| Jul 29, 2026 | -0.07% | Q2 FY2026 | Read transcript briefingQuarter in one view- Q2 FY2026: another record — revenue KRW 171.5T (+28% QoQ), OP KRW 89.5T (+56% QoQ), OPM 52% (from 43%), net profit KRW 71.6T (+52%), EPS KRW 10,849 both classes. FX added ~KRW 3.1T to OP QoQ (vs 1.8T in Q1), again centered on components. R&D hit a quarterly record KRW 16T (from 11T).
- Memory again did the work: record DRAM and NAND bit shipments; DRAM bits +low-teens% QoQ (beat vs single-digit guide), NAND +low-single (in line); DRAM ASP +mid-40s% QoQ, NAND +high-60s% QoQ — a sharp deceleration from Q1's +90s/+80s% but still steep.
- The quarter's biggest new disclosures: multiyear supply agreements finalized with the top 5 global data-center customers (5 more in final talks), 5-year rolling structure, ~60–70% of planned capacity to be contracted, substantial advance payments required (one-quarter already received); incentive accrual finally booked at ~10.5% of cumulative H1 OP; MX+Networks swung to a KRW 0.7T operating loss.
- HBM4: qualifications "winding down smoothly," Q3 HBM4 sales guided >3x QoQ, HBM4 to be "well over 60%" of HBM revenue in 2H (vs >50% from Q3 guided last quarter); 2027 HBM supply agreements already signed; HBM share targeted at parity with overall DRAM share.
- Foundry inflecting: advanced nodes (8nm and below) at full utilization, earnings "significantly improved before incentive-related charges," 2nm wins from major CSPs, Broadcom named in talks, 2nm project wins guided to >2x YoY in 2026, AI/HPC revenue mix from high-teens% (2025) to >30% (2026), advanced nodes >50% of revenue; turnaround "possible in the near term."
- CapEx KRW 16.8T (+5.5T QoQ; DS 15.4T, Display 0.7T) — the promised 2H-weighted ramp began. Q2 DPS KRW 374 (regular); special dividend now explicitly "under discussion" for this year.
- Not disclosed: segment OP (eighth consecutive quarter), absolute HBM revenue, foundry loss magnitude, MX units/ASP, strike impact (not mentioned at all on this call).
What management is focused on- Contractualizing the shortage: the call's centerpiece. Memory EVP Kim laid out the most detailed supply framework in this series — shortage persists through 2028, "more severe in 2027 than 2026," no meaningful incremental industry supply through 2028 given >3-year fab lead times; AI frontier-model developers now approaching Samsung directly (bypassing hyperscalers/neo-clouds) for multiyear deals.
- Business-model transformation narrative: multiyear LTAs framed as shifting memory "from one overexposed to supply-demand cycles to a more stable and predictable model" — minimum price floors on mainstream products, differentiated pricing by segment, advance payments as binding commitments.
- HBM share parity: explicit target of HBM market share "broadly in line with overall DRAM market share" in 2H26, balanced HBM/conventional mix maintained despite the margin inversion.
- Foundry pivot to AI/HPC: strategic shift toward high-growth segments with quantified mix targets (>50% advanced-node revenue, >30% AI/HPC in 2026) and capacity responses — Taylor Fab 2 construction to begin by end-2026, MP 2030; 1.4nm inquiries rising, additional fab capacity under review.
- DX damage control: MX loss acknowledged plainly; response is flagship-first mix, upselling (A57/A37), S26 FE, Z8 foldables, Intelligent eyewear "later this year," and end-to-end efficiency. VD pivoting to AI TV + TV Plus advertising/services under new leadership.
- New growth engines fleshed out: robotics consolidated into an "RX business office" reporting directly to the CEO (new hires named: EVP Dongkun Lee ex-Hyundai/Boston Dynamics, two professors), Gumi pilot line and data factory; HARMAN's ZF ADAS and Sound United acquisitions explained; agentic AI / "AI OS" as the MX software thesis.
Key numbers and quarter mechanics- Consolidated: revenue 171.5T (+28% QoQ); OP 89.5T (+56%); OPM 52% (+9pp); net profit 71.6T (+52%); EPS 10,849 (+52%); R&D 16T (record, from 11T); FX +~3.1T QoQ to OP.
- Memory Q2: DRAM bits +low-teens% QoQ (beat vs single-digit guide); NAND bits +low-single (in line); DRAM ASP +mid-40s% QoQ; NAND ASP +high-60s% QoQ; record bit shipments both. Q3 guides: DRAM bits +mid-single, NAND +high-single (inventories "significantly low").
- Incentive accrual: Q2 booked a cumulative H1 accrual equal to ~10.5% of cumulative H1 OP (implies roughly KRW 15.4T against ~146.5T H1 OP — arithmetic inference, not stated). Part capitalized into WIP inventory, so the Q2 P&L expense is lower than the accrual; the remainder hits COGS as inventory sells from Q3. This mechanically flattered Q2 margins and will weigh on coming quarters.
- MX: revenue KRW 32.3T (down from 37.5T in Q1); MX+Networks combined OP −0.7T (from +2.8T) — a swing to loss on memory costs despite QoQ volume growth and YoY revenue growth. FY26: shipments expected to decline, but full-year volume and ASP both guided up (premium mix).
- Foundry: revenue up QoQ on HBM base-die and U.S. customer demand; earnings "significant improvement before incentive-related charges" — i.e., the accrual hit foundry hard; advanced nodes at full utilization; 2H: 2nm gen-2 mobile MP, 4nm LPU ramp for a major customer, base-die scaling; double-digit-or-better YoY revenue growth expected.
- NAND mix: server SSD to exceed 60% of NAND sales in 2026 (+20pp YoY); QLC bit shipments to more than double 2H vs 1H; V10 (bonding, 3-stack) MP in August.
- Display: mobile improved QoQ; large up on gaming monitors; 8.6G IT OLED mass production started July; near-term fixed-cost drag flagged.
- VD: sales and profit up YoY (sporting-event pull-in, new categories), profit marginally down QoQ on memory costs.
- CapEx: Q2 16.8T (DS 15.4T, Display 0.7T); memory up on Pyeongtaek new fab + infrastructure; foundry up on Taylor ramp; H1 total 28T against a "substantial increase" over FY25's 52.7T — still heavily 2H-loaded.
- Capital return: Q2 DPS 374 both classes, payable August; special dividend for 2026 explicitly under board discussion; CFO flagged that LTA customer prepayments and employee-comp buybacks "may affect FCF" — a notable caveat on the 50%-of-FCF policy metric.
Product and launch scorecard- HBM4: qualifications winding down "smoothly"; Q3 sales guided >3x QoQ; 2H mix "well over 60%" of HBM revenue (raised from ">50% from Q3"); 1Z-nm capacity expansion and yields "as planned"; 2027 supply agreements already signed. Strongest delivery evidence yet, still no named customer.
- HBM4E: industry-first samples shipped to major customers in Q2 — delivered on last quarter's promise; cited as the basis for 2027 commercialization confidence.
- Multiyear LTAs: top-5 global data-center customers signed; 5 more in final talks; 5-year rolling terms with annual add-on negotiation; advance payments required, one-quarter received; price floors on mainstream products; 60–70% of planned capacity expected under contract. The quarter's most consequential commercial development.
- NAND/SSD: Gen 6 SSD positive customer feedback reiterated; 2Tb QLC done (March), 256TB SSD lineup; V10 MP August; server SSD >60% of NAND mix; KV-cache storage demand called out as a new AI-driven segment.
- Foundry: 2nm gen-2 mobile MP in 2H (on track); 2nm CSP/AI-HPC design wins entered design phase; Broadcom named in talks; 2nm wins to >2x YoY; Taylor Fab 1 operations by end-2026, Fab 2 construction start end-2026, MP 2030; mature-node shift to 8nm/17nm/CIS/eNVM plus silicon photonics.
- System LSI: record H1 revenue; next-gen flagship SoC secured (Exynos 2700 by implication, "development and production progressing smoothly"); custom SoC/total-solution push leveraging design+foundry+memory+packaging.
- MX: Z8 foldable series launched; S26 FE, Tab S12, Watch Ultra 2 coming 2H; Intelligent eyewear "later this year" — first timing commitment for AI glasses.
- Display: 8.6G IT OLED MP started July — on schedule; full-oxide technology for cost competitiveness.
- VD: Micro RGB/new categories credited for YoY profit gain; AI TV + TV Plus content/IP expansion as the new leadership's strategy.
Sell-side read-through- KB (Dongwon Kim) drew the special-dividend tease and the FCF caveat (LTA prepayments and employee-comp buybacks may affect FCF) — a subtle signal the 50%-of-FCF payout base could be inflated by customer prepayments, and management chose to flag it preemptively. Also got the 2028 shortage-duration call.
- HSBC (Ricky Seo) got the HBM share-parity target and the >3x QoQ Q3 HBM4 guide plus "well over 60%" 2H mix — the quarter's key HBM datapoints.
- Citi (Sei Cheol Lee) forced the incentive-provision quantification (~10.5% of H1 OP, partially capitalized into WIP — expense recognition deferred into Q3+) and a flat ADR denial ("not currently reviewing"), closing a media rumor.
- SK (Dong Hee Han) got the Broadcom name-drop and MX's service-monetization answer, which was deliberately vague ("phased approach," no short-term profit chase).
- JPMorgan (Jay Kwon) got the NAND AI thesis quantified (server SSD >60% of NAND mix, +20pp YoY; QLC bits doubling; V10 August) and the HARMAN M&A rationale (ZF ADAS, Sound United).
- Korea Investment (Min-sook Chae) extracted the LTA term sheet: top-5 signed, 5-year rolling, advance payments (one-quarter received), price floors, 60–70% capacity allocation — the most substantive answer of the call.
- Daiwa (S.K. Kim) got Taylor Fab 2's construction start (end-2026) and MP (2030) — new disclosure — plus the VD leadership-change strategy.
- BofA (Simon Woo) got foundry's full-utilization confirmation and a near-term turnaround hint, but the buyback-vs-employee-comp question was answered procedurally with no commitment.
- CGS (Kyunga Lee) got the 8.6G OLED margin caveat (initial fixed-cost drag) and the custom-SoC total-solution strategy.
- NH (Young Ho Ryu) got the AI OS/Knox differentiation pitch and the robotics org detail (RX office, named hires, Gumi pilot line).
- Missing challenges: no one asked about the May–June general strike's production or cost impact — total silence one quarter after it was the dominant risk; no segment OP request (eighth quarter); no HBM absolute revenue; no foundry loss quantification; no Tesla question (fourth quarter); no one pressed the effective margin impact of the capitalized incentive deferral; no one asked LTA pricing vs spot.
Management credibility- Delivered: DRAM bit beat (low-teens vs single-digit guide); NAND in line; HBM4E samples shipped in Q2 as promised; 8.6G OLED MP in July as flagged; CapEx stepped up +5.5T QoQ as the "back-loaded" explanation predicted; incentive accrual booked as promised ("will hit financials when finalized") with transparent quantum and accounting mechanics.
- Candor improvements: MX's swing to a 0.7T loss disclosed plainly with combined MX+Networks OP still given; foundry's "before incentive-related charges" qualifier is honest but also reveals the accrual materially dented foundry; the FCF caveat on prepayments is unusually forthcoming; ADR rumor addressed directly rather than dodged.
- Open items: strike impact never mentioned — a material omission given Q1's prominence; segment OP absent an eighth quarter; HBM still no named customer despite signed 2027 agreements; LTA advance-payment size withheld (NDA); foundry turnaround timing still unquantified ("near term"); MX service monetization remains conceptual.
- Consistency watch: HBM4 2H mix raised from ">50% from Q3" to "well over 60%" — upward revision consistent with ramp confidence; the "balanced mix" stance from Q1 held and now quantified as share parity; the Q1 claim of "signed multiyear agreements with some customers" escalated to top-5 signed with structural detail — verifiable progression; the 2027 margin-normalization thesis was not repeated this call (not contradicted, but not updated).
- New credibility risk: the incentive accounting (capitalized into WIP, expensed on sell-through) means Q2's 52% OPM overstates run-rate profitability — management explained it, but only under direct questioning; the FCF/prepayment caveat introduces policy-metric ambiguity just as the next return policy is being written.
What changed versus the prior quarter- Scale stepped up again: OP 57T → 89.5T, OPM 43% → 52%, net profit 47T → 71.6T — but ASP momentum decelerated hard (DRAM +low-90s% → +mid-40s%; NAND +high-80s% → +high-60s%) while bit growth accelerated (DRAM low-single → low-teens). The mix of price vs volume is shifting.
- The incentive-provision overhang resolved into a booked number (~10.5% of H1 OP) with deferred P&L recognition — known cost now, but its full income-statement weight lands from Q3.
- The strike vanished from disclosure: Q1's dominant risk (May 21–June 7 general strike) received zero mention — no impact statement, no resolution comment.
- LTAs went from "signed with some customers" to a full architecture: top-5 signed, 5-year rolling, advance payments (25% received), price floors, 60–70% of capacity — the shortage is now a contracted, prepaid business model.
- HBM4 mix guidance raised (>50% from Q3 → well over 60% in 2H) and 2027 HBM supply agreements already signed — the ramp question is becoming a share-gain question with a parity target.
- MX deteriorated from "single-digit profitability" to a 0.7T loss — the memory-cost transfer to DX is now visible in P&L, not just guidance.
- Foundry moved from "sequential improvement expected" to full advanced-node utilization, named customer talks (Broadcom), quantified mix targets, and a near-term turnaround hint; Taylor Fab 2 got dates for the first time.
- Capital return: special dividend for 2026 now explicitly on the table (Q1 said only "policy faithfully executed"); FCF-definition caveat introduced.
- Robotics graduated from narrative to organization: CEO-direct RX office, named external hires, Gumi pilot line — versus Q1's Jun Ho Oh mention (note: leadership name changed to Dongkun Lee; the call did not reconcile).
- New: ADR listing denied; HARMAN's ZF ADAS and Sound United acquisitions disclosed; Intelligent eyewear timing committed ("later this year").
Bull case- Two consecutive records with accelerating magnitude: 89.5T OP, 52% OPM, and Q3 bit guides (DRAM mid-single, NAND high-single) imply volume now adding to still-rising prices; management sees the shortage worsening in 2027 and persisting through 2028.
- The LTA architecture is a structural re-rating argument: 60–70% of capacity under 5-year rolling contracts with price floors and prepaid advances (25% received) converts cyclical earnings into contracted backlog — management's explicit intent, now with signed top-5 customers.
- HBM execution is compounding: >3x QoQ HBM4 sales guided for Q3, >60% 2H mix, 2027 agreements signed, HBM4E first-to-sample, share-parity target — the qualification overhang is fully behind.
- NAND is a confirmed second AI leg: server SSD >60% of NAND mix (+20pp YoY), QLC bits doubling, V10 MP August, KV-cache demand emerging.
- Foundry inflection is now quantified: full advanced-node utilization, 2nm wins >2x YoY, AI/HPC mix to >30%, advanced nodes >50% of revenue, near-term turnaround possible, Taylor Fab 2 committed.
- Capital return upside: special dividend explicitly under discussion for 2026 from a 71.6T-profit quarter; LTA prepayments could swell FCF.
- DX optionality: Intelligent eyewear this year, AI OS differentiation, robotics organization with credible hires, HARMAN ADAS entry.
Bear case- Earnings quality: ~10.5%-of-H1-OP incentive accrual partially capitalized into WIP means Q2's 52% OPM is flattered; the deferred expense hits COGS from Q3 — a mechanical margin headwind already scheduled. FX added another 3.1T.
- ASP momentum is decelerating sharply: DRAM +mid-40s% vs +low-90s%, NAND +high-60s% vs +high-80s% — the price-spike phase is maturing; if prices plateau, bit growth alone (mid/high-single) cannot sustain 56% QoQ OP growth.
- MX is now loss-making (−0.7T combined) with memory costs guided to persist through 2H — the internal transfer is destroying DX profitability in real time, and smartphone market volume is declining.
- The strike's impact is undisclosed — either immaterial (then say so) or being absorbed silently; the omission itself is a disclosure-quality concern.
- LTA lock-in cuts both ways: 60–70% of capacity committed at floor-protected but presumably below-peak pricing could cap upside if spot keeps rising; prepayments flatter FCF and the payout base without being earnings.
- CapEx is ramping into the teeth of the cycle: 16.8T in Q2, Taylor Fab 2 construction starting, >110T multi-year plan, 1.4nm capacity under review — FCF quality depends on prepayment durability.
- Disclosure remains poor: no segment OP (eight quarters), no HBM revenue absolute, no foundry loss figure, LTA economics under NDA.
- Consumer demand destruction is broadening: mobile/PC moderation acknowledged, Display citing memory-driven set-price pressure, System LSI flagging consumer softening, VD facing post-event slowdown.
Next-quarter watchlist- Incentive expense flow-through: how much of the capitalized accrual hits Q3 COGS; whether a second-half accrual adds on top; restated-like margin trajectory vs the flattered Q2.
- Strike resolution: any retrospective disclosure of production/cost impact; status of labor negotiations post-settlement; whether incentive terms are finalized.
- Memory pricing vs contracts: Q3 ASP trajectory after the deceleration; LTA pricing vs spot commentary; advance-payment balance growth; whether the 5 additional accounts close; any disclosure of contracted-volume pricing mechanics.
- HBM4: Q3 >3x QoQ sales delivery; 2H mix tracking to "well over 60%"; any named customer; HBM vs conventional DRAM margin update vs the 2027 normalization thesis; HBM4E commercialization progress.
- Foundry: 2nm gen-2 MP execution; conversion of Broadcom/CSP talks into awards; evidence toward the "near-term" turnaround; Taylor Fab 1 operations start and Fab 2 construction start by year-end; 1.4nm capacity decision.
- CapEx: 2H run-rate vs the "substantial increase" guide (H1 was only 28T); any FY number; memory/foundry split.
- Capital return: special dividend decision and size; next-policy (post-2026) framework; how LTA prepayments are treated in the FCF payout base; buyback vs employee-comp sequencing under the revised Commercial Act.
- MX: whether the loss narrows in 2H on foldables/S26 FE/eyewear launches; memory-cost pass-through evidence; service monetization specifics; Intelligent eyewear launch execution.
- Robotics: tangible milestones from the RX office and Gumi pilot line; reconciliation of the leadership naming (Jun Ho Oh vs Dongkun Lee).
- Earnings quality: segment OP restoration; MX unit/ASP disclosure; quantification of one-offs in the 71.6T net profit.
|
| Apr 29, 2026 | +0.81% | Q1 FY2026 | Read transcript briefingQuarter in one view- Q1 FY2026: record revenue KRW 134T (+43% QoQ vs the prior record of 93.8T); record OP KRW 57T (+185% QoQ); OPM 43% (vs 21.4%); net profit KRW 47T (2.4x QoQ); EPS KRW 7,123 common / 7,124 preferred. FX added ~KRW 1.8T to OP QoQ, again centered on components.
