8.0tn of revenue on smartphone OLED supply for new launches, though management expects 2027 demand to be weak. Visual Display and Appliances widened their loss to ₩0.2tn on the post-tournament television slowdown and higher raw-material and logistics costs. Harman's ₩0.6tn included one-off tariff refunds.
Assessment: The consumer complex remains at roughly breakeven in aggregate. The only signal worth noting is that the display business expects weak demand next year from higher device prices, which is the second-order cost of the memory shortage showing up inside the group.
Key Topics & Management Commentary
Overall Management Tone: Confident and more specific than in July on demand, contracts and HBM, and more guarded on everything that touches the shareholder. Management answered the currency question with a policy description rather than an explanation of the ₩9.5tn finance line, and the capital-return update restated an envelope the market already knew. Compared with July, the tone on supply duration was unchanged while the industry peer reporting a day earlier went two years further, and that difference was the one the market noticed.
1. The first ₩100 trillion quarter, earned against the currency
Operating profit of ₩107.44tn made Samsung the first technology company to exceed ₩100tn of quarterly operating profit. Nine-month operating profit of ₩254.16tn is 5.8x the whole of FY2025.
"On currency effects, weaker exchange rates for the USD, EUR, and major other currencies had a negative impact of around 10 trillion KRW on operating profit compared with the previous quarter, mainly in our component business."
— Daniel Oh, Head of Investor Relations
Assessment: The currency figure is the most important number on the call for judging the quality of the quarter. At second-quarter exchange rates, operating profit would have been near ₩117tn, about 31% above the second quarter. The won does not stay strong or weak forever, and we treat it as a valuation input rather than an operating one, which is why it moves our target below.
2. A finance loss turns the usual KRX earnings-quality problem upside down
For KRX memory names the recurring quality-of-earnings issue has been non-operating gains flattering net income, most visibly at SK hynix. Samsung has not had that problem in our coverage, and this quarter it reported the reverse: a ₩9.5tn net finance expense took pre-tax profit ₩8.9tn below operating profit, so net profit grew 4.9% on 20.1% operating growth.
"Rather than using costly derivatives to artificially hedge, we focus on leveraging natural hedges across our global value chain to structurally minimize the impact of FX volatility."
— Soon-Cheol Park, CFO
Assessment: The natural-hedge policy explains why the operating line carries the currency directly and why there is no derivative gain to offset it. It does not explain the finance line, which neither the deck nor the call broke out. Investors screening on EPS growth will read this quarter as a deceleration to mid-single digits. On the operating line it was an acceleration.
3. Ten signed contracts, and customers asking for more
The multi-year contract programme that drove our July upgrade doubled in signed counterparties. Ten more major accounts are in negotiation, and the interest is spreading from data-centre buyers into mobile, automotive and network applications.
"In terms of the number of customers under multi-year contracts, we've already concluded contracts with 10 major accounts. The number of other customers seeking more substantial long-term memory supply are increasing as well. At present, we are in talks with 10 other major accounts for multi-year contracts, and we are near the final phase for several of these clients."
— Jaejune Kim, EVP, Memory Global Sales and Marketing
Assessment: The direction of travel is the opposite of what our Underperform trigger was built to catch: counterparties are seeking extensions and extra volume, not renegotiating down. The 60-70% capacity target described as "gaining greater visibility" is now credible. The one caution is that customers requesting additional supply for years they have already contracted is what over-ordering looks like at its start, and we add it to our watch list below.
4. HBM sold out for 2027 at target prices
HBM4 tripled as guided and reached 60% of HBM revenue, with the fourth quarter left to carry the second half to July's "well over 60%". Parity with DRAM share came a quarter early, and next year's HBM volume is fully booked at agreed prices.
"As our competitive HBM offerings continue to receive high marks, our 2027 HBM supply volume is already sold out at an agreed target price that reflect our product competitiveness. Yet, we continue to receive requests from key customers for additional supply."
— Jaejune Kim, EVP, Memory Global Sales and Marketing
Assessment: Because HBM is contracted annually in advance, 2027 HBM revenue is now largely determined, and on the company's account it was set at a price reflecting product competitiveness rather than a discount to win share. Two years after being the supplier that missed HBM3E, Samsung is pricing HBM from a position of parity. Our target still places no premium on that.