- Memory is the entire story: second straight quarterly earnings record; DRAM blended ASP +low-90s% QoQ, NAND +high-80s% QoQ; server bits +low-teens% (DRAM) and +low-20s% (NAND) QoQ with record server revenue. DRAM bit guide met; NAND beat (high-single vs mid-single guide).
- HBM4 shipped commercially in February (world-first claim), capacity fully booked/sold out, >3x 2026 HBM sales guide reiterated, HBM4 to exceed 50% of HBM sales from Q3 and ~half for the full year; HBM4E (16Gbps, 4.0TB/s) samples within Q2.
- DX: MX revenue KRW 37.5T, MX+Networks combined OP KRW 2.8T ("single-digit profitability" despite memory-cost surge); S26 driving YoY ASP/revenue growth. VD profit improved QoQ but down YoY.
- CapEx KRW 11.2T (-9.2T QoQ; DS 10.2T, Display 0.6T) — front-loaded 2025 cleanroom investment explains the dip; full-year 2026 guided to a "substantial" YoY increase. Corporate value enhancement plan: >KRW 110T facilities + R&D commitment.
- Capital return: Q1 DPS KRW 372 (regular only — no special); remaining treasury shares from the 10T buyback fully canceled in April (73.4M common / 13.6M preferred, ~1.2%/1.7% of shares, ~KRW 14.6T value), retaining only KRW 1.6T for employee compensation.
- New risks surfaced: general strike announced May 21–June 7 (first strike disclosure in seven quarters); incentive provision under negotiation, not yet booked; Middle East oil/freight cost risk flagged.
What management is focused on- Record framing: CFO opened with "highest ever quarterly revenue and operating profit," attributing it to AI innovation and high-value portfolio execution — the victory lap continues but is now backed by a 43% OPM.
- Shortage as strategy: memory demand fulfillment rate "at a record low"; customers pre-booking 2027 demand; supply-demand gap "looking to widen further in 2027." Multiyear supply agreements confirmed as signed with some customers — a hardening from last quarter's "requests handled selectively."
- HBM4 ramp and premium pricing: management claims HBM4's differentiated performance is "translating into actual premium on pricing," with demand concentration and sold-out capacity; 2H26 volume scaling is the explicit plan.
- HBM vs conventional DRAM margin inversion addressed head-on: conventional DRAM currently higher-margin due to quarterly repricing vs HBM's annual pricing, but the gap is expected to "significantly reduce in 2027"; management committed to a balanced mix rather than chasing short-term DRAM margin.
- Cost defense everywhere downstream: MX, Display, VD, System LSI all framed around surviving memory-price inflation; MX explicitly guided to a YoY profitability decline in 2026.
- New growth narratives: humanoid robotics (Jun Ho Oh named as leader; manufacturing robots first, then home/retail; 2-track with global partners, M&A possible); data-center cooling via FläktGroup (market cited $4.7B 2024 → $16.6B 2030, ~24% CAGR; North America and Korea expansion); silicon photonics/CPO in foundry (strategic project win from a leading optical module player, MP starting 2H26).
- Labor risk management: strike response framed as dedicated teams within legal framework, prioritizing dialogue — first substantive labor comment in this transcript series.
Key numbers and quarter mechanics- Consolidated: revenue KRW 134T (+43% QoQ); OP KRW 57T (+185% QoQ); OPM 43% (+~21.6pp); net profit KRW 47T (2.4x QoQ); EPS 7,123/7,124. FX +~KRW 1.8T QoQ to OP.
- Memory Q1: DRAM bits met guide (low-single); NAND bits +high-single QoQ (beat vs mid-single guide); DRAM blended ASP +low-90s% QoQ; NAND ASP +high-80s% QoQ; server bits +low-teens% DRAM / +low-20s% NAND QoQ; record server revenue; record-low demand fulfillment rate.
- Q2 memory guides: DRAM bits +single-digit QoQ; NAND bits +low-single QoQ (inventory-constrained); memory prices expected to stay on the upward trend.
- MX Q1: revenue KRW 37.5T; MX+Networks combined OP KRW 2.8T; single-digit profitability; S26 sales expanding YoY in value terms. Q2: revenue down QoQ, profitability decline "inevitable" on component costs. No smartphone unit/ASP disclosure this quarter (vs 60M/$244 last quarter).
- Foundry Q1: earnings declined QoQ (seasonality) but double-digit YoY revenue growth; Q2 advanced nodes expected at full utilization with sequential earnings improvement, HBM4 base-die demand cited; 2nm gen-2 MP in 2H26; 1.4nm "progressing as planned."
- Display: mobile down QoQ (seasonality + memory price pressure on set demand); large display stable on gaming monitors; 8.6G IT OLED ramp as 2H revenue driver.
- VD: profit improved QoQ, declined YoY (stagnant demand, raw material costs); Q2 World Cup demand expected (more teams/matches than prior years).
- CapEx: Q1 KRW 11.2T (DS 10.2T, Display 0.6T); memory down QoQ on front-loaded Pyeongtaek cleanroom spend, equipment spend to rise through the year; foundry down on Taylor base effect, ramp investment rising from Q2; FY2026 "substantial" YoY increase, no number.
- Capital return: Q1 DPS KRW 372 both classes, payable May; cancellation completed in April — 73.4M common (1.2%) + 13.6M preferred (1.7%), ~KRW 14.6T at board-date close; KRW 1.6T retained for employee comp.
- Not disclosed: segment OP (seventh consecutive quarter), absolute HBM revenue, foundry loss magnitude, incentive provision size, MX unit shipments/ASP.
Product and launch scorecard- HBM4: delivered on the February ship commitment — world-first commercial shipment claimed, ramp on schedule, sold out, premium pricing confirmed, >3x 2026 HBM sales reiterated, >50% of HBM mix from Q3. Still no named customer, but "fully booked and sold out" plus premium pricing is the strongest evidence yet.
- HBM4E: specs disclosed for the first time — 16Gbps pin speed, 4.0TB/s bandwidth; samples shipping within Q2 (ahead of the prior "mid-2026" framing).
- SOCAMM2: industry-first mass-produced shipments started in Q1 — new delivery, previously only portfolio positioning.
- PCIe Gen 6 SSD: on-time development in Q1, now in customer qualification with "positive feedback"; early Gen 6 market lead targeted for 2H26; NVIDIA's CMX architecture (inference storage extending to NAND) cited as a demand driver — a notable named-customer ecosystem reference.
- QLC/TLC NAND: 2Tb QLC development completed in March; 256TB server SSD lineup expansion; V9 migration accelerating within QLC.
- S26: launched; YoY value growth claimed, ultra-model mix driving ASP; "second-generation custom AP" question from last quarter resolved indirectly — System LSI confirmed Exynos 2600 as the flagship predecessor and Exynos 2700 in smooth development with share-gain ambitions. Exynos is back in the flagship, confirmed by implication.
- Foundry: 2nm gen-2 MP 2H26 on track; active 2nm talks with multiple large AI/HPC customers; 4nm HBM4 base-die demand leading, supply expansion under consideration; silicon photonics strategic win with MP 2H26; Taylor Fab 1 equipment move-in ceremony April 23, operations 2026, MP 2027; Fab 2 in early review. Mature nodes: 8-inch PMIC/DDI/CIS phased closeouts, migration to 17nm specialty.
- Micro RGB TV: launching as the premium differentiator against the TCL–Sony JV; mini LED added for the volume zone.
- AI glasses and new form factors: referenced by MX as upcoming multimodal AI devices — no timing.
Sell-side read-through- Citi (Sei Cheol Lee) got two firsts: confirmation that multiyear memory supply contracts are signed with some customers (binding, higher-commitment structures), and the incentive-provision answer — under negotiation, not booked in Q1, will hit financials when finalized. An unbooked cost of unknown size now hangs over coming quarters.
- BofA (Simon Woo) forced the first strike disclosure in seven quarters: union rally April 23 at Pyeongtaek, general strike announced May 21–June 7; management committed only to "minimize potential production disruptions" — no quantification of exposure.
- Meritz (Sunwoo Kim) drew the capital-return posture: current policy will be "faithfully executed," next policy under board discussion, and the >KRW 110T value-enhancement investment plan detailed (Pyeongtaek, Taylor, Yongin cluster, M&A, venture).
- UBS (Nicolas Gaudois) probed Middle East supply risk — management says lines normal, safety stock secured, suppliers diversified (U.S./Japan), but flagged freight-cost inflation as the real transmission channel. MX's answer on margin defense was partnerships + efficiency, still unquantified.
- KB (Dongwon Kim) extracted the Q2 bit guides and the robotics strategy (Jun Ho Oh, manufacturing-first, 2-track partner approach).
- Daiwa (S.K. Kim) got the HBM4 mix disclosure (>50% of HBM sales from Q3, ~half full-year) and HBM4E specs — the quarter's most concrete HBM datapoints.
- Goldman (Giuni Lee) got foundry order color (2nm AI/HPC talks, 4nm base-die expansion consideration, silicon photonics MP 2H26) and a VD answer that was largely boilerplate.
- Korea Investment (Dana Chae) asked the HBM-vs-DRAM margin question directly and got the most candid answer of the call: margin inversion acknowledged, annual vs quarterly pricing mechanics explained, gap expected to narrow significantly in 2027, balanced mix defended as protecting AI infrastructure demand. Also got Exynos 2700 confirmation.
- NH (Young Ho Ryu) got MX's explicit admission of a 2026 profitability decline and the DA restructuring tease ("selective and focused," details later) plus FläktGroup cooling strategy.
- JPMorgan (Jay Kwon) pivoted to NAND-for-AI and got the NVIDIA CMX architecture thesis, Gen 6 readiness, and 2Tb QLC/256TB SSD detail.
- Samsung Securities (Jong Wook Lee) got the TCL–Sony JV response (Micro RGB + mini LED counter) and Taylor Fab 1/Fab 2 status plus mature-node closeout strategy.
- Missing challenges: no one pressed the size of the unbooked incentive provision; no one asked for segment OP (seventh quarter); no one quantified strike production exposure; no Tesla execution question (third quarter of silence); no one asked why Q1 CapEx fell 45% QoQ against a "substantial increase" full-year guide; no HBM absolute revenue request.
Management credibility- Delivered: HBM4 February shipments (committed last quarter, confirmed world-first commercial shipment); HBM4E samples pulled into Q2 (vs mid-2026 prior framing); NAND bit beat (high-single vs mid-single guide); DRAM bit guide met; treasury cancellation completed in Q1 as promised — and larger than the 6.6T flagged (14.6T value, full cancellation of remaining shares); S26 launched with Exynos 2600 implied as the custom AP.
- Candor improvements: the HBM/conventional DRAM margin inversion explained with pricing-mechanics detail (annual vs quarterly contracts) rather than glossed; MX's 2026 profitability decline stated plainly; strike and incentive provision acknowledged on first direct questioning; memory-cost demand destruction for mobile/PC reiterated by the memory team itself.
- Open items: incentive provision unquantified and unbooked — a known cost of unknown size; strike impact unquantified days before the announced May 21 start; segment OP absent for a seventh quarter; HBM still has no named customer despite sold-out claims; Tesla execution unaddressed for a third quarter; DA restructuring announced via media before the call, with details deferred.
- Consistency watch: last quarter's "multiyear contract requests handled selectively" hardened into signed contracts — verifiable escalation; the Q4 "server DDR may be prioritized over HBM near-term" evolved into a full margin-inversion explanation with a 2027 normalization thesis — management is now defending HBM mix rather than apologizing for it; the 6.6T cancellation figure from last quarter reconciled into a full 14.6T cancellation (the 6.6T appears to have been purchase-cost basis vs 14.6T market value — the call did not explicitly reconcile, but the full-cancellation outcome exceeds prior signaling).
- New credibility risk: Q1 CapEx of 11.2T annualizes well below a "substantial increase" over 52.7T — management's explanation (front-loaded cleanroom, equipment spend back-loaded) is plausible but makes FY2026 CapEx heavily 2H-weighted; watch execution.
What changed versus the prior quarter- The upcycle went vertical: OP 20.1T → 57T, OPM 21.4% → 43%; DRAM ASP acceleration from +~40% to +low-90s% QoQ; NAND from +mid-20s% to +high-80s%. Pricing is now doing almost all the work — bit growth remains supply-capped at low-single digits.
- HBM4 moved from "shipments start February" to shipped, sold out, premium-priced, and >50% of HBM mix by Q3 — the qualification overhang is now a ramp-execution question.
- The HBM-vs-DRAM margin tension resolved into an explicit inversion admission with a 2027 normalization forecast — last quarter's open question now has a management answer and a timeline.
- Multiyear contracts moved from "requests" to signed agreements — the shortage is being contractualized, raising both visibility and future capacity-commitment risk.
- Labor risk materialized: after six quarters of silence, a general strike (May 21–June 7) and an unbooked incentive provision are now live issues.
- CapEx whipsawed: 20.4T → 11.2T QoQ, explained as front-loading, with FY2026 guided to a substantial increase — the investment cycle is now back-loaded and larger (>110T multi-year plan disclosed).
- Capital return normalized: Q1 dividend back to regular 372 (no special), but the cancellation delivered at 14.6T — well above the 6.6T signaled — and the next policy (post-2026) is formally under board review.
- Exynos question resolved: 2600 confirmed as the S26 AP by implication, 2700 in development with share-gain intent — the two-quarter dodge ended.
- New structural moves: DA profitability-driven restructuring acknowledged; FläktGroup cooling expansion (North America, Korea subsidiary/factory); silicon photonics/CPO as a foundry growth vector; mature-node 8-inch closeouts announced.
Bull case- Pricing power is unprecedented in this series: DRAM ASP +low-90s% and NAND +high-80s% QoQ in a single quarter, with Q2 prices guided still upward, record-low fulfillment rates, and customers pre-booking 2027 — the shortage has duration and contractual underpinning via signed multiyear deals.
- HBM4 is now a ramp story, not a qualification story: world-first shipment, sold-out capacity, premium pricing, >3x 2026 sales, >50% mix by Q3, 4E sampling ahead of schedule with 16Gbps/4.0TB/s specs.
- The margin-inversion bear point has a management-dated expiry: HBM vs conventional DRAM gap expected to narrow significantly in 2027 as HBM reprices annually into a widening supply gap.
- NAND is becoming a second AI leg: NVIDIA CMX architecture extends inference storage to NAND; Gen 6 SSD qualification underway with positive feedback; 2Tb QLC and 256TB SSDs position for capacity demand.
- Foundry is inflecting: double-digit YoY growth, full advanced-node utilization guided for Q2, 2nm gen-2 MP in 2H26, HBM4 base-die demand pulling 4nm, silicon photonics MP 2H26, Taylor on schedule.
- Shareholder returns delivered above signaling: 14.6T cancellation completed, next policy under review from a position of 47T quarterly net profit.
- Earnings breadth beyond memory: System LSI improved on flagship SoCs (Exynos 2600 in S26), MX held single-digit margins through the worst memory-cost surge, VD profit improved QoQ.
Bear case- This is a price spike, not a volume story: DRAM bits +low-single, NAND +high-single, both supply-capped; OPM of 43% rests on ASPs that nearly doubled in one quarter — any price plateau flattens earnings immediately. FX added another 1.8T.
- Known unbooked costs: the incentive provision (size unknown, under negotiation) will hit a future quarter; the May 21–June 7 general strike carries unquantified production risk at the very fabs driving the beat.
- Memory is taxing the rest of the group: MX guided to a 2026 profitability decline, Display cited memory-driven set-demand weakness, System LSI flagged consumer softening — the internal transfer is now explicitly dilutive to DX.
- HBM margins are currently inferior to conventional DRAM — management's own admission — so the >3x HBM sales ramp is dilutive to memory mix margins until at least 2027 repricing.
- CapEx opacity persists: Q1 spend fell 45% QoQ while guiding a "substantial" FY increase with no number; the >110T multi-year plan plus Fab 2 review plus M&A appetite signals capital intensity that could compress FCF-based returns just as the post-2026 policy is written.
- Earnings quality still opaque: no segment OP for seven quarters, no HBM revenue absolute, no foundry loss figure, and now an unquantified provision overhang.
- Consumer end-demand is deteriorating: smartphone volumes guided to "significant" decline in 2026, tablets down, mobile/PC demand destruction from memory prices acknowledged by the memory team itself — the set businesses face a shrinking-volume, rising-cost pincer.
- Geopolitical/freight risk is now management-flagged: Middle East tensions, oil-driven freight inflation, and tariffs all cited as 2H headwinds.
Next-quarter watchlist- Strike and provision: outcome of the May 21–June 7 general strike — any production disruption quantified; incentive negotiation conclusion and the size/timing of the booked cost.
- Memory pricing durability: Q2 DRAM bit guide (single-digit) and NAND (low-single) vs actuals; ASP trajectory after +90s/+80s%; whether multiyear contract terms (pricing, volume, duration) get any disclosure; 2027 pre-booking evidence.
- HBM4 ramp: 2H volume scaling on schedule; Q3 mix crossing 50% of HBM sales; HBM4E sample shipments within Q2 as promised; any named customer; HBM vs conventional DRAM margin commentary vs the 2027 normalization thesis.
- Foundry: Q2 full-utilization claim vs actuals; 2nm gen-2 MP progress; conversion of "active talks" with large AI/HPC customers into awards; silicon photonics MP start; Taylor Fab 1 operations start; Tesla — press after three quarters of silence.
- CapEx trajectory: Q2 equipment-spend pickup as promised; any FY2026 number or range; DRAM/NAND/foundry split; progress on the >110T plan and Fab 2 decision.
- Capital return: next-policy (post-2026) framework signals; whether the Q4-style additional dividend recurs; any further buyback after the 10T program's completion.
- MX: S26 sell-through and ASP vs the value-growth claim; memory-cost pass-through evidence; the guided profitability decline's magnitude; A57/A37 and foldable traction; AI glasses timing.
- DA restructuring: specifics on the "selective and focused" portfolio actions; FläktGroup North America/Korea expansion milestones.
- Earnings quality: segment OP restoration; net income walk; quantification of one-offs; MX unit/ASP disclosure restoration.
- Corporate: Middle East freight-cost impact on logistics; robotics "tangible progress" evidence; any strike-related disclosure becoming a standing item.
|
| Jan 28, 2026 | +1.01% | Q4 FY2025 | Read transcript briefingQuarter in one view- Q4 FY2025: record revenue KRW 93.8T (+9% QoQ); record OP KRW 20.1T (+7.9T QoQ); OPM 21.4% (+7.3pp). SG&A KRW 24.2T (+2.9T QoQ; 25.8% of sales, +1pp). R&D KRW 10.9T in Q4; FY record KRW 37.7T.
- DS drove everything: revenue +33% QoQ on HBM and high-value products plus price strength; Memory set another all-time quarterly revenue record, beating the record set one quarter ago. DX revenue -8% QoQ (fading smartphone launch effect, U.S. tariff hit to appliances); DX OP declined.
- FX flipped to a tailwind: dollar strength added ~KRW 1.6T to company-wide OP, centered on components — versus roughly neutral in Q3 and -0.5T in Q2.
- Memory mechanics: DRAM and NAND bit growth both "matched guidance" (DRAM low-single, NAND decline); DRAM ASP +~40% QoQ; NAND net ASP +mid-20s% QoQ; server mix +10pp QoQ. Bit shipments hit a record high despite the capped guide.
- HBM4 is the headline: final qualification phase, production already commenced, 11.7Gbps top bin, shipments start February; 2026 HBM sales guided to more than triple YoY with all production-ready capacity fully booked by customer POs.
- CapEx: Q4 KRW 20.4T (+11.2T QoQ; DS 19.0T, Display 0.7T); FY25 KRW 52.7T (-1.0T YoY; DS 47.5T, Display 2.8T) — well above the prior KRW 47.4T FY guide. 2026 memory CapEx guided to increase again.
- Capital return: Q4 dividend KRW 566/567 (common/preferred) including a KRW 1.3T additional dividend to qualify for the government's separated-taxation scheme; FY25 FCF ~KRW 36.5T; KRW 6.6T of the 8.2T bought back in 2025 to be canceled; cancellation schedule to be disclosed in Q1 2026.
What management is focused on- Victory lap, extended: CFO framed 2025 as "the clear turnaround" delivered "as we promised" — highest annual revenue ever, record Q4 profit — and quoted customers saying "Samsung is back."
- HBM4 as proof of regained technology leadership: performance targets set above JEDEC from the outset, no redesign needed despite customers raising requirements, stable full-scale production, February shipments. Memory's stated 2026 goal: "regain our core technology leadership."
- Shortage economics: management repeatedly emphasized demand exceeding supply across HBM, conventional DRAM, and NAND "for the time being," with multiyear supply-contract requests from GPU/ASIC developers and hyperscalers — and an explicit intent to respond only "selectively" to hedge investment risk.
- Portfolio flexibility between HBM and server DDR: notably, management said near-term mix may tilt toward server DDR over HBM "from a profitability perspective," while committing to a balanced mix long term — the HBM-vs-DRAM margin tension from Q3 is now stated openly.
- One-stop solution as the strategic identity: logic + memory + foundry + advanced packaging integration (HBM base dies on logic, core dies on memory, 3D hybrid copper bonding) pitched as unique; foundry 2nm project awards expected +130% YoY in 2026.
- Cost defense in set businesses: MX, VD, Display, and System LSI all framed 2026 around surviving memory-cost inflation — strategic supplier partnerships, resource efficiency, premium mix — with no quantified margin offset anywhere.
- New growth engines: acquisitions (FläktGroup HVAC, ZF ADAS, Xealth, Masimo audio), humanoid robotics ("tangible progress" promised in 2026), NRD-K R&D complex expansion.
Key numbers and quarter mechanics- Consolidated: revenue KRW 93.8T (+9% QoQ); OP KRW 20.1T (+7.9T QoQ); OPM 21.4% (+7.3pp); SG&A KRW 24.2T (25.8% of sales); Q4 R&D KRW 10.9T; FY R&D KRW 37.7T (record). FX +~KRW 1.6T to OP.
- Divisional: DS revenue +33% QoQ; DX revenue -8% QoQ with OP down (MX and appliances). No segment OP disclosed — sixth consecutive quarter.
- Memory Q4: DRAM bits +low-single QoQ (in line with guide); DRAM ASP +~40% QoQ; NAND bits declined QoQ (in line with the ~-10% guide; planar NAND exit, migration losses, low inventory); NAND net ASP +mid-20s% QoQ (blended ASP muted by QLC mix and planar phase-out); server mix +10pp QoQ; total bit shipments a record high.
- Q1 2026 memory guides: DRAM bits +low-single QoQ (very low inventory); NAND bits +mid-single QoQ (base effect).
- MX Q4: 60M smartphones, 6M tablets, ASP $244 (vs 61M / 7M / $304 in Q3 — shipments and ASP down as guided); revenue and profit down QoQ, revenue up YoY; full-year double-digit profitability maintained; flagship volume and sales grew for the year. Q1 guide: shipments and ASP up on S26 launch; tablets flat.
- Foundry Q4: revenue up QoQ (U.S. and China demand; 2nm gen-1 ramp; 4nm HBM base-die shipments started) but earnings improvement "limited due to recognition of provisions" — a new, unexplained one-off. Q1: revenue down QoQ on seasonality. 2026 target: double-digit YoY revenue growth, continued earnings improvement; 2nm gen-2 MP in 2H26; 1.4nm MP targeted 2029, PDK 1.0 in 2H27; Taylor operations to start this year.
- System LSI Q4: earnings down QoQ (customer seasonality, launch-schedule adjustments); image sensors grew (200MP/50MP). First time on the call as a separate voice.
- CapEx: Q4 KRW 20.4T (DS 19.0T, Display 0.7T); FY25 KRW 52.7T vs the KRW 47.4T guide — a ~5.3T overshoot, driven by Q4 DS spending (memory advanced-node transitions, Taylor). Display FY 2.8T vs 3.3T guide (8.6G line complete).
- Capital return: FY25 FCF ~KRW 36.5T; 50% = ~KRW 18.3T return basis; regular dividends KRW 9.8T + additional KRW 1.3T; Q4 DPS KRW 566 common / 567 preferred (regular 371 + additional 196); payable April after March AGM. Of KRW 8.2T shares acquired in 2025, KRW 6.6T to be canceled (note: Q3 had flagged KRW 8.4T cancellation from the 10T program — the 6.6T figure appears to cover only the 2025 purchases; reconcile).
- Not disclosed: net income, segment OP, absolute HBM revenue, foundry loss/provision magnitude, strike status (sixth quarter), Tesla execution detail (again).
Product and launch scorecard- HBM4: the quarter's defining delivery. Final qualification phase, production commenced, 11.7Gbps highest bin, February shipments, 2026 HBM sales guided >3x YoY, capacity fully booked via POs, and 2027 supply discussions already underway. Still no named customer — but "fully booked with customer POs" is the strongest demand evidence yet.
- HBM4E: standard-product sampling ~mid-2026; custom HBM on 4E core dies in 2H26. HBM3E/4 16-high deliberately skipped for mass commercialization (limited demand; 4E 12-high covers density); TC-NCF 16-layer kept MP-ready as insurance; HCB samples shipped, partial commercialization at 4E stage.
- HBM3E: near-term capacity expansion explicitly directed at 3E demand while 1c capacity is built for 4/4E — 3E remains a revenue workhorse in 2026.
- 2nm foundry: gen-1 ramping with stabilization focus; gen-2 hitting yield/performance targets, MP 2H26; 2026 2nm project awards expected +130% YoY (HPC/AI-led). Tesla referenced only as "the Tesla award" in passing — still no execution update.
- S26: confirmed 1H26 launch with "second-generation custom AP" — the Exynos question from last quarter remains unresolved in language; "custom AP" is not a commitment to Exynos 2600. New camera sensors and agentic AI as selling points.
- TriFold: launched December 2025 — delivered on the "upcoming launch" commitment; new foldables planned for 2H26.
- Memory portfolio: SOCAMM 2, GDDR7, high-density DDR5 expansion; V9-based PCIe Gen 6 SSDs positioned for the 2H26 Gen 6 server ramp; Key-Value SSD for AI inference as a new demand category; QLC benefiting from nearline HDD substitution.
- Display: 8.6G IT OLED line complete, mass production starting 2026 — revenue growth driver; QD-OLED monitor expansion continues.
- VD: Micro RGB line expanding to 55–130 inches in 2026; Vision AI Companion showcased at CES; TV Plus and Art Store as service growth engines; 1H26 sports events (Winter Olympics, World Cup) as demand catalysts.
Sell-side read-through- Daiwa (S.K. Kim) extracted the quarter's key memory mechanics: DRAM ASP +~40% QoQ, NAND net ASP +mid-20s%, server mix +10pp, bits in line with guide — plus confirmation that bit shipments still hit a record.
- Citi (Sei Cheol Lee) asked the HBM4 qualification question directly and got the most substantive answer in this transcript series: final qualification phase, production started, 11.7Gbps, February shipments, 4E sampling mid-2026, 16-high strategy, and the >3x 2026 HBM sales guide with fully booked capacity. Notably, no one asked about NVIDIA by name this quarter — the qualification answer was given without naming customers.
- JPMorgan (Jay Kwon) drew out the 2026 memory operating stance: shortage across all categories, multiyear contract requests handled selectively, and the striking admission that mix may favor server DDR over HBM near-term on profitability — the HBM margin question the sell-side failed to press last quarter was effectively answered unprompted.
- Meritz (Sunwoo Kim) got 2026 memory CapEx direction: meaningful increase, weighted to preemptive fab/cleanroom space plus 1c DRAM and V9 NAND capacity acceleration; NRD-K expansion. No number given.
- UBS (Nicolas Gaudois) got foundry node detail (2nm gen-2 on track, 1.4nm MP 2029, PDK 1.0 2H27, +130% 2nm awards) and the one-stop turnkey framing; the VD answer was strategy boilerplate.
- BofA (Simon Woo) got the capital-return mechanics: FCF 36.5T, 18.3T basis, 6.6T cancellation, Q1 2026 board disclosure on cancellation timing; MX's memory-cost answer was partnerships + efficiency, unquantified.
- NH (Young Ho Ryu) got System LSI's cost-pressure read (mid/low-end squeezed, premium resilient) and confirmation that a post-2026 shareholder-return policy is under active review.
- Missing challenges: no one pressed the foundry "provisions" magnitude; no strike question (sixth quarter); no segment OP request; no one asked why FY25 CapEx overshot the guide by ~5T; no Tesla execution follow-up; no quantification of MX margin defense against memory inflation; no question on the additional dividend's one-off vs recurring nature.
Management credibility- Delivered: Q4 DRAM/NAND bit guides met exactly as stated; server SSD mix shift "substantial" as guided (+10pp); MX shipment/ASP decline as guided; HBM4 moved from "samples at all customers" (Q3) to production commenced and February shipments — on the customer-timeline commitment; TriFold launched December as promised; FY25 CapEx cut delivered (52.7T, -1.0T YoY) though above the 47.4T guide; dividend trajectory continued (370→566 including the special).
- Candor improvements: DRAM ASP +40% and NAND ASP +mid-20s% disclosed with mix caveats (QLC, planar exit) explained; the HBM-vs-server-DDR profitability trade-off stated openly rather than discovered; foundry provisions acknowledged as limiting earnings improvement; mobile/PC demand-destruction risk from memory prices flagged by the memory team itself.
- Open items: foundry provisions unexplained and unquantified — the one-off opacity pattern continues; HBM4 still has no named customer despite "fully booked" claims; the 6.6T cancellation vs the previously flagged 8.4T needs reconciliation; "Samsung is back" customer quote is anecdote, not evidence; S26 "custom AP" language continues to dodge the Exynos commitment question.
- Consistency watch: Q3's "2026 HBM bit plan backed by secured demand" hardened into "fully booked with customer POs" and a >3x sales guide — a verifiable escalation; the Q3 HBM-capacity-gating concern resolved into simultaneous 3E expansion and 1c build-out, funded by the Q4 CapEx surge; the FY25 CapEx overshoot vs guide is the quarter's main guidance miss and went unaddressed on the call.
What changed versus the prior quarter- The recovery accelerated rather than plateaued: OP 12.2T→20.1T, OPM 14.1%→21.4%; Q3's price-led thesis compounded — DRAM ASP went from +mid-teens to +~40% QoQ.
- HBM4 went from development-complete/samples to production commenced with a February ship date and a >3x 2026 sales guide — the qualification overhang substantially de-risked in one quarter, though customer names remain absent.
- The HBM-vs-DRAM margin tension moved from implicit (Q3 capacity gating) to explicit: management said server DDR may be prioritized over HBM near-term on profitability — effectively confirming last quarter's unexamined margin-compression warning.
- CapEx pivoted hard: Q4 spend of 20.4T (more than double Q3's 9.2T) pushed FY25 to 52.7T vs the 47.4T guide; 2026 memory CapEx guided up again with emphasis on preemptive cleanroom/fab space — the investment cycle is now running hot.
- FX swung from roughly neutral to +1.6T — a meaningful, possibly underappreciated, contributor to the OP beat.
- Foundry's narrative gained a caveat: revenue up and 2nm ramping, but "provisions" limited earnings improvement — the clean loss-reduction story from Q3 acquired a new one-off.
- Capital return shifted structure: the additional 1.3T dividend is tax-scheme-driven (payout ≥25%, +10% YoY growth requirements), and the cancellation quantum was restated as 6.6T of 2025 purchases with timing deferred to a Q1 2026 board decision.
- System LSI appeared as a distinct executive voice for the first time; the Exynos flagship question remains unanswered heading into the S26 launch quarter.
Bull case- Pricing power is compounding: DRAM ASP +40% QoQ with further increases implied, NAND net ASP +mid-20s%, shortages across all memory categories "for the time being," multiyear contract requests incoming, and Q1 guides still supply-capped — the upcycle has both price and duration.
- HBM4 is de-risking on schedule: production started, February shipments, 11.7Gbps top bin, capacity fully booked by POs, 2026 HBM sales >3x YoY, and 2027 supply talks already underway — the qualification narrative has shifted from "if" to "how much."
- Capacity is being built ahead of demand: Q4 CapEx 20.4T, preemptive cleanroom strategy, 1c DRAM and V9 NAND acceleration, 3E expansion plus 4/4E build-out in parallel — Samsung is buying shortage leverage.
- Foundry momentum is broadening: 2nm gen-1 ramping, gen-2 on target for 2H26, +130% 2nm project awards expected, 4nm HBM base-die shipments started, Taylor operational this year, double-digit 2026 revenue growth targeted.
- Cash generation now funds returns at scale: 36.5T FCF, 18.3T return basis, 11.1T total dividends, 6.6T cancellation queued, and a post-2026 policy under active review from a position of strength.
- The set businesses hold premium ground: flagship volume/sales grew in 2025, foldables strong, S26 cycle starting, Micro RGB creating a new TV category, 8.6G IT OLED entering production.
Bear case- A large slice of the beat is price and FX, not volume: DRAM bits grew only low-single, NAND bits declined, and +1.6T of OP came from currency — if ASP momentum stalls, the earnings trajectory flattens quickly.
- HBM economics remain suspect: management openly said server DDR may be prioritized over HBM near-term on profitability — HBM's margin premium appears gone even as HBM4 ramps; the >3x sales guide may carry lower incremental margins than the headline suggests.
- Memory is eating its own customers: MX, System LSI, and Display all flagged memory-cost inflation as a 2026 headwind; mobile/PC demand destruction (higher end prices, lower content per box) was flagged by the memory team itself — the shortage is a group-level transfer, not pure upside.
- Foundry's improvement stalled on unexplained provisions: earnings improvement "limited" in Q4, Q1 revenue guided down, loss magnitude still undisclosed, and Tesla execution remains unaddressed two quarters after the award.
- Earnings quality still opaque: no segment OP for six quarters, no net income discussion, one-off magnitudes never quantified, and now a new provisions line with no detail.
- CapEx overshoot without explanation: FY25 landed ~5T above guide; if 2026's "meaningful increase" is similarly unbounded, FCF-based returns could compress just as the return policy is renegotiated.
- DX is the weak link: revenue -8% QoQ, OP down, appliances hit by tariffs, VD facing stagnant TV demand and Chinese pricing — the consumer franchise is not participating in the recovery.
- The additional dividend is scheme-driven, not policy-driven: it exists to meet a government tax qualification (25% payout, +10% growth) — its recurrence is conditional, not committed.
Next-quarter watchlist- HBM4 conversion: first revenue recognition post-February shipments; any named customer or qualification confirmation; 4E sampling on time mid-2026; whether the >3x 2026 HBM sales guide is reiterated; HBM vs server DDR mix commentary as the profitability signal.
- Memory pricing and bits: Q1 DRAM bit guide (low-single) and NAND (mid-single) vs actuals; ASP trajectory after +40%/+mid-20s%; multiyear contract signings and terms; any sign of mobile/PC demand destruction quantified.
- Foundry: the provisions — magnitude, nature, recurrence; 2nm gen-1 ramp yields and gen-2 MP progress; the +130% 2nm awards claim vs actual bookings; Tesla execution — press after two quarters of silence; Taylor operational start.
- CapEx discipline: 2026 memory CapEx number or range; reconciliation of the FY25 overshoot vs the 47.4T guide; DRAM vs NAND vs foundry split.
- Capital return: Q1 2026 board disclosure on the 6.6T cancellation schedule; reconciliation with the previously flagged 8.4T; whether the additional dividend becomes recurring; early signals on the post-2026 policy framework.
- MX: S26 launch reception, shipment/ASP delivery vs the up-guide; the AP question — does "second-generation custom AP" mean Exynos 2600; memory-cost pass-through evidence in pricing or margins; TriFold sell-through.
- Earnings quality: segment OP restoration; net income disclosure; any quantification of one-offs across the past three quarters.
- Corporate: strike status — press after six quarters of silence; humanoid robotics "tangible progress" claim; integration of the four 2025 acquisitions.
|
| Oct 29, 2025 | +4.93% | Q3 FY2025 | Read transcript briefingQuarter in one view- Q3 FY2025: revenue KRW 86.1T (+15.4% QoQ); OP KRW 12.2T (+7.5T QoQ); OPM 14.1% (+7.9pp). SG&A KRW 21.3T, but SG&A/sales fell 3.1pp to 24.8%. YTD R&D hit a record KRW 26.9T.
- The CFO's "Q2 bottom, 2H rebound" call was delivered: OP more than doubled QoQ, led by DS (memory all-time-high quarterly sales, improved foundry utilization, "significantly reduced" inventory value adjustments vs Q2) plus DX (Fold 7 mix).
- Divisional revenue: DS +19% QoQ (memory record quarter on HBM3E + server SSDs); DX +11% QoQ (foldables launch, flagship sales). No segment OP disclosed again.
- FX: won strength hurt components but was "largely offset" by DX positives — minimal net OP impact, a big change from Q2's -0.5T hit.
- Memory mechanics: DRAM bit growth mid-teens % QoQ (vs +high-single guide — a beat); DRAM ASP +mid-teens % QoQ (HBM mix + broad price rises); NAND bits +~10% QoQ (vs +mid-single guide — beat); NAND ASP +mid-single-digit QoQ. Inventories reduced further; industry inventory now "subnormal."