5. A volume step-up in the fourth quarter
After two quarters of price-led growth with bits short of guidance, management guided a sharp acceleration in fourth-quarter shipments, with demand for products tied to new CPU and GPU platforms still outpacing production.
"Therefore, in the fourth quarter, we expect QoQ bit growth to increase by mid to high teen percentage for DRAM, and mid-teens increase for NAND."
— Jaejune Kim, EVP, Memory Global Sales and Marketing
Assessment: This is the strongest bit guide in our coverage period. Mid-to-high-teens DRAM bit growth on top of any further price increase would lift fourth-quarter operating profit well above the third quarter's, even after another currency headwind if the won holds near ₩1,338. It also marks the shift from price to volume that historically precedes a slowdown in price, so we read it as good for the next two quarters and neutral for the cycle.
6. The 2028 horizon, and why it read as cautious
Management repeated the July view that supply stays tight into 2028 and gave a detailed argument for why record industry capital spending will not translate quickly into bits: spending on clean-room space, higher construction and equipment prices, prepayments to equipment suppliers, longer lead times, and larger die sizes as HBM4E and HBM5 ramp.
"Considering the long lead time from new fab construction to actual wafer production, the increase in CapEx in 2027 is unlikely to result in a meaningful boost to bit production in 2028."
— Jaejune Kim, EVP, Memory Global Sales and Marketing
Assessment: The supply argument is the most rigorous version we have heard from this management. The market's problem was relative: the industry peer that reported a day earlier said demand would outstrip supply through 2030, and Samsung's horizon stopped two years shorter. We prefer the shorter, better-argued claim; a supplier that names the limit of its visibility is more credible than one that does not. But we note that "unchanged" from July is not "extended", and the shares are priced on extensions.
7. Capital spending finally quantified, and higher than we thought
After four quarters without a figure, the company gave full-year 2026 capital spending of ₩84.4tn, up ₩31.7tn from 2025, with ₩76.8tn in Device Solutions. Nine-month spending was ₩56.9tn, implying about ₩27.5tn in the fourth quarter.
"For the full year of 2026, we expect total CapEx to reach 84.4 trillion KRW or 31.7 trillion KRW increase from last year."
— Daniel Oh, Head of Investor Relations
Assessment: Disclosure delivered, and the number is 20-40% above our ₩60-70tn estimate. Spending up 60% in a year at peak margins is the classic precondition for a future oversupply, and the contracted, floor-priced volume is what stops us from treating it as one.
8. ₩30 trillion paid, ₩55-75 trillion still to come, and a January date
The board approved a third-quarter dividend of ₩4,601 per share, about 12 times the regular quarterly payout, totalling ₩30tn. Of the ₩90-110tn 2026 return envelope announced in August, ₩35tn is planned or completed and ₩55-75tn remains, to be returned through dividends and a substantial buyback and cancellation programme. The amount, the method and the successor policy will be set at a late-January 2027 board meeting.
"The new shareholder return policy, along with the final amount of shareholder returns and the specific execution plans for the remaining funds for 2026 will be finalized at the board meeting scheduled for late January 2027, with an announcement to follow."
— Soon-Cheol Park, CFO
Assessment: The January meeting first mentioned in August is now fixed for late January, and the envelope it will fill is confirmed. The disappointment is understandable: the dividend was pre-announced in August, nothing new was sized, and the record date for the ₩4,601 was September 30, so a buyer today does not receive it. But ₩55-75tn is still to be returned within the 2026 envelope, roughly ₩8,400-11,500 per share across common and preferred, or 3.6-5.0% of the reaction close. That is the most concrete catalyst in the story, and it now has a date.
9. The preferred-share discount
Management tied the buyback mix explicitly to the gap between common and preferred shares. The preferred line closed at ₩183,900 on the print day, 20.7% below the common.
"In addition, we recognize the growing market interest in the price gap between common and preferred shares, and we are committed to actively considering and adjusting the proportion of preferred shares in our share repurchase and cancellation program based on market conditions with a view of ensuring equal respect to all other shareholders, optimizing capital allocation, and enhancing corporate value."
— Soon-Cheol Park, CFO
Assessment: Buying the preferred shares at a 21% discount to the common is the highest-return use of buyback capital available to the company, since both lines carry the same dividend. Weighting the cancellation programme toward them would be the clearest evidence yet that capital allocation is being run for per-share value.