- HBM: bit shipments +mid-80s% QoQ; mix now fully transitioned to HBM3E (ex-tail legacy). HBM4 development finished, samples at all customers, 11Gbps+ claimed, MP ready on customer timelines.
- CapEx KRW 9.2T (-1.9T QoQ, -3.3T YoY; DS 7.8T, Display 0.8T). FY25 guide KRW 47.4T (-6.3T YoY). 2026 memory CapEx guided to a "significant" YoY increase; foundry CapEx to rise back to "2024 levels."
- Capital return: Q3 dividend KRW 370/share (vs 367 in Q2), KRW 2.45T payable late November; KRW 10T buyback completed September 29 (ahead of schedule); KRW 8.4T of treasury shares to be canceled, KRW 1.6T reserved for employee stock comp; new all-employee PSU program announced, additional buybacks to fund it.
What management is focused on- Vindication framing: CFO opened by acknowledging market/shareholder concern "through the previous quarter" and declaring Q3 a "clear rebound" that "meaningfully met" expectations — a direct victory lap on the Q2-bottom call.
- AI demand as the through-line: every business (memory, foundry, display, MX, VD) framed around AI; memory demand for 2026 described as arriving "much stronger and faster than usual," with customer demand exceeding supply even after planned capacity expansion.
- Supply discipline as strategy: memory explicitly running "profitability-focused operations" — Q4 DRAM bit growth guided to only low-single-digit and NAND bits to decline ~10% QoQ, prioritizing price/mix over volume into a shortage.
- 2026 CapEx pivot: after a -6.3T YoY cut in 2025, memory CapEx guided to a significant increase (DRAM-weighted, 1b-nm portfolio, advanced-node transitions, some construction); foundry to finish Taylor construction and prep 2nm gen-2 / 1.4nm, back to 2024 spending levels.
- HBM4 positioning: performance targets set "above customer requirements" (11Gbps+, low power); 1c-nano capacity expansion flagged as customers raise performance asks; 2026 HBM bit plan already has "significant customer demand" secured, with further expansion under review — but weighed against now-highly-profitable conventional DRAM.
- Employee alignment: PSUs for all employees (3-year stock-performance-linked, 3-year installment vesting) plus OPI stock comp extended company-wide from January 2026 — framed as long-term value alignment, and the mechanism for using the 1.6T treasury-share reserve.
- Cautious 2H26: CFO deferred detailed 2H26 outlook to the Q2 2026 call, citing tariff uncertainty; memory team flagged geopolitical risks (tariffs, AI-chip export controls) for 2H26 specifically.
Key numbers and quarter mechanics- Consolidated: revenue KRW 86.1T (+15.4% QoQ); OP KRW 12.2T (+7.5T QoQ); OPM 14.1% (+7.9pp); SG&A KRW 21.3T (24.8% of sales, -3.1pp QoQ); YTD R&D KRW 26.9T (record).
- Divisional revenue: DS +19% QoQ; DX +11% QoQ. Memory set an all-time quarterly sales record. No segment OP disclosed (fifth consecutive quarter without it on the call).
- Memory Q3: DRAM bits +mid-teens QoQ (beat vs +high-single guide); DRAM ASP +mid-teens QoQ; NAND bits +~10% QoQ (beat vs +mid-single guide); NAND ASP +mid-single QoQ; inventories down further. Q2 inventory adjustments "significantly reduced" as guided — still no magnitude given.
- HBM Q3: bits +mid-80s% QoQ; HBM3E now effectively 100% of the mix (ex-legacy tail). No absolute HBM revenue disclosed; the Q2-guided "3E >high-90s% of HBM revenue" was not explicitly confirmed with a number.
- Q4 memory guides: DRAM bits +low-single-digit QoQ; NAND bits ~-10% QoQ (V8/V9 migration losses + low inventory); server SSD mix to rise "substantially"; prices expected to rise further across DRAM and NAND.
- MX Q3: 61M smartphones, 7M tablets, ASP $304 (vs 58M / 7M / $270 in Q2 — shipments and ASP up as guided); "robust double-digit profitability" maintained; Fold 7 drove double-digit foldable growth in shipments and value YoY. Q4 guide: smartphone shipments and ASP down QoQ, tablets down QoQ.
- Foundry: revenue flat QoQ (China export-control headwinds offset by U.S. customers and memory products); loss significantly reduced on lower one-offs, better utilization, cost savings; record-high order backlog; first-gen 2nm product entered mass production. Q4: 2nm ramp, further earnings improvement expected.
- CapEx: Q3 KRW 9.2T (DS 7.8T, Display 0.8T); 9M25 KRW 32.3T (-3.6T YoY; DS 28.5T, Display 2.1T); FY25 guide KRW 47.4T (-6.3T YoY; Display 3.3T, -1.6T). Transcript states DS FY25 CapEx of "KRW 4.9 trillion, down 5.4 trillion" — almost certainly a garble for ~KRW 40.9T given 9M DS CapEx was already 28.5T; verify against the release.
- Capital return: dividend KRW 370/share, KRW 2.45T, late November; KRW 10T buyback completed Sept 29; KRW 8.4T cancellation decided "at an appropriate time"; KRW 1.6T for stock comp; future buybacks planned to fund PSUs.
- Not disclosed: net income, segment OP, absolute HBM revenue, one-off magnitudes (again), foundry loss magnitude, strike status (fifth consecutive quarter), Tesla order execution detail.
Product and launch scorecard- HBM3E: the recovery is now the P&L driver — bits +mid-80s% QoQ, mix fully 3E, memory's record quarter attributed to 3E plus server SSDs. But the NVIDIA qualification question was asked directly and refused on NDA grounds — the single most-wanted data point remains unconfirmed.
- HBM4: development complete, samples with all customers, 11Gbps+ on low power claimed, MP-ready on customer timelines; 1c-nano capacity expansion planned. Language is confident but still no named customer, no qualification confirmation, no firm MP date.
- 2nm foundry: first-gen 2nm product in mass production (a real milestone vs Q2's "reliability complete"); record order backlog; gen-2 2nm on plan; Taylor fab construction to finish with 2026 operations; HBM4-based die and 4nm performance/power variants in the pipeline. Tesla was not mentioned once this quarter — a notable silence after being Q2's headline.
- Foldables: Fold 7 credited for DX profit growth and double-digit foldable shipment/value growth YoY — tracking the annual double-digit target. Tri Fold now "upcoming launch" (reconfirmed); Galaxy XR described as "recently released" — both Q2 commitments delivered or on track.
- Exynos: adopted in Flip 7 and some A-series this year, but S26 AP evaluation "still underway" — no flagship commitment for 2026, a softening versus the 1H26 Exynos 2600 flagship target stated in Q2.
- Memory portfolio: QLC ultra-high-density SSD ramp with large data centers starting Q4; 16TB+ TLC SSDs for AI servers; 10.7Gbps LPDDR5X and UFS 5.0 for on-device AI; GDDR7, high-density DDR5 as 2026 focus.
- Display: 8.6G IT OLED line investment "nearing completion," products in 2026; QD-OLED monitor momentum (double-digit large-panel sales growth QoQ); automotive rigid OLED expansion.
- VD: micro RGB as the new form-factor answer to Chinese price competition; Vision AI Companion (Bixby conversational AI) launched September, expanding in 2026; TV Plus/services as profit driver. Profit still down YoY in Q3 on competition costs.
Sell-side read-through- Citi (unnamed analyst) got the quarter's key mechanics: DRAM bits +mid-teens, ASP +mid-teens; NAND bits +~10%, ASP +mid-single; Q4 guides (DRAM bits low-single, NAND -10%); plus the buyback completion and 8.4T/1.6T cancellation split from the CFO.
- KB (Kim Dongwon) asked the NVIDIA HBM3E qualification question directly — refused on NDA. What was extracted: HBM bits +mid-80s% QoQ, full 3E mix transition, HBM4 samples at all customers with 11Gbps+, and the CFO's disclosure that 2026 HBM bit plans are already backed by secured customer demand with expansion under review — but HBM vs conventional DRAM relative profitability now explicitly gates further HBM capacity adds. That last point is new and important: conventional DRAM's price spike is competing with HBM for wafers.
- JPMorgan (Jay Kwon) got the 2026 memory outlook: demand exceeds supply even at maximum production; legacy (DDR4/LPDDR4X/GDDR6) constraints persist; NAND nearline-HDD substitution (QLC SSD) tailwind; but an explicit 2H26 caution flag on tariffs/export controls. Also drew out the full PSU/OPI stock-comp mechanics.
- SK Securities (Han Dong-hee) got foundry loss-reduction drivers (one-off roll-off, utilization, cost savings) and Q4 2nm ramp confirmation; MX's answer on memory-cost pressure was flagship mix + component standardization — no quantified margin defense.
- UBS (Nicolas Gaudois) got the 2026 memory CapEx signal: "significant" YoY increase, DRAM-weighted, 1b-nm-based bit growth plus construction; NAND advanced-process share to rise only "after confirming market demand" — an asymmetry worth noting.
- Daiwa (S.K. Kim) got foundry CapEx direction: 2025 cut explained by line-conversion focus; 2026 back to ~2024 levels with Taylor completion and 2nm gen-2/1.4nm prep. Display answer was IT/auto OLED strategy boilerplate.
- Goldman (Giuni Lee) got two useful disclosures: Galaxy AI usage metrics (60% weekly / 80% monthly feature usage) and the Exynos S26 non-commitment; the TV answer conceded share pressure from Chinese entry-level pricing and laid out micro RGB + real-QLED marketing + services as the response.
- Missing challenges: no one pressed the size of Q2/Q3 inventory adjustments (third straight quarter); no strike question (fifth quarter); no foundry loss magnitude despite the improvement narrative; no Tesla order follow-up despite it being Q2's centerpiece; no one asked why the Q2-guided "3E >high-90s% of HBM revenue" wasn't confirmed with a figure; no segment OP request; no question on the HBM3E margin-compression warning from last quarter — management's sharpest Q2 concession went entirely unexamined.
Management credibility- Delivered: the Q2-bottom/2H-rebound call was emphatically validated (OP +7.5T QoQ); one-offs "significantly reduced" as guided; DRAM and NAND bit guides beaten; MX shipment/ASP recovery delivered; buyback completed ahead of schedule (Sept 29 vs Oct 8 target); dividend raised again (367→370); 2nm first product in MP; Tri Fold and XR both reconfirmed/delivered; 2025 CapEx cut executed as flagged.
- Candor: 2H26 caution flagged unprompted (tariffs, export controls); HBM capacity expansion explicitly conditioned on relative profitability vs conventional DRAM — an honest signal that HBM economics are being weighed, consistent with last quarter's margin-compression warning; Exynos S26 non-answer was at least plainly stated rather than spun.
- Open items: one-off magnitudes still never quantified (Q2 or Q3) — the "clean" run-rate remains unknowable; NVIDIA HBM3E qualification status refused; HBM4 has samples and specs but no named customer or MP date; Tesla order received zero follow-up commentary; foundry loss magnitude undisclosed; the FY25 DS CapEx figure in the transcript (4.9T) is internally inconsistent and must be verified.
- Consistency watch: Q2's "HBM3E supply exceeds demand, margin gap narrows sharply" warning was neither repeated nor updated — yet Q3's profitability-first bit discipline (Q4 DRAM bits low-single, NAND -10%) and the HBM-vs-DRAM profitability gating look like the behavioral response to exactly that warning. Also, last quarter's "2026 CapEx higher" for foundry is now quantified as "2024 levels."
- Transcript garbles to verify: DS FY25 CapEx "KRW 4.9 trillion" (likely 40.9T); "GDDR74 servers" (GDDR7 for servers); "SOCAR" (likely SoC/CAMM or similar); "17 nano CLS" (likely 1.4nm class); "LC SCDs" (QLC SSDs); "NPI" vs "OPI" incentive naming; "weighted" (weighed); "moving to sales" (monitor sales).
What changed versus the prior quarter- The rebound arrived on schedule: OP 4.7T→12.2T, OPM 6.3%→14.1%; the Q2 one-offs shrank as promised; FX went from -0.5T headwind to roughly neutral.
- Memory flipped from volume recovery to price/mix leverage: Q2 was bit-driven (NAND +high-20s%); Q3 delivered ASP power (DRAM +mid-teens) and Q4 guides deliberately restrict bits (DRAM low-single, NAND -10%) into a shortage — the pull-in payback concern from Q1/Q2 is moot; the constraint is now supply, not demand.
- HBM narrative shifted from mix transition to capacity allocation: Q2 was about 3E share of volume; Q3 is about whether to expand HBM4/1c capacity given conventional DRAM's "sharp improvement in profitability" — last quarter's margin-compression warning is now shaping capital allocation.
- 2026 CapEx pivoted from cuts to increases: FY25 -6.3T YoY confirmed, but memory guided to a significant 2026 increase and foundry back to 2024 levels — the investment cycle is turning back up on AI demand visibility.
- Tesla disappeared: Q2's named $16.5B anchor order got no mention; foundry narrative this quarter was utilization, one-off roll-off, 2nm MP, and record backlog instead.
- Exynos softened: Q2's "Exynos 2600 first 2nm flagship, 1H26 customer target" became "S26 AP evaluation still underway" — a visible walk-back of flagship commitment.
- Shareholder returns entered a new phase: buyback complete, cancellation quantum fixed (8.4T), and a new all-employee PSU program announced with future buybacks to fund it — returns now linked to a compensation structure, not just a one-off program.
- Strike silence extended to a fifth quarter; segment OP silence also extended.
Bull case- The earnings recovery is broad and mechanical: record memory sales, ASPs rising across DRAM and NAND with further increases guided, industry inventories subnormal, and 2026 customer demand already exceeding Samsung's maximum planned supply — pricing power should compound through 2026.
- HBM is now a volume and mix driver with an upgrade path: bits +mid-80s% QoQ, full 3E transition, HBM4 samples exceeding specs (11Gbps+) at all customers, 2026 HBM volumes already backed by secured demand, and 1c capacity expansion ready to commit.
- Supply discipline supports margins: deliberately capping Q4 bit growth (DRAM low-single, NAND -10%) into rising prices maximizes the ASP lever; legacy DRAM/planar NAND tightness adds a second profit engine as the industry exits those nodes.
- Foundry is inflecting: loss significantly reduced, record order backlog, first 2nm product in MP, Taylor completing for 2026, 2026 CapEx restored to 2024 levels — utilization and node momentum moving together for the first time in this transcript series.
- Capital return is structural, not episodic: 10T buyback done early, 8.4T cancellation queued, dividend raised three quarters running (365→367→370), and a PSU program that requires future buybacks — an embedded return mechanism.
- Product cadence delivered: Fold 7 drove DX profit, Tri Fold and XR on track/released, S26 with second-gen custom AP and next-gen AI teased, micro RGB creating a new TV category, 8.6G IT OLED entering production in 2026.
Bear case- The NVIDIA question remains unanswered: direct qualification ask refused on NDA; HBM4 has no named customer or MP date; if Samsung's HBM4 window slips while competitors qualify, the 2026 HBM demand "secured" claim covers volume but not necessarily the leading-edge sockets.
- HBM economics are being rationed: management is explicitly weighing HBM4 capacity against now-more-profitable conventional DRAM — consistent with last quarter's warning that the 3E margin premium is collapsing; HBM may be a lower-margin business than the recovery narrative assumes.
- Earnings quality still opaque: one-off magnitudes never disclosed across two quarters; the clean Q2 base and the size of Q3's "significantly reduced" adjustments are both unknowable; no segment OP for five quarters.
- Q4 guides imply a sequential step-down in momentum: DRAM bits low-single, NAND bits -10%, MX shipments and ASP down QoQ — Q4 holds up only via price, which is fine until it isn't.
- MX margin squeeze is live: memory prices rising "steeper" in Q4 directly raise MX BOM costs; management's defense is mix and component standardization, with no quantified offset; competition intensifying in premium.
- Foundry improvement is partly one-off arithmetic: Q3's loss reduction leans on the absence of Q2's write-downs; the business is still loss-making (magnitude undisclosed), Tesla got no update, and China export controls remain a recurring sales restriction.
- 2H26 is explicitly flagged: tariffs, export controls on high-end AI chips, and a CFO who deferred the 2H26 outlook — management is telling you visibility ends in mid-2026.
- VD and the China problem: TV profit down YoY again on Chinese entry-level pricing; the response (micro RGB, marketing, services) is unproven at scale.
Next-quarter watchlist- HBM4 conversion: any named customer, qualification confirmation, or MP date; whether 1c-nano expansion is formally committed; the HBM-vs-DRAM capacity allocation decision and what it implies about relative margins; whether the NVIDIA question gets answered.
- Memory pricing follow-through: Q4 DRAM/NAND ASP realization vs the "prices rise further" guide; whether the bit caps (DRAM low-single, NAND -10%) hold or are beaten again; server SSD mix shift and QLC ramp with large data centers; 2026 contract pricing commentary.
- One-off and disclosure cleanup: demand the Q2/Q3 inventory-adjustment magnitudes; watch for segment OP restoration; verify the garbled FY25 DS CapEx figure (4.9T vs ~40.9T) against the release.
- 2026 CapEx concreteness: memory "significant increase" — get a number or range; DRAM vs NAND split; foundry "2024 levels" — reconcile with Taylor completion timing and 2nm gen-2/1.4nm prep.
- Foundry: 2nm ramp yields and customer names; Tesla order status — press after a quarter of silence; Taylor operational timeline; loss trajectory toward breakeven; record backlog conversion.
- MX: Q4 shipment/ASP decline vs guide; memory-cost pass-through strategy for S26; S26 AP decision — Exynos 2600 in or out; Tri Fold launch execution and pricing; foldable double-digit annual growth target — final tally.
- Capital return: 8.4T cancellation timing; size/timing of buybacks to fund the PSU program; board posture on returns beyond the 2024-26 policy window.
- Macro: Section 232/tariff developments and any quantified DX impact; China export-control evolution; the promised detailed 2H26 outlook at the Q2 2026 call.
- Corporate: strike status — press after five quarters of silence; Galaxy AI usage metrics (60%/80%) as a leading indicator for S26 AI monetization.
|
| Jul 30, 2025 | +0.85% | Q2 FY2025 | Read transcript briefingQuarter in one view- Q2 FY2025: revenue KRW 74.6T (-5.8% QoQ); OP KRW 4.7T (-2.0T QoQ); OPM 6.3% (-2.2pp). SG&A KRW 20.8T (-0.6T QoQ). FX flipped to a headwind: won strength cost ~KRW 0.5T OP QoQ (vs +0.5T tailwind in Q1).
- Mix flipped back: DS revenue +11% QoQ (HBM3E, DDR5, foundry customer expansion) while DX -16% QoQ (post-S25 launch fade, TV competition). Despite DS revenue growth, DS profit was hit by memory inventory value adjustments and one-off costs from China export restrictions in non-memory.
- Memory mechanics: DRAM bits +low-teens QoQ (in line with guide); conventional DRAM ASP +low-single-digit QoQ vs a guided -mid-single — a clear beat, with the legacy DRAM (DDR4/LPDDR4X) price spike starting mid-Q2. NAND bits +high-20s% QoQ (beat vs +mid-teens guide); NAND ASP -low-to-mid-single QoQ, some products up; NAND inventory down significantly for a second straight quarter, DRAM inventory below normal.
- HBM: bits +30% QoQ; HBM3E now a high-80s% share of HBM volume; 3E revenue share guided to exceed high-90s% in Q3; "meaningful expansion" of 3E sales in 2H vs 1H. First absolute HBM disclosure in three quarters.
- CapEx KRW 11.1T (-0.9T QoQ; DS 9.8T, Display 0.8T). Foundry CapEx down QoQ; Taylor 2025 spend within existing plan, higher CapEx flagged for 2026.
- Capital return: Q2 dividend KRW 367/share (vs 365 in Q1), KRW 2.45T payable late August; KRW 6.1T of the 10T buyback acquired through May; third/final tranche KRW 3.9T approved July 8 (1.1T for stock comp, 2.8T shareholder return); full program to complete October 8, ahead of the November 14 target.
- CFO called Q2 the bottom: "rebound in the second half following a bottoming out in the second quarter."
What management is focused on- Q2-as-bottom framing: CFO explicitly guided to 2H rebound with "earnings expected to improve steadily," and said DS is "truly preparing for a turnaround" after taking inventory adjustments and addressing "operational inefficiencies" — unusually direct acknowledgment of DS underperformance.
- Tariff/Section 232 watch: Korea-U.S. trade agreement welcomed as uncertainty-reducing, but the mid-August Section 232 announcement on semiconductors and derivative products (smartphones, tablets, PCs, monitors) is flagged as potentially significant; Samsung is lobbying both governments directly and indirectly.
- HBM normalization: mass-production approvals being secured "customer by customer"; 1z-nm mother die MP approval obtained; HBM4 samples on 1z process shipped to some major clients; hybrid copper bonding in volume-production discussions with key clients.
- Foundry inflection via Tesla: $16.5B next-generation order disclosed by name — the first named anchor customer for advanced nodes and Taylor; framed as the utilization/profitability inflection.
- M&A as growth engine, now with receipts: Rainbow Robotics (control, end-2024), Masimo Audio and FläktGroup (signed May), Xealth (started July); >$120M venture investments across ~40 companies in 1H — "largest half-year venture investment in Samsung's history."
- MX: foldables-led 2H (Fold7/Flip7 launched, double-digit foldable growth target vs last year), TriFold and XR headset confirmed for this year, S25 FE earlier launch, Xealth to turn Samsung Health into a connected-care platform.
- Memory pricing pivot: management now expects DRAM market prices to rise "by a greater margin" in 2H and a NAND price rally from Q3, with legacy DRAM/planar NAND rising most on tight supply — while conceding HBM3E supply has grown faster than demand and the HBM3E-vs-conventional-DRAM margin gap will "narrow sharply."
Key numbers and quarter mechanics- Consolidated: revenue KRW 74.6T (-5.8% QoQ); OP KRW 4.7T (-2.0T QoQ); OPM 6.3%; SG&A KRW 20.8T; FX -~KRW 0.5T OP QoQ.
- Divisional revenue: DS +11% QoQ; DX -16% QoQ. No segment OP disclosed on the call again.
- Memory Q2: DRAM bits +low-teens QoQ (in line); conventional DRAM ASP +low-single QoQ (vs -mid-single expected at quarter start); NAND bits +high-20s% QoQ (beat); NAND ASP -low-to-mid-single, some products up; DRAM inventory below normal, NAND inventory down sharply two consecutive quarters.
- One-offs: memory inventory value adjustments plus non-memory one-off costs from China export restrictions; memory one-offs "expected to significantly be reduced in Q3" — reduced, not eliminated.
- HBM Q2: bits +30% QoQ; 3E share of HBM volume high-80s%; Q3 guide: 3E >high-90s% of HBM revenue; 2H 3E sales meaningfully above 1H.
- Q3 memory guides: DRAM bits +high-single-digit QoQ; NAND bits +mid-single-digit QoQ — a sharp deceleration from Q2's +high-20s% NAND, consistent with inventory now normalized.
- MX Q2: 58M smartphones (vs 61M Q1), 7M tablets, ASP $270 (vs $326 Q1); revenue and OP up YoY; double-digit profitability maintained. Q3 guide: smartphone shipments and ASP up QoQ, tablets down QoQ.
- Foundry: 3nm GAA mobile in full-scale volume production; "significant" QoQ revenue increase on U.S./China key customers; but China AI-chip export controls caused sales restrictions and inventory adjustments, and mature-node underutilization still weighed on profit. 2nm gen-1 reliability complete; 2nm gen-1 mobile MP drives 2H revenue improvement.
- System LSI: record 1H revenue (GAA flagship SoC MP + 200MP sensors); advanced-product development costs capped profitability. Exynos 2500 in the July foldable; Exynos 2600 to be first 2nm GAA flagship, targeting 1H26 key-customer flagships.
- Display: mobile panel revenue up QoQ (new-customer products, IT/auto); large panels up on QD-OLED TV/monitor; 500Hz gaming monitor launched.
- VD: premium mix (Neo QLED, OLED, Super Big) expanded but profit down YoY on stagnant demand and competition costs.
- Not disclosed: net income, segment OP, absolute HBM revenue, one-off magnitudes, strike status (fourth consecutive quarter), foundry loss magnitude.
Product and launch scorecard- Enhanced HBM3E: the conversion is now visible — bits +30% QoQ, 3E at high-80s% of volume, MP approvals accumulating "customer by customer," Q3 revenue share >high-90s% guided. This is the first quarter the recovery thesis has hard numbers behind it. Caveat: management itself flagged 3E supply outpacing demand industry-wide and a sharply narrowing margin premium vs conventional DRAM.
- HBM4: 1z-based samples shipped to some major clients; base die on advanced logic process; 2x HBM3E performance claimed; supply expansion aligned to demand; hybrid copper bonding in customer discussions for volume production. MP timing language ("as HBM4 demand enters full scale") is less pinned than last quarter's "MP 2H25."
- Tesla foundry order: $16.5B, next-gen product on advanced process, named on the call; Taylor ramp from 2026, 2026 CapEx guided higher. The single most concrete foundry win disclosed in this transcript series.
- 2nm: gen-1 reliability complete (repeated from Q1), full-scale mobile MP in 2H; new "two-track" strategy (lead-customer node hardening, then platform opening) — an implicit admission the single-track approach struggled with yield/maturity/IP.
- Exynos: 2500 re-entered a flagship (July foldable) — the design-win evidence missing last quarter; 2600 committed as first 2nm GAA flagship with 1H26 customer target. Credibility partially restored.
- Foldables: Fold7 (8.9mm folded/4.2mm unfolded, 215g, 200MP camera) and Flip7 launched; initial results "in line with expectations," double-digit foldable growth targeted for the year; 500k folding test passed. TriFold explicitly confirmed for this year — a shift from last quarter's non-answer.
- XR headset: confirmed for this year, Google partnership, multimodal AI.
- Memory portfolio: LPDDR5X for AI servers shipped initial mass product to a major customer; SOCAM MP starting Q3; 64/128TB QLC SSD customer approvals obtained; 24Gb GDDR7 leadership claimed.
- Display: LEAD polarizer-free tech (>30% panel power cut) and 500Hz QD-OLED monitor as differentiation; 2H rests on major-customer flagship launches.
Sell-side read-through- BofA (Woo Dong-je) got the tariff update: Korea-U.S. deal reduces uncertainty, but Section 232 results (mid-August, covering semis plus smartphones/tablets/PCs/monitors) are the real swing factor; Samsung lobbying both governments.
- Daiwa (S.K. Kim) extracted the Tesla confirmation ($16.5B, named) plus Taylor specifics: 2025 spend within plan, 2026 CapEx higher, ramp from 2026. Best foundry disclosure of the call.
- JPMorgan (Jay Hyun Kwon) got the ASP detail: conventional DRAM +low-single QoQ vs -mid-single expected, with the legacy price spike from mid-Q2 and fuller ASP reflection in Q3; NAND ASP -low-to-mid-single; and confirmation Q2 one-offs "significantly reduced" in Q3 — not zero.
- KB (Kim Dongwon) pressed M&A and got the full scorecard (Rainbow, Masimo Audio, FläktGroup, Xealth, $120M+ ventures) but no forward deal commitment; Display answer was LEAD/foldable-durability boilerplate.
- NH (Ryu Young-ho) got foldable specifics: Fold7 dimensions/weight, "in line with expectations" initial sales, double-digit annual foldable growth target — a quantified commitment that can be checked.
- Citi (Sei Cheol Lee) got the HBM numbers (bits +30%, 3E high-80s% of volume, >high-90s% of revenue in Q3, 1z mother-die MP approval, HBM4 samples shipped) and hybrid copper bonding status — the most analytically valuable exchange.
- Korea Investment (Minsook Chae) drew out the foundry two-track strategy — a candid structural answer on how Samsung plans to fix node execution.
- SK (Han Dong-hee) got MX 2H profit framing (flagship mix, cost cuts, but tariff/FX/memory-price headwinds acknowledged) and the Exynos 2500/2600 confirmation.
- Goldman (Giuni Lee) closed with the 2H memory outlook and got the call's most important concession: HBM3E supply has grown faster than demand, pricing pressure expected, and the HBM3E-conventional DRAM margin gap will "narrow sharply," requiring a "more balanced product mix."
- Missing challenges: no one asked the size of the memory inventory adjustment or non-memory one-offs; no strike question (fourth straight quarter); no foundry loss magnitude despite Tesla; no one asked whether HBM3E margin compression changes the HBM revenue-recovery math; no segment OP requested; no question on why Q3 NAND bit guide decelerates to mid-single after +high-20s%.
Management credibility- Delivered: the Q1-promised HBM step-up materialized (bits +30% QoQ, 3E mix disclosed for the first time); DRAM/NAND bit guidance met/beat; MX double-digit profitability held again; buyback accelerated (full 10T to complete October 8, ahead of schedule); Exynos 2500 re-entered a flagship as targeted; 2nm reliability milestone repeated as complete; TriFold moved from evasion to confirmed 2025 launch.
- Candor improved: HBM3E oversupply and margin-compression risk disclosed unprompted; DS "operational inefficiencies" acknowledged at CFO level; one-off persistence into Q3 admitted ("significantly reduced," not eliminated); foundry two-track strategy implicitly concedes past execution problems.
- Open items: the Q2 profit decline is blamed on one-offs and FX, but no magnitudes given for the inventory adjustment or export-control costs — the "clean" Q2 run-rate is unknowable from the call; HBM4 MP timing language softened from "MP 2H25" to alignment with demand; foundry loss magnitude still never disclosed; Tesla order terms undisclosed (understandable, but margin/duration unknown).
- Consistency watch: last quarter's "gradual 2H improvement conditional on easing uncertainties" is now a firmer "Q2 bottom, 2H rebound" — a confidence upgrade that the Q3 bit guides (DRAM +high-single, NAND +mid-single) only partially support; the 2H pull-in payback warning from Q1 was not revisited even though Q2 NAND +high-20s% looks partly like the pull-in they flagged.
- Transcript garbles to verify: "third and finance share repurchase" (final), "executing digit buybacks" (the), "expand timing supply" (timely), "1z-nano" (nm), "Bank of Korea (sic) [Bank of America]" operator correction, "Kwon Yun" correction, "SI maturity," "PPAC," "Awesome Intelligence" capitalization, indiscernible DA VP name.
What changed versus the prior quarter- HBM moved from promise to numbers: Q1 gave "samples shipped, gradual contribution"; Q2 gives bits +30% QoQ, 3E high-80s% of volume, >high-90s% of revenue guided for Q3, 1z mother-die MP approval, HBM4 samples out. First mix disclosure after two quarters of refusal.
- New risk introduced: HBM3E industry oversupply and sharp margin convergence with conventional DRAM — the recovery is now quantified but its profitability is explicitly questioned by management itself.
- Memory pricing inflected: conventional DRAM ASP turned positive QoQ (vs guided decline) on the legacy supply squeeze; 2H price increases guided for DRAM and a NAND rally from Q3 — a clear upgrade from Q1's "declines slowing."
- One-offs appeared: inventory value adjustments and China export-control costs hit DS in Q2 — a new earnings-quality issue absent from Q1, with partial persistence into Q3.
- Foundry got its anchor: Tesla $16.5B named order, Taylor ramp 2026, 2026 CapEx guided higher — versus Q1's generic "order intake up QoQ."
- Exynos silence broken: 2500 in the July foldable, 2600 as first 2nm flagship targeting 1H26 — the disclosure gap from Q1 closed.
- TriFold: from "reviewing, no commitment" to confirmed 2025 launch alongside XR.
- M&A moved from "continuously exploring" to four named transactions plus record venture investing.
- FX flipped from +0.5T tailwind to -0.5T headwind; CapEx decline flattened (-0.9T vs -5.8T in Q1).
- Pull-in payback language dropped: Q1's explicit 2H payback warning was not repeated despite Q2's outsized NAND bit growth — an inconsistency worth pressing.
- Strike silence extended to a fourth quarter; Taylor silence broken via the Tesla answer.
Bull case- The HBM recovery is now evidenced, not promised: bits +30% QoQ, 3E mix disclosed and guided to >high-90s% of revenue in Q3, MP approvals accumulating, HBM4 samples at major clients, hybrid copper bonding in customer discussions — every milestone from the last two quarters is converting.
- Memory pricing has inflected: conventional DRAM ASP already positive in Q2 with fuller market-price reflection coming in Q3; 2H DRAM price increases and a NAND rally guided; legacy DRAM/planar NAND tightness favors incumbents; inventories below normal (DRAM) or normalized (NAND) — the setup for a 2H margin recovery is mechanical.
- Q2 is a credible bottom: the profit miss is attributed to one-offs (inventory adjustment, export-control costs) that management says shrink significantly in Q3, plus a 0.5T FX swing; underlying DS revenue grew 11% QoQ.
- Foundry has a real anchor customer: $16.5B Tesla order validates advanced-node competitiveness, supports Taylor utilization from 2026, and management expects follow-on wins; 2nm mobile MP starts 2H.
- Capital return is ahead of schedule: full 10T buyback completing October 8 (vs November 14 target), 6.1T already acquired, cancellations proceeding, dividend ticked up to KRW 367.
- Optionality broadened: TriFold and XR confirmed for 2025, foldables targeting double-digit growth, Exynos 2600 on 2nm for 1H26, SOCAM MP in Q3, QLC high-density SSD approvals, HVAC (FläktGroup/Lennox) and digital health (Xealth) building new legs.
Bear case- HBM profitability is now the question, not volume: management concedes 3E supply exceeds demand industry-wide, pricing pressure is coming, and the 3E-conventional margin gap will "narrow sharply" — Samsung is ramping 3E to >90% of HBM revenue just as its premium erodes.
- Q2 earnings quality is murky: one-off magnitudes undisclosed, so the clean run-rate is unknown; one-offs persist (smaller) into Q3; the "Q2 bottom" call rests on adjustments management won't quantify.
- Q3 bit guides decelerate hard: DRAM +high-single, NAND +mid-single after Q2's +low-teens/+high-20s — consistent with Q1's pull-in payback warning that management conspicuously did not repeat; the 2H demand story may be flattered by borrowed 1H demand.
- Section 232 is a live, dated risk: mid-August decision covering semiconductors and end products (smartphones, tablets, PCs, monitors) could hit DS and DX simultaneously; management is lobbying, which signals real exposure.
- MX momentum is fading into 2H headwinds: shipments 61M→58M, ASP $326→$270 QoQ; 2H smartphone demand guided to contract slightly YoY; memory price rebound raises MX component costs just as tariffs loom; double-digit OPM defense gets harder.
- Foundry remains loss-making with no magnitude disclosed; Tesla revenue is 2026-weighted; China export controls already caused sales restrictions and inventory adjustments in Q2 — a recurring, not one-time, headwind.
- Disclosure gaps persist: no segment OP, no net income discussion, no HBM absolute revenue, four quarters of strike silence, multiple transcript garbles requiring release verification.
Next-quarter watchlist- HBM economics: whether Q3 3E revenue share exceeds the high-90s% as guided; any sign of 3E price cuts or margin compression vs conventional DRAM; HBM4 qualification progress and a re-pinned MP date; hybrid copper bonding conversion to a production commitment; custom HBM4/4E contracts for 2026.
- One-off cleanup: size of Q3 inventory adjustments vs Q2 (management guided "significantly reduced"); whether China export-control costs recur; demand the Q2 one-off magnitudes.
- Memory pricing follow-through: Q3 conventional DRAM ASP capture of the Q2 market-price spike; NAND price rally confirmation; legacy DRAM/planar NAND tightness persistence; server SSD (Gen 5 TLC, 64/128TB QLC) revenue ramp.
- Pull-in payback test: Q3 bit results vs the decelerating guides (DRAM +high-single, NAND +mid-single); any evidence the Q2 NAND surge was preproduction; customer inventory commentary.
- Section 232 outcome: mid-August announcement scope and rates; Samsung's mitigation (production relocation, pricing); impact split between DS and DX.
- Foundry: 2nm gen-1 mobile MP start and yield commentary; Tesla order execution timeline and any additional named wins; Taylor 2026 CapEx magnitude; loss trajectory — press again for a number.
- MX: Fold7/Flip7 sell-through vs the double-digit growth target; TriFold and XR launch execution and pricing; S25 FE early launch; Q3 shipment/ASP recovery vs guide; OPM under rising memory costs.
- System LSI: Exynos 2500 expansion beyond the foldable; 2600 development milestones toward 1H26 flagship entry.
- Capital return: October 8 buyback completion; cancellation timing for the remaining repurchased shares; board posture on further returns post-program; M&A pipeline concreteness after four 2025 deals.
- Corporate disclosure: strike status — press after four quarters of silence; verify garbled figures (buyback tranche splits, dividend, ASP, bit-growth numbers) against the release.
|
| Apr 29, 2025 | +0.26% | Q1 FY2025 | Read transcript briefingQuarter in one view- Q1 FY2025: revenue KRW 79.14T (+4% QoQ), a new all-time quarterly high; OP KRW 6.7T (+0.2T QoQ), OPM 8.4% (flat vs 8.6% prior). SG&A KRW 21.4T (-0.6T QoQ; 27.1% of sales, -2pp). FX tailwind ~+KRW 0.5T OP (vs +0.7T in Q4).