10. 2027 guided as another record year
The CFO guided group performance above 2026, with memory demand far exceeding supply and foundry growth accelerating, while the consumer businesses face continued cost pressure and some demand adjustment in mobile and PC.
"Driven by strong growth momentum in the DS division, we are positioned to exceed our 2026 performance and achieve another year of record results."
— Soon-Cheol Park, CFO
Assessment: This is as close as a Korean issuer comes to a full-year guide, and it is unconditional on the semiconductor side. It matters less for our target, which is built on normalised earnings, than for the cash it implies: a second year of peak earnings adds materially to a net cash position already at ₩197.6tn, and half of free cash flow goes back to shareholders under the current policy.
11. Foundry capacity and the Taylor ramp
Foundry is looking to add investment in new fabs in the US and Korea, plus advanced packaging, because 2nm and AI demand is growing faster than capacity. Taylor Fab 1 is in equipment installation and line setup.
"To meet future demand from global customers, we are evaluating further options to add capacity, including building additional fabs."
— Sukchae Kang, EVP, Foundry
Assessment: The July call committed to starting Taylor Fab 2 construction by year-end. This call talked about "additional fabs" in general and did not update that specific commitment. Read with the ₩84.4tn capital budget, foundry is the part of the spending plan most exposed to a demand disappointment, because unlike memory it has no floor-priced contracts under it yet.
Guidance & Outlook
| Item | Q3 2026 actual | Q4 2026 / 2027 guide | Direction |
|---|---|---|---|
| DRAM bit shipments | +low to mid-single digit % QoQ | +mid to high teens % QoQ in Q4 | Accelerating |
| NAND bit shipments | +mid-single digit % QoQ | +mid-teens % QoQ in Q4 | Accelerating |
| Memory pricing | DRAM +~20%, NAND +high teens | Not guided; demand to outpace production | Unquantified |
| HBM | HBM4 >3x QoQ, 60% of HBM | 2027 sold out at target price; HBM4E mass production H1 2027 | Booked |
| Supply tightness | Tight | Into 2028; 2027 tighter than 2026 | Unchanged |
| Multi-year contracts | 10 accounts signed | 10 more in talks; 60-70% of capacity targeted | Expanding |
| Foundry | Earnings improved significantly | Double-digit revenue growth QoQ and YoY in Q4; nodes below 8nm >100% utilised in 2027 | Improving |
| MX | (₩0.6tn) with Networks | Q4 revenue and profit down QoQ; cost pressure into 2027 | Pressured |
| SDC | ₩1.1tn operating profit | Limited Q4 improvement; weak 2027 demand | Softening |
| Group 2027 | n/a | Exceed 2026; another record year | Up |
| FY2026 CapEx | ₩29.0tn in Q3; ₩56.9tn 9M | ₩84.4tn (DS ₩76.8tn, SDC ₩3.9tn) | Higher than our estimate |
| 2026 shareholder return | ₩30tn Q3 dividend | ₩55-75tn remaining; set at late-January 2027 board | Deferred |
As a KRX issuer Samsung guides bit shipments and frames price qualitatively, so there is no revenue or margin range to hold it to. What the guide does say is that the business moves from price-led to volume-led growth in the fourth quarter, with demand still running ahead of production, and that management expects 2027 to be another record year.
Implied fourth quarter: mid-to-high-teens DRAM bit growth and mid-teens NAND bit growth, with prices still rising, point to fourth-quarter operating profit above the third quarter's ₩107.44tn even if the won holds near ₩1,338 and adds another currency headwind. We model ₩110-125tn.
Street at: before the print, consensus had fourth-quarter operating profit above ₩100tn, a full year near ₩375tn and 2027 clustered around ₩550tn. Our full-year range of ₩365-380tn brackets the Street.
Guidance style: unchanged and conservative in form. Both July bit guides were missed this quarter, which argues for treating the fourth-quarter bit guide as a ceiling set by capacity rather than a floor.
Analyst Q&A Highlights
How long the shortage lasts, and whether record capex changes it
The call opened on the horizon beyond 2027. Management answered with a structured case for why capital spending will not become bits on the usual lag, and kept its conclusion at 2028 with an explicit caveat about macro events.