- The mix flipped: DX revenue +28% QoQ (S25 flagship + premium appliances) carried the quarter while DS revenue -17% QoQ on the HBM demand deferral flagged last quarter. The guided Q1 HBM air pocket landed as forecast.
- Memory mechanics: DRAM bit shipments beat guidance by a high-single-digit % (bit growth +low-single-digit QoQ vs -high-single-digit guide — a beat); server DRAM bits +mid-teens QoQ; DRAM ASP -mid-single-digit QoQ. HBM sales down QoQ (export controls + wait-for-enhanced-HBM3E deferral), as guided. NAND bits -~10% QoQ (vs -low-teens guide — slight beat), ASP down "mid-10% level" on server SSD price erosion.
- CapEx KRW 12.0T (-5.8T QoQ; DS 10.9T, Display 0.5T) — memory and foundry both cut sequentially on "flexible" supply-demand alignment; Display down post-8.6G IT OLED completion.
- Capital return: Q1 dividend KRW 365/share (both classes), KRW 2.45T payable late May; first KRW 3T buyback tranche completed and cancelled (2.7T common + 0.3T preferred by mid-Feb); second KRW 3T tranche of the remaining 7T now in progress; KRW 2.5T of phase-2 shares to be cancelled; expanded stock-based compensation announced.
- Transcript garble flags: "KRW 10.9 trillion invested in DS dividend" (division); Foundry section says "fourth quarter" (clearly Q1); CFO outlook swaps DS/DX division references; "risk apical tariffs" (reciprocal); "on-device sale trend" (AI). Verify against the release.
What management is focused on- Tariff/trade-risk framing dominated: CFO opened on "global trade tensions," "elevated uncertainties throughout the year," and scenario-based countermeasures using global production footprint flexibility. Full-year improvement is now explicitly conditioned on "easing current uncertainties" and weighted to 2H — a hedge, not a commitment.
- HBM recovery narrative: Q1 was the bottom; enhanced HBM3E samples already shipped to major accounts, gradual revenue contribution from Q2, "step-by-step recovery with every quarter." HBM4 MP 2H25 on plan; custom HBM4/4E revenue from 2026.
- Pull-in demand acknowledged: management explicitly flagged that tariff-pause preproduction is pulling component demand into Q2 and "may likely have some negative impact on demand in the second half" — unusual candor that tempers the Q2 beat setup.
- Robotics/AI as new-growth pillar repeated (CFO level now, not just Q&A): internal development plus external partnerships, record 2024 R&D spend cited as evidence.
- Foundry loss-narrowing plan: utilization recovery via advanced-node ramp (2nm GAA gen-1 MP starting 2H25, reliability assessment complete) plus mature-node cost competitiveness; HPC/AI order intake up QoQ.
- MX: sustain double-digit profitability via flagship mix (S25 Edge launch), premium ecosystem, and cost/resource efficiency against rising component prices and tariff risk.
- Shareholder value execution: buyback cadence continuing (3T done, next 3T in progress), cancellations confirmed, stock-comp expansion framed as alignment.
Key numbers and quarter mechanics- Consolidated: revenue KRW 79.14T (+4% QoQ, record); OP KRW 6.7T (+0.2T); OPM 8.4%; SG&A KRW 21.4T (27.1% of sales); FX +~KRW 0.5T OP.
- Divisional revenue: DX +28% QoQ; DS -17% QoQ (HBM deferral). No segment OP splits disclosed on the call.
- Memory Q1: DRAM bits +low-single-digit QoQ (beat vs -high-single-digit guide); server DRAM bits +mid-teens QoQ; DRAM ASP -mid-single-digit QoQ. NAND bits -~10% QoQ (slightly above -low-teens guide); NAND ASP down mid-10s% — server SSD prices fell hardest because industry supply concentrated into server SSDs while datacenter projects stayed postponed. HBM sales down QoQ, as pre-flagged.
- Q2 memory guides: DRAM bit growth +low-10s% QoQ; NAND bit growth +mid-teens% QoQ; mobile/PC DRAM pricing to recover first; NAND price declines slowing, some products flat-to-up. Management warns Q2 strength partly reflects pull-in demand.
- MX Q1: 61M smartphones, 7M tablets, ASP $326 (up from $260 on S25 mix); revenue and OP up QoQ; "solid double-digit profitability" maintained, OP improved YoY — the double-digit OPM commitment held in Q1.
- Foundry: utilization down on weak mobile/PC demand from major customers → fixed-cost burden widened losses; HPC/AI order intake up QoQ (2nm, 4nm, sub-5nm); 2nm gen-1 GAA reliability assessment complete, MP 2H25.
- Display: mobile panels up YoY (flagship demand, FX) but down QoQ seasonally; large panels improved QoQ on QD-OLED monitor momentum (double-digit monitor sales growth). Q2 mobile outlook "conservative" on tariff uncertainty.
- VD: Q1 profitability improved QoQ (price increases, material cost cuts) but down YoY (stagnant demand, competition costs).
- CapEx: Q1 KRW 12.0T (DS 10.9T, Display 0.5T), -5.8T QoQ; memory and foundry both reduced sequentially.
- Not disclosed: net income, segment OP, absolute HBM revenue, inventory levels, strike status (third consecutive quarter of silence), Taylor fab.
Product and launch scorecard- Enhanced HBM3E: samples shipped to major customers; revenue contribution from Q2, gradual ramp. This is the make-or-break item — the Q1 gap was blamed on customers waiting for exactly this product, so Q2/Q3 must show the conversion. No mix or absolute-revenue disclosure again.
- HBM4: development on customer timelines, MP 2H25 reiterated; custom HBM4/4E in discussion with multiple clients, revenue from 2026. Consistent with prior commitments.
- DRAM: server strength (bits +mid-teens QoQ); high-density focus (128GB+ DDR5); 1b-nm transition continuing; 10.7Gbps LPDDR5X for on-device AI; LPW/LP-PIM and LPDDR6 in discussion for post-2026.
- NAND: V8 migration accelerating for cost; Gen 5 PCIe server SSD supply expansion planned with demand recovery; Q1 ASP hit hard by server SSD price concentration — the franchise remains pressured.
- Foundry: 2nm gen-1 GAA reliability complete, MP 2H25 (slipped from "MP 2025" language precision — still within year); gen-2 2nm and optimized 4nm design-enablement infrastructure building; Tier-1 HPC/auto order focus. Losses widened; 2H recovery rests on utilization.
- MX: S25 series momentum "maintained predecessor's strong sales"; S25 Edge launching Q2; "Awesome Intelligence" AI features cascading to A-series; new foldables 2H with form-factor-specific AI; XR headset launch planned; Tab S11 2H; new Galaxy Watch.
- Foldables/new form factors: pressed directly on tri-fold (Huawei), management gave no commitment — "reviewing products with new form factors," quality/usability bar first. Foldable shipment growth deceleration acknowledged implicitly, no target given.
- Display: 8.6G IT OLED investment complete; 8.6G/6G lines to address IT/auto; 27" UHD 160ppi QD-OLED monitor launched; ultra-high-refresh gaming monitors coming.
- System LSI: Q1 sales slightly up (image sensors, LSI) despite delayed flagship SoC adoption at key customer; 200MP sensor adoption growing in China; no Exynos 2500 update this quarter — a disclosure gap after last quarter's 2H design-win target.
Sell-side read-through- BofA (Dong Woo) opened on tariffs/supply chain — got the key facts: semiconductors, smartphones, tablets currently exempt from reciprocal tariffs, but product-specific tariff investigations underway; production relocation "considered if necessary." Generic otherwise.
- JPMorgan (Jay Hyun Kwon) extracted the quarter's best memory detail: server DRAM bits +mid-teens, DRAM ASP -mid-single, NAND ASP -mid-10s with the server-SSD supply-concentration explanation, and confirmation the HBM gap matched the January projection.
- Daiwa (Sang Kyu Kim) forced capital-return specificity: KRW 2.5T cancellation of phase-2 shares, remaining 4T under board review; M&A "continuously exploring" — still no deal. Also got MX's profit drivers (flagship mix, component deflation, spec standardization).
- KB (Kim Dongwon) got the HBM recovery cadence (Q1 bottom, quarterly step-up, HBM4 MP 2H25, custom HBM revenue 2026) and pressed TV share loss — answer was premium/AI lineup defense plus 75"+ entry-level response, no share target.
- SK Securities (Han Dong Hee) asked the Foundry loss question directly — management admitted utilization drop and fixed costs as the cause, guided only to "gradual" 2H narrowing. No magnitude given.
- Goldman (Lee Giuni) got the Q2 memory guide plus the pull-in demand warning — the most analytically useful exchange; Display answer was technology/lineup boilerplate.
- Hyundai (Geun Chang Roh) asked the sharpest product question (foldable deceleration, tri-fold vs Huawei) and got a non-answer on tri-fold timing — reads as "not ready, not committed."
- Citi (Lee Sei Cheol) closed on DeepSeek/edge-AI memory demand — answer: workstation AI PCs (128–512GB DRAM) showing first traction; conventional AI PC killer-app still missing; Windows 10 EOL replacement expected to be AI-PC-led; mobile density/power roadmap (LPDDR5X → LPDDR6, LPW/PIM).
- Missing challenges: no one asked whether enhanced-3E sampling implies the original 3E lost the key customer; no quantification of the HBM shortfall or Q1 HBM revenue; no strike question (third straight quarter); no Taylor question (third straight); no Exynos 2500 follow-up; no one pressed the 2H pull-in payback size; no foundry loss magnitude requested.
Management credibility- Delivered: the Q1 HBM decline they pre-announced landed (DS -17% QoQ); DRAM and NAND bit results beat the weak Q1 guides; MX double-digit profitability commitment held in Q1; buyback execution continued (3T completed/cancelled, next 3T in progress, 2.5T cancellation approved); 2nm GAA reliability milestone hit on schedule.
- Candor improved further: pull-in demand risk to 2H disclosed unprompted; Foundry loss drivers (utilization, fixed costs) stated plainly; NAND server-SSD price concentration explained with real mechanics; tariff exemption status and investigation risk disclosed factually.
- Open items: enhanced HBM3E is still samples-to-revenue promise — the entire 2025 HBM recovery thesis now rests on Q2–Q3 conversion; "step-by-step recovery" is unquantified; no 3E mix disclosure for the second straight quarter (the vanished ~50% target remains unaddressed); Foundry loss magnitude never given; tri-fold evasion suggests form-factor response is behind Chinese rivals.
- Consistency watch: last quarter's "recovery from Q2" is now "gradual improvement into 2H assuming easing uncertainties" — the recovery timeline slipped a half; full-year outlook is explicitly conditional, weakening prior-quarter confidence; System LSI went silent on Exynos 2500 after admitting flagship delay.
- Garbles requiring release verification: DS CapEx line ("dividend"), Foundry "fourth quarter" reference, CFO's DS/DX outlook swap, dividend/buyback figures.
What changed versus the prior quarter- HBM: from "Q1 air pocket, optimized 3E MP late Q1, full-swing Q2" to "Q1 bottom confirmed, enhanced 3E samples shipped, gradual contribution from Q2, step-by-step quarterly recovery." The "full swing" language softened to "gradual" — a modest walk-back of the Q2 ramp aggressiveness.
- Memory demand: Q1 bit results beat weak guides on tariff-driven pull-in and faster inventory normalization; but management introduced a new 2H risk — payback from preemptive purchasing. The recovery story now carries a self-identified air-pocket risk in 2H.
- Recovery timeline: pushed from "Q2 recovery" to "2H gradual improvement, conditional on uncertainties easing."
- Capital return: moved from "7T under review" to active execution — second 3T tranche in progress, 2.5T cancellation approved, stock-comp expansion added. The most concrete shareholder-value quarter to date.
- CapEx turned down: Q1 12.0T vs 17.8T, with memory and foundry both cut — supply discipline is now showing up in spending, not just words.
- Tariffs became the dominant macro variable: exemption status disclosed, product-specific investigations flagged, production relocation explicitly on the table — new versus last quarter's generic monitoring language.
- New disclosures: server DRAM bit growth (+mid-teens), NAND ASP magnitude (mid-10s decline), AI-PC workstation memory traction (128–512GB), 2nm reliability completion, Q1 MX ASP $326.
- Silences extended: strike and Taylor now three quarters unmentioned; Exynos 2500 dropped from the script after one quarter of candor.
Bull case- The worst is confirmed behind: Q1 HBM bottom called and delivered, enhanced 3E sampling major customers with Q2 revenue starting, quarterly step-up guided, HBM4 MP 2H25 and custom HBM revenue 2026 — the full stack of recovery milestones is intact and on schedule.
- Q1 proved demand resilience: record revenue, DRAM/NAND bit beats, server DRAM +mid-teens, MX double-digit margin held, OP up sequentially despite the DS decline — the non-HBM businesses are carrying earnings.
- Q2 guides are strong: DRAM bits +low-10s%, NAND +mid-teens%, mobile/PC pricing recovering first, NAND price declines ending — if even partly pull-in-free, memory earnings inflect in Q2.
- Supply discipline is now mechanical: CapEx cut 5.8T QoQ, flexible memory investment, V8/1b-nm migration bit loss — supports the 2H NAND tightening and DRAM pricing recovery thesis.
- Capital return is compounding: 3T cancelled, next 3T buying, 2.5T more cancellation approved, 4T still under review, plus stock-comp alignment — tangible per-share value support with remaining catalysts.
- Edge-AI optionality: AI-PC workstation memory demand (128–512GB) emerging, Windows 10 EOL replacement cycle, mobile density upgrades — a second memory demand leg beyond AI servers.
Bear case- The Q2 strength is partly borrowed: management itself flagged pull-in demand from the tariff pause and warned of 2H payback — the strong Q2 bit guides may flatter the underlying trend, and the 2H recovery is explicitly conditional on trade uncertainties easing.
- HBM credibility is still unproven: "gradual contribution" and "step-by-step recovery" replaced "full swing"; no mix, no absolute revenue, no named customer; export controls remain a live headwind; the qualification question at the key customer remains unanswered for a second quarter.
- NAND economics are deteriorating: ASP down mid-10s in Q1 with server SSD prices falling hardest due to industry supply concentration — the highest-value NAND segment is the weakest, and recovery depends on datacenter projects resuming.
- Foundry losses widened with only "gradual" 2H narrowing promised, no magnitude; 2nm MP is 2H-weighted, leaving 1H utilization weak; mobile/PC end-demand from major customers still soft.
- Tariff escalation risk is unresolved: current exemptions cover semis/smartphones/tablets, but product-specific investigations are underway — a negative outcome hits both DS and DX simultaneously, and component-cost inflation already pressures MX margins.
- Form-factor competitiveness: no tri-fold answer while Huawei ships; foldable growth decelerating; XR unproven. MX's premium defense relies on S25 Edge and incremental foldables.
- Disclosure quality: three quarters of silence on strike and Taylor; segment OP, net income, HBM revenue all undisclosed; multiple transcript garbles require verification before modeling.
Next-quarter watchlist- HBM conversion: enhanced-3E revenue contribution in Q2 vs "gradual" guidance; any mix/absolute disclosure; whether the quarterly step-up materializes in Q3; HBM4 MP milestones for 2H25; custom HBM4/4E contract announcements; export-control developments.
- Pull-in payback: evidence of how much Q2 bit strength (+low-10s% DRAM, +mid-teens% NAND guides) is preproduction vs end-demand; 2H order patterns as tariff deadlines hit; customer inventory levels post-normalization.
- Memory pricing: mobile/PC DRAM price recovery in Q2 as guided; whether NAND prices actually flatten/turn up; server SSD price stabilization and datacenter project resumption.
- Foundry: 2nm GAA gen-1 MP start in 2H; order-intake conversion to revenue; utilization and loss trajectory — demand a magnitude; any Tier-1 HPC/auto wins named.
- MX: S25 Edge launch reception; Q2 shipment/ASP decline vs guide; double-digit OPM under rising component costs; new foldables' competitiveness vs Chinese tri-folds; XR headset launch execution.
- System LSI: Exynos 2500 design-win evidence for 2H flagships — back on the checklist after this quarter's silence.
- Capital return: completion of the second 3T tranche; board decision on the remaining 4T (timing, scale, cancellation); any M&A concreteness; whether buyback counts inside the 50% FCF policy.
- Corporate disclosure: strike status and Taylor — press after three quarters of silence; verify all garbled figures (DS CapEx, Foundry quarter reference, DS/DX outlook lines, buyback/cancellation amounts) against the release.
|
| Jan 30, 2025 | -4.07% | Q4 FY2024 | Read transcript briefingQuarter in one view- Q4 FY2024: revenue KRW 75.8T (-4% QoQ); SG&A KRW 22.0T (+1.2T QoQ, 29.1% of sales, +2.7pp — R&D-heavy); operating profit KRW 6.5T (-2.7T QoQ), OPM 8.6% (-3.0pp). FX flipped back to a ~+KRW 0.7T tailwind (strong USD offset by other currencies). FY2024 revenue KRW 300.9T, second-highest ever (behind 2022).
- Transcript garble flag: CFO says "DS Division revenue increased 3%" then "DS Division revenue decreased 10%... due to competition and pending smartphone launch" — the second is clearly DX. Verify segment splits against the release.
- Memory: record Q4 revenue on DRAM ASP strength, but OP down slightly QoQ (R&D + advanced-node ramp costs). DRAM bits down low-teens QoQ, NAND bits down low-single (transcript garbled — verify); DRAM ASP "rose around 20% QoQ" (likely garbled — verify; aging-stock sales limited ASP); NAND ASP down mid-single-digit. HBM sales 1.9x QoQ, "slightly below initial forecast."
- The HBM story shifted again: Q4 3E sales exceeded HBM3 and supply broadened to multiple GPU/datacenter clients, but management disclosed a new Q1 air pocket — export controls plus customers deferring orders to wait for the "optimized" HBM3E (MP late Q1, full-swing Q2). 2025 HBM bit supply 2x 2024 was re-affirmed.
- Q1 guide is weak across the board: DRAM bits -high-single-digit, NAND bits -low-teens, conventional DRAM/server DDR5/SSD prices falling, HBM sales down QoQ, Foundry utilization weak, Display conservative. Recovery framed as starting Q2.
- Capital return: dividend KRW 363/364 (common/preferred); FY2024 FCF KRW 19.9T, 50% pool ~KRW 9.9T vs KRW 9.8T dividends — explicitly "no meaningful surplus" for early returns. KRW 10T buyback: first KRW 3T tranche 89.3% executed; remaining KRW 7T timing/method "under review."
What management is focused on- New CFO Sooncheol Park opened with an unusually direct acknowledgment of "challenges" and "current difficulties," promising swift resolution and better IR communication — a tone reset consistent with the buyback and value-up pressure.
- HBM damage control, round two: the narrative is now "optimization-driven pause" — enhanced HBM3E MP late Q1, ramp Q2, 8→12-high transition accelerating, 16-high samples already shipped for validation, HBM4 (1c-nm) MP 2H25 on plan, custom HBM4/4E talks ongoing.
- Portfolio purification hard deadline quantified: DDR4/LPDDR4 at low-30% of 2024 sales, to be cut "dramatically" to single-digit share in 2025; supply concentrated to major mobile/consumer/auto customers only. This is the clearest answer yet on the China legacy-supply threat.
- 2025 supply discipline reiterated and mechanized: bit growth "significantly limited" for both DRAM and NAND; V6→V8 double-stack conversion causes larger equipment-replacement bit loss; management explicitly accepts short-term bit loss for node migration.
- Memory demand recovery call: mobile/PC inventory adjustment ending after Q1, demand recovery from Q2; NAND supply-demand tightening by early 2H25 at the latest (three years of conservative industry CapEx plus recent production cuts).
- Foundry: 2nm gen-1 MP 2025, gen-2 PDK 1H25/MP 2026; 4nm in HPC mass production on "stable yields"; mature-node profitability and specialty (mmWave/RF) commercialization as the near-term lever; admits slower recovery at "some major customers" limits 1H improvement.
- MX: S25 launched (Jan 23, San Jose) as the AI-led flagship cycle; explicit commitment to "secure a double-digit operating margin" despite component-cost inflation; double-digit flagship revenue growth targeted for 2025; foldable lineup diversification "later this year"; XR device showcased at Unpacked.
- New growth narrative: robotics — Rainbow Robotics consolidated as a subsidiary, Future Robotics Office under direct CEO oversight (Prof. Jun-Ho Oh), humanoid development ambition stated. First time robotics is an earnings-call topic.
Key numbers and quarter mechanics- Consolidated: revenue KRW 75.8T (-4% QoQ); SG&A KRW 22.0T (+1.2T; 29.1% of sales, +2.7pp); OP KRW 6.5T (-2.7T); OPM 8.6% (-3.0pp). FX ~+KRW 0.7T OP. FY2024 revenue KRW 300.9T.
- Memory Q4: record quarterly revenue; OP down slightly QoQ (R&D, next-gen R&D facilities, advanced-node ramp costs). DRAM bits down low-teens QoQ (vs -mid-single-digit guide — a miss); NAND bits down low-single (vs +low-single guide — a miss); DRAM ASP "+~20% QoQ" (garble risk — verify; aging-stock sales capped ASP); NAND ASP -mid-single-digit. HBM revenue 1.9x QoQ, below internal forecast; ≥128GB DDR5 module sales 2.5x QoQ.
- Q1 memory guides: DRAM bits -high-single-digit QoQ; NAND bits -low-teens QoQ; DRAM ASP and performance "slight decline"; conventional, server DDR5, and SSD market prices all falling in Q1; HBM sales down QoQ (export controls + wait-for-optimized-3E deferral).
- MX Q4: 52M smartphones (-6M QoQ), 7M tablets (flat), ASP $260 (down from $295) — the guided post-launch decline landed; sales and profit down QoQ. FY flagship sales grew double-digit (S24). Q1: shipments and ASP up QoQ on S25; tablets flat.
- System LSI: Q4 earnings declined (mobile weakness, R&D); W1000 wearable SoC on 3nm in MP, CES 2025 innovation award. Admits "delayed entry into the flagship SoC market" — Exynos 2500 now targets design wins for 2H flagship models, implying S25 is not Exynos 2500-based.
- Foundry: Q4 sales up slightly (a few strong US customers) but profit down on low utilization and advanced-node R&D. Q1: market contracts seasonally, earnings weak on fixed-cost burden. 2025: sub-5nm market double-digit growth; YoY revenue growth targeted via advanced-node mix.
- Display: mobile profit down QoQ (weak smartphones, competition); IT/Auto sales +double-digit QoQ; large panels revenue +double-digit QoQ, overall results "improved slightly." Q1 conservative.
- VD: Q4 sales up, mix improved (Neo QLED/OLED/75"+), but profitability down slightly on competition-driven costs. TV market up slightly YoY, entry-led.
- CapEx: Q4 KRW 17.8T (+5.4T QoQ; DS 16.0T, Display 1.0T). FY2024 KRW 53.6T (DS 46.3T, Display 4.8T) vs 53.1T in 2023 — below the ~56.7T projection given last quarter. 2025 memory CapEx "similar" again; Foundry CapEx cut in 2024.
- Shareholder returns: FY2024 FCF KRW 19.9T; 50% pool ~9.9T ≈ 9.8T dividends → no surplus. Buyback: KRW 3T tranche 89.3% complete (as of "January 2024" — clearly 2025, transcript error); KRW 7T remainder under review, including whether it counts toward the 50% FCF policy.
- Not disclosed: net income, segment OP splits, absolute HBM revenue, inventory levels, strike status (second consecutive quarter of silence).
Product and launch scorecard- HBM3E: mixed quarter. Delivered: 8/12-high in MP with 3E sales exceeding HBM3 in Q4; supply expanded to multiple GPU providers and datacenter clients. Missed: total HBM 1.9x QoQ "slightly below initial forecast"; no 3E mix disclosure vs the ~50% Q4 target — the target simply vanished, replaced by a new Q1 demand gap (export controls + customers waiting for optimized 3E). Optimized 3E: MP and first supply late Q1, "full swing" Q2 — consistent with the prior 1H25 enhanced-3E commitment but now with a disclosed air pocket first.
- HBM roadmap: 16-high 3E samples already shipped to major customers for validation (ahead of any commercial demand — a credibility-positive milestone); HBM4 on 1c-nm, MP 2H25 reiterated; custom HBM4/4E discussions "as originally planned." 2025 HBM bit supply 2x 2024 re-affirmed after going unmentioned last quarter.
- DRAM: 256GB module on 32Gb 1b-nm DDR5 enters MP/sales in Q1 (new); ≥128GB DDR5 modules 2.5x QoQ; 1b-nm conversion accelerating; LPDDR5X expanding to PC/server.
- NAND: V7 QLC server SSD sales expanding; V8/V9 conversion accelerating; server SSD demand hit by datacenter project delays (GPU supply constraints) — the prior +~10% QoQ server SSD guide appears missed (shipments "limited vs initial expectations"). No 64TB/128TB QLC update this quarter — a disclosure gap.
- Foundry: 2nm gen-1 PDK out (1H24), MP 2025; gen-2 PDK 1H25, MP 2026 (new granularity); PPA/MPW with major customers, product-level design started "for certain clients"; Tier-1 order talks; 4nm HPC in MP on stable yields. Second-gen 3nm wearable: W1000 confirmed in MP via System LSI — closes last quarter's open item.
- Exynos 2500: still not landed — "delayed entry into the flagship SoC market" admitted; targeting 2H flagship design wins via optimization and Foundry collaboration. Effectively confirms S25 runs without it.
- MX: S25 launched Jan 23 (One UI 7.0, multimodal Galaxy AI, Knox Vault personal data engine, enhanced NPU/DRAM bandwidth); XR device showcased at Unpacked; foldable lineup diversification promised "later this year" (the teased new form factor remains undated); Galaxy Trade-In program new.
- Display: 8.6G IT OLED line "being prepared as planned"; 160ppi UHD and 500Hz gaming monitors, B2C monitor entry in 2025; new brighter QD-OLED TV ("security OLED TV" is a garble).
- VD: "Vision AI" branding for AI screens; Super Big TV ("above nine inches" — likely 98"+) push; The Frame with Neo QLED and wireless; TV Plus ad expansion.
- Robotics: Rainbow Robotics consolidated; CEO-level Future Robotics Office; humanoid ambition — strategy statement only, no financials.
Sell-side read-through- KB (Dongwon Kim) opened on shareholder returns — the buyback's remaining KRW 7T, its relationship to the 50% FCF policy, and value-up status. Answers were non-committal ("under review," "shortest possible timeline"), and the CFO disclosed the uncomfortable math: no FCF surplus in 2024 for early returns. The memory follow-up extracted the Q4 bit/ASP mechanics and the HBM 1.9x/below-forecast admission.
- SK Securities (Han) asked the HBM question plus DeepSeek. The answer produced the quarter's biggest disclosure: the Q1 HBM demand gap (export controls + customers deferring to wait for optimized 3E), late-Q1 MP/Q2 full-swing timing, faster 8→12-high transition, 2x 2025 bit supply, 16-high samples shipped. DeepSeek answer was a hedge — "long-term opportunities and short-term risk," too early to call.
- Citi (Lee) pressed Trump-policy risk (generic monitoring/response answer, no specifics) and got the full Q1 memory guide: DRAM bits -high-single-digit, NAND -low-teens, price declines now extending to server DDR5 and SSDs — a notable deterioration from "server pricing relatively strong."
- UBS (Gaudois) asked only Foundry 2nm — got gen-1/gen-2 split, PDK/MP timing, and "product-level design started for certain clients," the most concrete customer-engagement language yet.
- Goldman (Lee) extracted the 2025 memory framework (Q2 recovery, bit-loss acceptance, NAND tightening by early 2H) and MX's AI answer (NPU/DRAM bandwidth, Knox Vault, agent vision).
- Daiwa (S.K. Kim) forced the China-legacy answer: DDR4/LPDDR4 from low-30% of 2024 sales to single-digit in 2025 — the most decisive portfolio-purge commitment to date.
- Korea Investment (Chae) got MX's 2025 framing: premium growth decelerating, mass-market AI proliferation, double-digit flagship revenue growth target, no pin-to-pin price competition; VD's answer was differentiation + efficiency boilerplate.
- BofA (Woo) asked robotics — a soft closer that let management unveil the Rainbow Robotics consolidation and humanoid office.
- Missing challenges: no one asked why the ~50% Q4 3E mix target went undisclosed; no one quantified the HBM shortfall vs forecast; no strike question (second straight quarter); no Taylor fab question; no one pressed the DRAM ASP "+20%" figure or the bit-guide misses; no one asked whether the optimized-3E deferral implies the original 3E failed qualification at the key customer; no MX margin miss question despite the prior "close to double-digits" softness.
Management credibility- Delivered: S25 launched on schedule (1H25 commitment); Q4 MX shipment/ASP decline landed as guided; 2nm PDK in customer hands with MP 2025 reiterated; W1000 3nm wearable MP confirmed; buyback execution tangible (89.3% of first KRW 3T); FY CapEx landed below the guided 56.7T at 53.6T; 2025 HBM 2x bit supply re-affirmed.
- Missed/slipped: DRAM bits (-low-teens vs -mid-single-digit guide) and NAND bits (-low-single vs +low-single guide) both missed; HBM 1.9x below internal forecast; the ~50% Q4 3E mix target was neither confirmed nor denied — simply not disclosed, which reads as a miss; server SSD momentum broke on datacenter project delays; Exynos 2500 flagship entry explicitly delayed (now 2H design-win target); a new Q1 HBM air pocket appeared one quarter after "sales expansion starting Q4."
- Candor improved: CFO opened with direct acknowledgment of difficulties; the Q1 HBM gap and its causes (export controls, customer deferral to optimized product) were disclosed proactively in Q&A; "delayed entry into the flagship SoC market" is unusually plain language; the no-surplus-FCF math was stated rather than spun.
- Soft spots: the optimized-3E story is doing double duty — it explains both the Q4 forecast miss and the Q1 gap, and management did not address whether it signals a qualification problem at the major customer; "slightly below initial forecast" is unquantified; value-up program remains placeholder; Trump-policy and DeepSeek answers were content-free by design; multiple transcript garbles (DS/DX revenue lines, DRAM ASP +20%, "January 2024" buyback date, "security OLED TV," "product obsession plans") require release verification.
- Consistency watch: last quarter's "inventory normalization by end of year" was not explicitly confirmed as completed — aging-stock sales continued in Q4; the 2028 Foundry targets remain absent; "double-digit operating margin" for MX is now an explicit 2025 commitment (upgrade from "close to double-digits") — hold them to it.
What changed versus the prior quarter- HBM narrative reset again: from "qualification phase complete, Q4 expansion, ~50% 3E mix" to "1.9x but below forecast, 3E > HBM3, and a new Q1 demand gap from export controls plus customers waiting for optimized 3E (MP late Q1, full swing Q2)." The mix target disappeared from disclosure; the 2x 2025 supply commitment returned.
- Memory momentum broke harder: Q3's "record revenue, OP down slightly" became Q4 record revenue with bit guides missed on both DRAM and NAND, and Q1 guided to bit declines on both plus price declines spreading to server DDR5/SSDs — the server pricing exception is gone.
- Recovery timeline pushed out: Q4's "improvement in DS" became Q1 "limited improvement, semiconductor weakness," with recovery now a Q2 story and NAND tightening an early-2H story.
- China response hardened: from "scaling down legacy" to a quantified exit — DDR4/LPDDR4 low-30% of 2024 sales → single-digit in 2025.
- Exynos reality admitted: from "close collaboration to secure supply, targeting flagship adoption next year" to "delayed entry into the flagship SoC market," 2H design-win target — S25 implication is clear though unstated.
- Capital return moved from placeholder to partial delivery: KRW 10T buyback announced in November, first 3T nearly done, but the FCF math (no 2024 surplus) and the uncommitted 7T temper it; value-up program still undetailed.
- New items: new CFO on the call; robotics strategy (Rainbow Robotics consolidation, CEO-level humanoid office); DeepSeek and Trump-policy risk entered the Q&A; 2nm gen-2 timeline (PDK 1H25, MP 2026) disclosed; MX double-digit OPM explicitly committed for 2025.
- Strike and Taylor: absent for a second consecutive quarter — now a pattern, not an omission.
Bull case- The HBM ramp is delayed, not lost: 3E outsold HBM3 in Q4, supply broadened to multiple GPU/datacenter customers, 16-high samples already in customer hands, optimized 3E MP late Q1 with full-swing Q2, HBM4 on 1c-nm for 2H25, and 2025 bit supply still guided to 2x 2024. If the Q2 ramp lands, the Q1 gap is a timing issue.
- The legacy exit is now quantified and aggressive: cutting DDR4/LPDDR4 to single-digit share in 2025 insulates Samsung from Chinese oversupply and concentrates it where pricing is strongest — HBM, DDR5 (128/256GB modules 2.5x QoQ), LPDDR5X, GDDR7, QLC SSDs.
- Supply-side setup for 2H25 is favorable: three years of conservative industry NAND CapEx plus new production cuts, Samsung's own bit growth "significantly limited" by conversion loss, and mobile/PC inventory adjustment ending after Q1 — management's Q2 recovery/2H NAND tightening call is internally consistent.
- Capital return is real and accelerating: KRW 10T buyback with 3T nearly executed and cancelled, on top of KRW 9.8T dividends — and the remaining 7T plus value-up details are pending catalysts.
- Optionality broadened: 2nm gen-1 MP 2025 with product-level customer designs underway, 4nm HPC in stable-yield MP, S25 AI cycle with a double-digit OPM commitment, XR showcased, and a consolidated robotics/humanoid platform under CEO oversight.
Bear case- HBM execution risk keeps compounding: three straight quarters of slippage (MP timing → mix/qualification → forecast miss plus a self-inflicted Q1 demand gap). The "customers waiting for optimized 3E" framing invites the question management didn't answer: did the original 3E fail to hold the key customer? The ~50% mix target's disappearance from disclosure suggests it was missed.
- Q1 is guided down everywhere at once: DRAM bits -high-single-digit, NAND -low-teens, prices falling even for server DDR5/SSDs, HBM sales down QoQ, Foundry utilization weak, Display conservative — with OP already at 8.6% margin and SG&A at 29.1% of sales, 1H25 earnings risk is substantial.
- Bit guides were missed on both DRAM and NAND in Q4, and the server SSD franchise — last quarter's bright spot — is now hit by datacenter project delays tied to GPU supply, outside Samsung's control.
- Export controls are now a direct, disclosed HBM demand headwind with "unexpectedly high" volatility — a structural risk to the 2x 2025 supply plan that management cannot hedge.
- Exynos 2500's delayed flagship entry confirms System LSI's flagship SoC gap persists into 2H25 at earliest; Foundry profit fell on utilization with only "a few US customers" carrying sales; 2028 targets remain buried; Taylor unmentioned.
- Capital return has a ceiling: 2024 FCF left no surplus above the 9.8T dividend, the remaining 7T buyback is uncommitted in timing and structure, and value-up is still words.
- Strike silence for two quarters; transcript garbles (DS/DX lines, DRAM ASP +20%, dates) require verification before modeling.
Next-quarter watchlist- HBM proof points: optimized-3E MP in late Q1 and the promised Q2 full-swing ramp; whether the Q1 demand gap closes as guided; any disclosure of 3E mix (the vanished ~50% target); 16-high validation progress; HBM4 milestones toward 2H25 MP; whether 2x 2025 bit supply survives export-control volatility; any named-customer or absolute-revenue disclosure.
- Memory mechanics: Q1 DRAM bits vs -high-single-digit guide, NAND vs -low-teens; whether server DDR5/SSD price declines materialize as feared; confirmation (finally) that inventory normalization completed at year-end; aging-stock drag on ASPs; R&D/ramp cost trajectory vs the "OP down slightly" run-rate.
- Q2 recovery call: evidence that mobile/PC inventory adjustment ended after Q1; datacenter project resumption as GPU supply improves; NAND supply-demand tightening signals toward early 2H.
- Legacy exit: DDR4/LPDDR4 share trajectory toward single-digit in 2025; any margin benefit visible by 2H.
- Foundry: 2nm gen-1 MP execution in 2025 and gen-2 PDK in 1H25; conversion of "product-level design for certain clients" into named order wins; utilization and loss trajectory; Taylor — press again; whether 2028 targets ever return.
- System LSI: Exynos 2500 design-win evidence for 2H flagships; 200MP 0.5µm sensor contract; whether the earnings downturn persists as guided.
- MX: S25 early sell-through and the double-digit OPM commitment under component-cost inflation; foldable lineup diversification timing (the teased new form factor); XR launch cadence; double-digit flagship revenue growth tracking.
- Corporate: remaining KRW 7T buyback structure/timing and whether it counts inside the 50% FCF policy; value-up program specifics — now two quarters pending; strike status — demand disclosure after two quarters of silence; verification of all garbled figures (Q4 bit growth, DRAM ASP, DS/DX revenue lines, buyback completion date) against the release.
|
| Oct 30, 2024 | -0.23% | Q3 FY2024 | Read transcript briefingQuarter in one view- Q3 FY2024: revenue KRW 79.1T (+7% QoQ); gross profit KRW 30.0T (up only slightly QoQ); GM 37.9% (-2.0pp QoQ) on sharply lower inventory-valuation reversals and aging-inventory depletion; SG&A KRW 20.8T (+1.5T, 26.3% of sales, R&D at a new high again); operating profit KRW 9.2T (-KRW 1.3T QoQ), OPM 11.6% (-2.5pp). FX flipped to a ~KRW 0.5T OP headwind (won strength) vs +0.8T tailwind in Q2.