Q: "For 2027, I think the dominant view is projecting that there will be a memory shortage next year. How about looking beyond 2027? What is your outlook for memory in 2028?"
— Dong Kim, KB Securities
A: "First of all, much of the CapEx increase will be budgeted toward acquiring new space. Also, higher construction costs may be making the CapEx numbers look bigger than they actually are." … "On balance, with all things considered, we expect tight supply conditions to continue in 2028."
— Jaejune Kim, EVP, Memory Global Sales and Marketing
Assessment: The capital-spending argument is the strongest available rebuttal to the oversupply bear case, and it is specific. The plan to build clean rooms first and install equipment flexibly is also the right hedge, since it keeps the expensive decision reversible. The answer did not extend the horizon, and on a day when the peer extended to 2030 that was enough to cost the shares.
The multi-year contract book
Questioning asked for an update on the contracts disclosed in July. Management doubled the signed count, described a broadening customer base and reported demand for extensions above its own expectation.
Q: "At your second quarter earnings call, I think you did provide a detailed status update on multi-year contracts. So I imagine there may have been many changes since then. To the extent possible, could you share some more details?"
— Sunwoo Kim, Meritz Securities
A: "We are also seeing much stronger demand for additional extensions than we had initially expected. Also, we have been receiving requests for additional supply for periods that are already covered under existing contracts."
— Jaejune Kim, EVP, Memory Global Sales and Marketing
Assessment: The best answer on the call for our thesis. The answer gave no update on price floors or prepayments, the two terms that make the contracts economically meaningful, and that omission is now the main gap in the disclosure. The extension demand, though, is the market voting with multi-year commitments on how tight supply will be.
HBM4 status and 2027 negotiations
Questioning asked for the HBM position and the state of next year's negotiations. Management confirmed every July target and said the 2027 book is closed at target pricing.
Q: "Could you provide a status update on your HBM business? How are your negotiations for 2027 proceeding also for HBM?"
— Seicheol Lee, Citi Securities
A: "Thanks to differentiated performance of our HBM4 sales increased by more than three-fold QoQ in the third quarter. Also HBM4 mix increased to 60% of total HBM sales. This is consistent with our HBM4 sales target, which we shared with you at our second quarter earnings call."
— Jaejune Kim, EVP, Memory Global Sales and Marketing
Assessment: A supplier that names a target, hits it and says so is building the credibility it will need when the cycle turns. The answer did not engage with reports that a lead customer may shift some platforms to shorter stacks, which is the next HBM question the market will ask.
Agentic AI as a NAND demand driver, and the cost squeeze in mobile
A two-part question tested whether personal AI agents create real storage demand and how the mobile business is coping with memory costs. Management gave a mechanism for the first and a holding pattern for the second.
Q: "What is your outlook on this type of development driving NAND demand, primarily for server SSDs?"
— Jay Kwon, JPMorgan Securities
A: "Also, with the introduction of cloud-based personal AI agent services, we have observed a new trend of physically allocating dedicated virtual machines to individual users in the cloud. We're seeing a growing preference for QLC SSDs to reduce service latency and improve throughputs. As a result, as the number of personal AI agent service users increases, we expect a proportional increase in QLC SSD demand as well."
— Jaejune Kim, EVP, Memory Global Sales and Marketing
Assessment: Demand that scales with user count rather than with hyperscaler capital budgets is a more durable kind of demand, if it materialises. On mobile, management expects high memory costs to persist into 2027 and a shrinking unit market, which confirms that the consumer divisions will stay a drag for as long as the memory thesis works.
Foundry capacity in Korea and the US
Questioning asked whether order wins will outrun capacity. Management confirmed they already are and described Taylor's progress and further capacity options.
Q: "As you significantly increase your order wins, you may experience some shortage in capacity. Are you planning any additional capacity expansions for either Korea or the U.S. market?"
— Simon Woo, Bank of America Securities
A: "For the Taylor site, we have commenced clean room operations at Fab 1, started equipment installations, also wafer input for line setup is underway, and specialists have been sent from the head office, and the project is proceeding as planned."
— Sukchae Kang, EVP, Foundry
Assessment: Clean-room operations and line-setup wafer input mean Taylor Fab 1 is physically under way, and management called the project on plan. But the release now lists the start of its operations in the 2027 outlook, where July said "by 2026", and nobody asked which plan. The broader capacity answer is where the risk sits: foundry is evaluating more fabs on order momentum alone, without the contract floors that underwrite memory spending.