- The miss is explicitly acknowledged: management states DS one-time costs (incentive provisioning) were "more significant than the gap between operating profit and consensus estimates." Memory profit fell QoQ despite revenue growth — lower write-backs, incentive provisioning, weak USD.
- Divisional revenue: MX +13% QoQ (new model launches), DS +3%, Display +5%. Q4 guide: semis improve, finished goods "lackluster," DX sales to decline slightly QoQ.
- The headline event: HBM3E commercialization delay at a major customer, pre-announced with the prelims. Q3 HBM3E mix came in at low-to-mid-teens % of HBM sales — below the mid-teens Q3 guide — and the Q4 target was cut from high-60s % to ~50%. Management claims "an important phase in the qualification process" is now complete, with sales expansion starting Q4.
- CapEx: Q3 KRW ~12.4T (transcribed "1-2.4 trillion" — verify; DS 10.7T, Display 1.0T). First-ever FY guide on this call: 2024 CapEx ~KRW 56.7T (+3.6T YoY; DS 47.9T down slightly, Display 5.6T up). 2025 memory CapEx "similar to this year"; Foundry CapEx declining this year and capacity expansion to be "thoughtful" next year.
- Dividend KRW 361/share (~KRW 2.45T), payable end of November; FY2024 total KRW 9.8T. A "value-up program" review was acknowledged with no details — a new shareholder-return thread.
- Transcript-quality flags: numerous garbles ("TSG SSDs," "QH" for QLC, "V8 based TCI Gen 5," "HBM33E," "NOx Matrix" for Knox, "BSPTN" likely BSPDN, "tonometer products," "KRW 30.3 billion" DS YTD CapEx, "luggage products" for legacy, "vehicle purchasing demand" likely "weak"). Verify all figures against the release.
What management is focused on- Damage control on HBM: the prepared remarks and Q&A are built around explaining the HBM3E delay while defending the ramp — qualification "important phase" complete, Q4 sales expansion, enhanced HBM3E variants for next-gen GPU programs with MP in 1H25, HBM4 MP still targeted 2H25, custom HBM talks with multiple clients.
- Portfolio purification in memory: depleting aging/legacy inventory (which depressed Q3 ASPs and margins), "flexibly rolling down" legacy DRAM/NAND production, accelerating 1b-nm DRAM and V8/V9 NAND conversion, and completing inventory normalization "by the end of the year."
- Explicit 2025 supply discipline: bit growth constrained by conversion bit-loss; "production de-growth to be limited" for both DRAM and NAND; focus on high-margin mix over bit share; CapEx flat-ish with priority to HBM packaging, R&D complex, and clean rooms — not capacity expansion.
- Foundry pivot to 2nm execution: 2nm GAA PDK released, MP target 2025 reiterated, PPA/MPW evaluation work with customers, advanced packaging (2.3D/2.5D/3D) and strategic partnerships as the customer-acquisition lever; Memory+Foundry collaboration on an HBM prototype (base die) to win AI/HPC customers. Notably, custom-HBM base-die foundry choice will follow customer preference "regardless of whether it involves an internal or external partner" — a striking flexibility statement.
- MX: profitability "close to double-digits" in Q3; Q4 shipments/ASP guided down QoQ post-launch; 2025 strategy is Galaxy AI-led flagship mix (S25 in 1H25), foldable differentiation (slimmer/lighter Fold, new form factor "being prepared"), and cost defense against rising component prices.
- AI-everywhere narrative: a full CX/MDE segment on SmartThings (360M users), Knox security, Bixby, Family Care, SmartThings Pro B2B — positioning the 500M-device installed base as the AI moat.
- Value-up program: management acknowledged "significant interest" and says a thorough review of options is underway, details to come — the first time capital-return enhancement is on the agenda.
Key numbers and quarter mechanics- Consolidated: revenue KRW 79.1T (+7% QoQ); GP KRW 30.0T; GM 37.9% (-2.0pp); SG&A KRW 20.8T (+1.5T; 26.3% of sales); OP KRW 9.2T (-1.3T); OPM 11.6% (-2.5pp). FX ~-KRW 0.5T OP.
- Memory Q3 actuals: DRAM and NAND ASPs both +high-single-digit QoQ (below internal budget due to aging-inventory sales); DRAM bits flat QoQ (missed the low-single-digit growth guide); NAND bits down single-digit (below bit guidance, blamed on mobile inventory adjustment and profitability focus); memory revenue +mid-single-digit in USD. HBM revenue +>70% QoQ (vs "doubling" trajectory); server DDR5 +mid-teens; server SSD +mid-30s and a record, now ~50% of NAND sales; DDR5 >80% of server DRAM sales; 128GB 1b-nm module now high-single-digit % of server DDR5 sales.
- HBM mix: HBM3E at low-to-mid-teens % of HBM sales in Q3 (below the mid-teens guide); Q4 target ~50% (cut from high-60s %). Both 8-high and 12-high HBM3E now in MP and generating sales.
- Q4 memory guides: DRAM bits to decline to mid-single-digit negative (HBM crowding out DDR5/LPDDR5X, Chinese mobile legacy inventory adjustment, base effect); NAND bits ~low-single-digit growth; server SSD revenue +~10% QoQ on a bit basis with QLC upside.
- MX: 58M smartphones (+4M QoQ), 7M tablets, ASP USD 295 (up from 279); revenue and OP both up QoQ; profitability "close to double-digits." Q4: shipments and ASP down QoQ; full-year flagship sales targeted at double-digit growth.
- System LSI: Q3 sales up (inventory-reduction push) but earnings down on one-time incentive costs; SoC up on flagship design-ins, sensors down on customer inventory, DDI up on new projects. 2025 smartphone market growth <1% (vs 2.5% in 2024).
- Foundry: Q3 earnings declined QoQ on incentive provisioning and engineering-run wafer inventory provisions; order targets met, particularly 5nm; 2nm PDK released. 2025 foundry market seen growing double-digit.
- Display: mobile >90% of sales, double-digit smartphone-panel sales growth, profit up QoQ; large panels volume up but earnings down QoQ. Q4 outlook "quite conservative" on competition.
- VD: profitability improved both YoY and QoQ (Neo QLED/OLED/Big TV mix, service business) despite TV market down slightly YoY.
- CapEx: Q3 ~KRW 12.4T (DS 10.7T, Display 1.0T); 9M24 KRW 35.8T (-0.9T YoY); FY2024 projection KRW 56.7T (DS 47.9T, Display 5.6T); 2025 memory CapEx similar to 2024; Foundry CapEx down in 2024, expansion cautious in 2025.
- Not disclosed: net income, segment OP splits, absolute HBM revenue, strike status (not mentioned at all this quarter), inventory levels in absolute terms.
Product and launch scorecard- HBM3E: the quarter's defining slip. 8-high MP did start in Q3 (delivering the rescheduled commitment), and both 8/12-high are in MP with sales — but a major customer's commercialization was delayed, 3E mix missed guide (low-to-mid-teens vs mid-teens), and the Q4 mix target was slashed from high-60s % to ~50%. Mitigants claimed: key qualification phase complete, Q4 sales expansion, multi-customer project broadening, enhanced 3E for next-gen GPUs with 1H25 MP. The 2H "3.5x 1H" HBM revenue framing from last quarter was not repeated — treat as abandoned or at risk; only +>70% QoQ was disclosed.
- HBM4: development "according to plan," MP 2H25 reiterated. Custom HBM: multiple clients, and base-die foundry selection explicitly customer-driven (internal or external) — a notable strategic concession.
- Server SSD/QLC: record server SSD revenue (~50% of NAND sales); 64TB QLC MP this quarter (delivers the 2H MP commitment); 128TB sampling this quarter with full lineup MP up to 128TB by 1H25 (128TB slipped from "Q4 lineup addition" to sampling in Q4, MP by 1H25); V8-based PCIe Gen5 expansion; V6→V8 migration accelerated. The "4x YoY server SSD" full-year guide was not re-stated.
- DRAM: 1b-nm 32Gb DDR5 128GB module now high-single-digit % of server DDR5 sales (scaling as promised); 10.7Gbps LPDDR5X and GDDR7 moving to "mass volume business."
- Foundry nodes: 2nm GAA PDK released (delivered), MP 2025 reiterated; second-gen 3nm wearable MP was not explicitly confirmed as started — silence worth pressing; backside PDN ("BSPTN," likely BSPDN) and automotive cited as 2nm-era differentiators; 2028 targets not repeated.
- Exynos: next-gen flagship SoC (Exynos 2500, unnamed) — "close collaboration with Foundry to secure sufficient supply," targeting flagship-series adoption next year; current-gen Exynos 2400 supply expanding. No yield or customer specifics; customer details refused.
- MX: Fold6/Flip6/S24 driving Q3; a new foldable form factor teased ("preparing... when quality meets expectations"); S25 confirmed for 1H25 with "more complete Galaxy AI"; XR devices referenced as upcoming ecosystem products.
- Display/VD: QD-OLED monitor lineup broadening (new high-res, diverse refresh rates) and B2C monitor entry in 2025; VD's "nine inches above Super Big TV" (likely 98"+) push; TV Plus ad platform expansion reiterated.
- Absences: strike never mentioned; Taylor fab never mentioned; Harman, Networks, DA/BESPOKE absent; 8.6G IT OLED not discussed beyond generic "IT/auto growth."
Sell-side read-through- UBS (Gaudois) asked the question that mattered: the HBM3E delay at a major customer. The answer produced the quarter's key disclosures — qualification phase complete, Q4 expansion, ~50% Q4 3E mix (a cut, delivered without acknowledging it as a cut), enhanced 3E for next-gen GPUs in 1H25, and the custom-HBM foundry-flexibility statement.
- Daiwa pressed the Q3 memory miss directly ("below market expectations") and got the fullest mechanics: flat DRAM bits, NAND below bit guide, ASPs below budget on aging-inventory sales, server SSD at 50% of NAND, 128GB module mix. The Exynos follow-up got only a supply-security assurance — no adoption confirmation for S25.
- KB got the Q4 bit guides: DRAM bits turning negative mid-single-digit — a notable deterioration signal — and NAND only low-single-digit; the foldable answer teased a new form factor.
- Citi extracted the 2025 framework: decoupling persists at least through 1H25, industry supply constrained by conversion bit-loss and clean-room limits, Chinese legacy supply rising, memory CapEx flat with conversion-over-capacity priority; Foundry 2nm detail (PPA analysis, MPW, packaging partnerships).
- Goldman asked Foundry CapEx (declining this year, cautious expansion next) and VD mid/low-end competition (premium leadership plus expanded entry lineups — a margin-relevant concession).
- Missing challenges: no one asked about the strike (last quarter's pre-submitted topic vanished); no one quantified the HBM revenue shortfall vs the "doubling quarterly" trajectory; no one asked the identity or revenue at risk of the delayed major customer; no Taylor question; no one pressed why the Q4 3E mix target fell from high-60s % to 50%; no MX margin commitment question this time (management volunteered only "close to double-digits").
Management credibility- Delivered: HBM3E 8-high MP did begin in Q3 as rescheduled; 64TB QLC MP in Q4 window confirmed; 2nm PDK released; MX Q3 shipments/ASP rebounded as guided (58M, $295); dividend at policy rate; FY CapEx guidance finally provided after two quarters of refusal.
- Missed/slipped: HBM3E mix below the mid-teens Q3 guide; Q4 3E mix target cut to ~50% from high-60s % — the second consecutive HBM timeline/mix disappointment, and the "revenue doubling each quarter / 2H ~3.5x 1H" framing quietly dropped (+>70% QoQ actual); DRAM bit guide missed (flat vs low-single-digit growth); NAND below bit guide; 128TB SSD moved from "Q4 lineup" to Q4 sampling/1H25 MP; Q3 OP below consensus with one-offs exceeding the miss gap.
- Candor improved in places: the miss was addressed head-on; the delay was pre-announced with prelims; aging-inventory depletion and its ASP/margin cost were disclosed; mobile/PC softness and Chinese legacy supply admitted; Display's conservative outlook stated plainly.
- Soft spots: qualification progress described only as "an important phase completed" — still no named customer, no revenue-at-risk quantification; one-off costs (incentives, engineering-run provisions) surfaced across DS divisions simultaneously and were not pre-signaled; "close to double-digits" MX profitability is a new, softer formulation; the value-up program is pure placeholder; multiple transcript garbles require release verification.
- Consistency watch: last quarter's "complete inventory normalization" language is now a firm "by end of year" commitment; HBM 2025 supply "2x 2024" was not repeated this quarter — check whether it still stands; the 2028 Foundry targets (4x customers/9x sales) went unmentioned.
What changed versus the prior quarter- HBM narrative reset: from "3E mix high-60s % by Q4, revenue doubling quarterly, 2H ~3.5x 1H" to "3E low-to-mid-teens in Q3, ~50% in Q4, +>70% QoQ" — a material de-rating of the near-term ramp, offset by new detail (enhanced 3E for next-gen GPUs, 1H25 MP; custom-HBM foundry flexibility).
- Memory momentum broke: Q2's price-led beat (OP +3.8T) became a Q3 miss (OP -1.3T) with GM -2.0pp, write-back tailwind shrinking, one-off incentive costs, and FX flipping from +0.8T to -0.5T.
- Bit trajectory deteriorated: Q3 DRAM bits flat (guide missed) and Q4 guided negative mid-single-digit — HBM crowding-out plus Chinese mobile inventory adjustment now bite volumes, not just mix.
- Inventory story advanced: from "sales exceeded production" to active aging-inventory depletion (margin-dilutive) with a hard end-of-year normalization deadline.
- CapEx transparency arrived: first FY guide (~56.7T) plus 2025 directional guidance (memory flat, Foundry down/cautious) — a disclosure upgrade.
- MX inflected up as guided: 54M→58M units, ASP $279→$295, profit up QoQ — but the profitability language softened to "close to double-digits," and Q4 is guided back down.
- The strike disappeared from disclosure entirely after being a headline risk last quarter — unresolved.
- New items: value-up program review announced; enhanced HBM3E product line created; new foldable form factor teased; smartphone 2025 growth guided to <1% (System LSI) — a cautious demand signal; VD returned to profit improvement YoY/QoQ.
Bull case- The Q3 miss is framed as one-off-heavy (incentives, engineering-run provisions, lower write-backs) rather than demand-driven: memory revenue still grew mid-single-digit USD, ASPs rose high-single-digit on both DRAM and NAND, and server/AI mix hit records (server SSD ~50% of NAND, DDR5 >80% of server DRAM).
- HBM's worst may be behind: both 3E stacks in MP, the key qualification phase complete, Q4 expansion starting, ~50% 3E mix in Q4, enhanced 3E for next-gen GPUs in 1H25, HBM4 on track for 2H25, and custom HBM with foundry flexibility widening the addressable customer base.
- 2025 supply discipline is bullish for pricing: conversion bit-loss caps industry bit growth, clean-room constraints persist, Samsung itself guides to limited production growth and flat CapEx — decoupling between tight advanced nodes and oversupplied legacy favors Samsung's mix.
- Q4 setup: DS results guided to improve; server SSD revenue +~10% QoQ; 64TB QLC MP and 128TB sampling this quarter; inventory normalization completed by year-end removes the margin drag.
- Structural optionality: 2nm GAA PDK in customer hands for 2025 MP, Memory+Foundry HBM prototype collaboration, MX flagship/AI cycle (S25, new foldable form factor, XR), and a value-up program review that could unlock capital return.
Bear case- HBM execution risk is now demonstrated, not hypothetical: two straight quarters of 3E slippage (MP timing, then mix/qualification), the Q4 mix target cut by ~20pp, and the major-customer delay still unnamed and unquantified — Samsung is ceding the most valuable memory window to competitors while its 2025 supply commitments (2x) went un-reaffirmed.
- Earnings quality deteriorated: GM -2.0pp, OP -1.3T QoQ, with the P&L propped in prior quarters by write-backs now shrinking; one-off costs surfaced across Memory, System LSI, and Foundry simultaneously; FX is now a headwind.
- Volume signals are weak: DRAM bits guided negative in Q4, NAND below guide, mobile/PC demand soft, Chinese legacy supply rising, and System LSI sees 2025 smartphone growth <1% — the non-AI memory market is rolling over while AI revenue is delayed.
- Foundry is still losing ground financially: earnings declined QoQ on provisions, CapEx is being cut, capacity expansion deferred, and the 2028 targets were quietly dropped this quarter; Taylor remains unmentioned.
- Finished goods face a soft Q4 (DX sales guided down) and 2025 component-cost inflation; MX profitability language slipped to "close to double-digits"; Display is "quite conservative" on competition; large panels' earnings fell QoQ.
- The strike is unresolved in disclosure — no update, no cost, no status.
- Transcript garbles (CapEx figures, product names, QLC/TSG references) require verification before modeling.
Next-quarter watchlist- HBM proof points: Q4 3E mix vs the ~50% target; whether the delayed major customer's sales expansion actually materialized; enhanced-3E MP timing in 1H25; HBM4 development milestones toward 2H25 MP; any named-customer qualification; whether 2025 HBM supply (2x 2024) is re-affirmed; absolute HBM revenue disclosure.
- Memory P&L: Q4 DRAM bits vs negative mid-single-digit guide; NAND vs low-single-digit; ASP trajectory amid legacy roll-down; confirmation that inventory normalization completed by year-end; write-back/provision run-rate; one-off cost recurrence.
- AI-SSD: server SSD +~10% QoQ bit-revenue guide; 64TB QLC MP ramp; 128TB sampling progress and 1H25 full-lineup MP; whether the 4x YoY server SSD full-year guide was met.
- One-offs and quality: incentive provisioning magnitude, engineering-run wafer provisions, and whether Q4 DS improvement is clean of further one-offs; FX direction.
- MX: Q4 shipment/ASP decline vs guide; full-year flagship double-digit growth claim; S25 launch (1H25) and Exynos 2500 adoption breadth; new foldable form factor timing; 2025 component-cost mitigation.
- Foundry: 2nm MP readiness for 2025 and customer PPA/MPW traction; second-gen 3nm wearable MP confirmation (missed this quarter); loss trajectory; Taylor status — press again; whether 2028 targets return.
- Display/VD: mobile OLED share defense in a "conservative" Q4; large-panel earnings; 8.6G IT OLED cadence; VD premium mix and TV Plus monetization.
- Corporate: value-up program specifics (the highest-impact pending disclosure); strike status — demand an update after this quarter's silence; verification of all garbled figures (Q3 CapEx, YTD DS/Display splits, product names) against the release.
|
Corbanu