Currency hedging
Questioning raised the earnings drag from the weaker dollar directly. The CFO described a natural-hedge policy and explained why derivatives are not used at Samsung's scale.
Q: "First, the recent decline in the dollar on exchange rate has been weighing on your earnings. Could you walk us through your FX hedging strategy and how you manage currency risk?"
— Dong Han, SK Securities
A: "Through these natural hedges, we are striving to minimize open FX positions. For any remaining positions, we sell dollars early and take other steps to keep the FX exposure period as short as possible."
— Soon-Cheol Park, CFO
Assessment: The policy is sensible for a company whose annual dollar receivables run to hundreds of trillions of won. Its consequence is that shareholders carry the currency through the income statement, and this quarter that cost about ₩10tn at the operating line. The answer did not address the ₩9.5tn finance expense, which was the question behind the question.
Foundry differentiation and the base-die advantage
Questioning asked what differentiates Samsung's foundry against stronger competition. Management pointed to the memory-foundry combination and disclosed a 1.4nm test-chip plan with a customer.
Q: "Given more intense competition, could you highlight what your technological differentiation may be?"
— Sungkyu Kim, Daiwa Capital Markets
A: "By jointly developing an HBM4 base die based on the 4nm process, we together have achieved industry-leading HBM4 performance. Notably, we also achieved industry-leading base die yields in the Foundry industry, although we are still early in mass production."
— Sukchae Kang, EVP, Foundry
Assessment: The base die is the one product where Samsung's integrated model is a genuine advantage over a pure-play foundry, because the customer is Samsung's own memory business. It is a real but captive franchise. The external test is 2nm HPC volume from third-party customers in the second half of 2027.
What They're NOT Saying
- Where the price floors sit: the term that drove our upgrade was not mentioned on the call or in the deck. A second quarter without a level or a relationship to spot leaves the contracts' downside protection unmeasured.
- The size of the advance payments: not updated since "one-fourth received" in July. The deck's balance sheet does not isolate them: the line for unearned revenue and other advances fell to ₩3.8tn from ₩4.5tn, so investors cannot see the balance from the filings either.
- What drove the ₩9.5tn finance expense: the deck shows the line and the call discussed hedging policy, but neither said whether the loss was currency translation, investment valuation or something else.
- Why bits missed the July guide: DRAM and NAND shipments both landed below guidance, and management did not explain the shortfall, although its own description of capacity constraints suggests supply.
- The incentive accrual: neither the third-quarter accrual nor how much of the capitalised second-quarter amount went through cost of sales was discussed, after an unusually detailed explanation in July.
- Foundry profitability: earnings "improved significantly" again, with no statement of whether the division made money and no breakeven quarter.
- Taylor timing: the July commitment to begin Fab 2 construction by the end of 2026 was not repeated, and the start of Fab 1 operations, "by 2026" in July, now sits in the 2027 outlook without comment.
- The competitive news of October: neither management nor the analysts raised the reported Chinese advance in DRAM cell density or the reported shift by a lead accelerator customer toward shorter HBM stacks, both reported this month.
Market Reaction
All price levels are FMP-verified KRX closing prices; the local benchmark is the KOSPI (^KS11). Samsung discloses before the Korean open, so the reaction is the print-day session.
- Pre-print setup: shares closed at ₩242,000 (~$180.88) on October 28, up 9.0% that session as the market rallied on SK hynix's results. They entered the print up 101.8% year to date and 140.8% over the trailing twelve months, but down 11.2% over thirty days and 33.2% below the 52-week closing high of ₩362,500. The KOSPI was up 42.7% year to date.
- Between the prints: from the ₩207,000 close of our July report session the shares reached ₩285,500 on September 23, then fell 2.4% on the October 8 provisional, 7.4% on October 13 when China's CXMT announced its DRAM cell-density progress, and 9.5% on October 15 to a ₩210,000 low close, with the KOSPI falling 5.5% and 7.6% on the last two of those sessions.
- Reaction session (October 29): opened at ₩241,000 (-0.4%), traded between ₩226,000 and ₩249,000, and closed at ₩232,000 (~$173.41), down 4.1% or ₩10,000.
- Volume: 25.0 million shares against a 30-day average of 18.0 million, about 1.4x.
- Relative: the KOSPI fell 1.2% and SK hynix rose 2.6% the same session, so Samsung underperformed the index by about 2.9 percentage points and its closest comparable by about 6.8.
- Next session (October 30): shares fell a further 3.0% to ₩225,000 as the KOSPI dropped 3.5% on renewed AI-spending worries after a US platform company's results; SK hynix fell 7.1%.
- Since our July note: from ₩207,000 to ₩232,000, up 12.1%, against the KOSPI up 6.2% and the S&P 500 up 3.8% over the same sessions.
The drop has three explanations, and the order matters. The first is positioning: the stock rose 9.0% the day before on a peer's results, so the two sessions together still left it 4.5% higher than on October 27. The second is the payout: the market had waited since August for the size of the remaining 2026 return and got a date instead. The third is relative: the chip division's operating profit came in a little below the Street's divisional number, and Samsung's 2028 supply horizon compared poorly with its peer's 2030.
None of the three says anything about the business deteriorating. A record quarter, a doubled contract book and a mid-to-high-teens bit guide were sold because the shareholder did not get paid on the day, in a market whose index had fallen 22% between October 2 and October 26 and then rebounded 10% in two sessions. That volatility is the setting for every judgment in this note.
The more useful observation is the relative move since July. The upgrade has outperformed both the S&P 500 and the KOSPI by a wide margin despite the October sell-off, which is the result the rating predicted. The question is how much of the remaining case is left at ₩232,000, and the valuation section answers it.
Street Perspective
Debate: Does a shorter shortage horizon mean Samsung sees the turn first?
Bull view: Samsung's 2028 is a conservative, sourced view built on clean-room lead times and die-size physics, and 2027 is guided tighter than 2026. A competitor talking to 2030 is selling; Samsung is underwriting. Either way the next two years are tight.
Bear view: The largest memory producer, with the most capacity and the most capital, sees the shortage ending two years before its peer does. Pricing gains have already slowed from roughly 90% a quarter to 20%, and industry capital spending is rising sharply into 2027. The supplier closest to the volume is the one to listen to.
Our take: The bears read the difference correctly and draw the wrong conclusion. A 2028 horizon from a supplier with ten signed multi-year contracts is a statement about how far contracts cover, not a forecast of collapse in 2029. We value the shares on normalised earnings precisely so that the horizon debate does not decide the rating.
Debate: Is the capital return a catalyst or a disappointment?
Bull view: ₩55-75tn of 2026 returns remain, much of it as buyback and cancellation, with a new policy in January and a stated intent to buy back discounted preferred shares. It completes the largest shareholder return in Korean corporate history, arriving into a de-rated share price.
Bear view: The policy is still half of free cash flow, half what US memory peers pledge, and the company has deferred every concrete number twice. The ₩30tn dividend was pre-announced and the record date has passed. Until January, there is nothing new to own.
Our take: The bears are right that nothing was sized, and wrong that nothing is coming. The envelope is set by the board and the date is fixed. We now treat the January meeting as the defining event for the rating and add an explicit trigger on it below.
Debate: Is this the peak, with AI spending fatigue as the trigger?
Bull view: 2027 HBM is sold out at target prices, ten customers have signed multi-year contracts and want more, fourth-quarter bits are guided up by the mid-to-high teens, and management guides another record year. The October sell-off priced a downturn the order book does not show.
Bear view: Price increases are decelerating, a Chinese competitor is closing the technology gap, the largest AI spenders are being punished by their own shareholders for capital intensity, and capital spending across the industry is rising 60% at the top. The cycle turns when the buyers' economics do, not when the sellers' contracts expire.
Our take: The bear case is about 2028 and later; the bull case is about the next four quarters. Our target already assumes the bears are eventually right about earnings, by valuing normalised rather than peak profit, so a peak-cycle argument alone is not a reason to sell at 12x normalised earnings with ₩30,000 a share of net cash on the balance sheet.
Model Update Needed
| Item | Prior estimate (July) | Revised | Reason |
|---|---|---|---|
| Q3 2026 operating profit | ₩85-95tn | ₩107.44tn actual | Price held and the incentive drag did not show, despite a ~₩10tn currency headwind |
| Q4 2026 operating profit | n/a | ₩110-125tn | DRAM bits +mid-to-high teens, NAND +mid-teens; further currency headwind if the won holds |
| FY2026 operating profit | ₩320-360tn | ₩365-380tn | 9M already ₩254.16tn |
| FY2026 EPS (company basis) | ₩38,000-44,000 | ₩41,000-44,000 | 9M EPS ₩29,434 plus Q4 of ₩12,000-14,500 |
| FY2027 operating profit | n/a | Above FY2026 | Management guides another record year |
| USD/KRW assumption | ₩1,421 | ₩1,338 | USD/KRW 5.8% lower between prints (won about 6% stronger) |
| Normalised mid-cycle EPS | ₩18,000-22,000 | ₩17,000-21,000 | Prior range scaled by the currency move (x 0.942); higher capex and buybacks treated as offsetting |
| FY2026 CapEx | ₩60-70tn, low confidence | ₩84.4tn | Company figure; DS ₩76.8tn |
| MX operating result | Loss-making through H2 | Unchanged | (₩0.6tn) in Q3 with Networks; Q4 profit guided lower |
| Foundry breakeven | Possible during 2027 | Unchanged | Q4 double-digit growth; sub-8nm over-utilised in 2027; profitability not stated |
| Remaining 2026 returns | Not modelled | ₩55-75tn | Company figure; size set in late January 2027 |
Valuation. At the ₩232,000 reaction close the shares trade at 5.3x to 5.7x our FY2026 EPS range and 5.1x annualised third-quarter EPS. On normalised earnings, the basis of every target in our coverage, the multiple is 11.0x to 13.6x our revised ₩17,000 to ₩21,000 range. On July's unchanged ₩18,000 to ₩22,000 range the multiple is 10.5x to 12.9x today, against 9.4x to 11.5x at July's ₩207,000. The re-rating since the upgrade is therefore about 12% on a like-for-like basis, matching the share-price move. Net cash of ₩197.6tn is roughly ₩30,000 per share, about 13% of the price, and our target gives it no separate credit.
Price target: ₩245,000, trimmed from ₩260,000, implying about 6% upside from the ₩232,000 reaction close and about 9% from the ₩225,000 close of October 30. The method is unchanged: 13x the midpoint of our normalised range, which is ₩18,833 after the currency scaling (₩244,800, rounded). The only input that moved materially is the currency. A dollar bought 5.8% fewer won at the third-quarter print than at the second, and since memory is priced in dollars and much of the cost base is in won, we scale July's normalised range down by that 5.8%. The capital-spending figure, above our estimate, raises future depreciation; the buyback, at least half of the ₩55-75tn still to come, lowers the share count. We treat the two as offsetting rather than claim a precision we do not have.
Why the rating holds at single-digit upside to target. Expected twelve-month return has two parts our target does not capture. The first is the ₩55-75tn of 2026 returns, about 3.6-5.0% of the price, plus whatever the January policy sets for 2027. The second is a second year of peak earnings, guided by management, that the normalised framework deliberately values at zero. Together they put expected return ahead of what we expect from the S&P 500, and above the KOSPI overlay, which is why we maintain Outperform at lower conviction rather than move to Hold.
Thesis Scorecard Post-Earnings
Scored against the standing thesis and the commitments management made on the July call.
| Thesis Point | Status | Notes |
|---|---|---|
| Bull 1: Early-cycle earnings power | Retired | Retired in July and superseded by Bull 5. No change. |
| Bull 2: Structural shortage | Confirmed | Tight into 2028, 2027 tighter than 2026, with the capital-spending argument made in detail. Horizon unchanged, not extended. Tag stays ON TRACK. |
| Bull 3: HBM4 | Exceeded | HBM4 >3x QoQ and 60% of HBM; parity with DRAM share a quarter early; 2027 sold out at target prices; HBM4E mass production H1 2027. Tag stays ON TRACK. |
| Bull 4: Legacy node scarcity | Confirmed | Prices rose across all applications; mobile memory shortage persists; supply response focused on HBM, server DRAM and SSD. Tag stays ON TRACK. |
| Bull 5: Contracted demand and price floors | Strengthened | Ten accounts signed (five in July), ten more in talks, extension demand above expectation, scope widening beyond data centre. Floors and prepayments not updated. Tag stays ON TRACK. |
| Bear 2: Non-memory drag | Materialising, capped | DX (₩0.8tn) loss; MX loss narrowed but Q4 guided lower; costs high into 2027. DS is 99.1% of group operating profit. Tag stays MATERIALIZING. |
| Bear 3: Cyclical peak and CapEx-driven depreciation | Escalating | FY2026 CapEx ₩84.4tn, +60% YoY and above our estimate; foundry evaluating more fabs; price gains slowing. Offset by doubled contract coverage. Tag stays EMERGING, closer to MATERIALIZING. |
| Bear 4: Labour disruption and unrecognised provision | Contained | No disruption reported. Tag stays CONTAINED. |
| Bear 5: Deferred cost distorts Q3 | Did not show | Gross margin slipped only 30bps and operating margin rose 280bps. Effect unquantified. Tag EMERGING to CONTAINED. |
| Bear 6 (new): Won strength | Materialising | About ₩10tn operating headwind QoQ; dollar down 12.3% between quarter-ends; natural hedges only; ₩9.5tn finance expense. New at MATERIALIZING. |
Management commitments from the July call, graded
| Commitment | Outcome |
|---|---|
| Disclose where the price floors sit | Not delivered. Not addressed on the call or in the deck. |
| Q3 gross margin to dip on the deferred incentive cost | Smaller than flagged. Gross margin -30bps; operating margin +280bps. |
| HBM4 sales >3x QoQ in Q3; well over 60% of HBM mix in H2 | On track. More than threefold; 60% in Q3, with Q4 still to come. |
| HBM share at parity with DRAM share in H2 | Delivered early, in Q3. |
| Q3 bits: DRAM +mid-single digit, NAND +high single digit | Missed. DRAM +low to mid-single digit; NAND +mid-single digit. |
| Server SSD >60% of NAND; QLC bits to double H2 vs H1; V10 mass production in August | Partly. Large additional QLC demand cited; mix not updated; V10 described as "production ready". |
| Foundry turnaround "possible in the near term"; 2nm orders more than double; Taylor Fab 2 construction by end-2026 | Partly. Earnings improved significantly and 2nm doubling reaffirmed; profitability not stated; Fab 2 not updated. |
| Successor shareholder-return policy | Partly. 2026 envelope set in August; remaining ₩55-75tn and new policy dated to late January 2027. |
| FY2026 CapEx to be quantified | Delivered. ₩84.4tn, above our estimate. |
| MX cost burden to continue through H2 | As guided. Loss narrowed to ₩0.6tn with Networks; Q4 guided lower. |
| No ADR issuance under review | Not raised on this call. |
The triggers we published in July, graded
| Trigger (as published) | Status |
|---|---|
| Move back to Hold "if the disclosed floors prove to sit far below mid-cycle economics" | Not fired. Floors still undisclosed, so the test cannot yet be run. |
| Move back to Hold "if 2027 pre-bookings show evidence of double-ordering" | Not fired. No evidence; requests for extra volume in contracted years are now on our watch list. |
| Move to Underperform "on contract renegotiation by any of the signed counterparties" | Not fired. Counterparties are seeking extensions, the opposite direction. |
| New: move to Hold if the late-January 2027 board sets the remaining 2026 return at the bottom of ₩55-75tn without a substantial cancellation programme, or a successor policy below 50% of free cash flow | Added today. Our base case is a figure inside the range with a substantial cancellation. |
Overall: Thesis strengthened. Of eleven July commitments, three were delivered or beaten, including HBM share parity and the capital-spending figure we had asked for across four notes; one, the HBM4 mix, is on track with a quarter to run; three were partly met, one played out as guided, two were missed and one was not raised. The two misses were the price-floor disclosure and the bit guide, and a new currency risk replaces the accounting distortion we had flagged. The contract pillar behind the upgrade doubled in signed counterparties.
Action: Maintain Outperform, target ₩245,000 from ₩260,000, conviction lowered to 6 from 7. The business delivered more than we asked of it, and the target is lower only because the won is stronger. Upside to target is now single-digit, so the rating rests on the ₩55-75tn of returns still to come and on a second year of peak earnings that our framework does not count. The late-January board meeting is the next decisive event, and we have published a trigger on it. The July triggers stand unchanged.