| Jul 28, 2026 | -21.24% | — | Read transcript briefingQuarter in one view- Q2 2026: the supercycle compounded again. Revenue KRW79.3trn (+51% QoQ, +257% YoY) — second consecutive record. OP KRW60.5trn (+61% QoQ, +557% YoY); OPM 76% (+5pp QoQ, all-time high). D&A KRW4.0trn; EBITDA KRW64.6trn (81% margin); pre-tax KRW122.7trn; net KRW93.9trn (118% net margin).
- Both guides hit; ASPs decelerated but stayed extreme. DRAM bits +high-single-digit QoQ (as guided); DRAM ASP +~30% QoQ (vs +mid-60% in Q1 — sharp deceleration, and Daiwa flagged it as below market expectations). NAND bits +mid-teens QoQ (as guided); NAND ASP +mid-50% QoQ (vs +mid-70%). eSSD revenue doubled QoQ; Solidigm ≥30TB eSSD revenue more than tripled QoQ.
- Earnings quality distortion went from large to dominant. Non-op +KRW62.2trn, including KRW63.3trn from sale/valuation of investment assets (Kioxia stake sale confirmed in Q&A) plus KRW1.1trn FX gains. Net profit (93.9trn) exceeds operating profit (60.5trn) by 55% — the majority of the headline print is one-off.
- Balance sheet: cash incl. ST investments KRW88trn (+33.6trn QoQ); debt KRW18.6trn (-0.7trn); net cash KRW69.4trn (from 35trn — 69% of the way to the >100trn target in one quarter); D/E 7% (from 12%).
- Capex finally quantified: 2026 capex "high KRW40 trillion range" — the first absolute figure in at least six quarters, driven by M15X pull-forward and Yongin Fab 1 (early 2027 cleanroom). New announcements: P&T7 (advanced packaging), M17 (new NAND fab), and a new domestic cluster beyond Yongin.
- ADR executed: listed on Nasdaq July 10; described as the largest-ever US IPO by a foreign company. Conversion limit 17,790,000 shares.
- Q3 2026 guide: DRAM bits +~10% QoQ; NAND bits +low-single-digit QoQ. No ASP guidance.
What management is focused on- Defending AI-capex durability against a new bear angle. The opening question (J.P. Morgan) raised data-center leasing and efficient models as slowdown risks. Management reframed both as utilization/monetization transitions, citing "explosive demand for recent high-efficiency AI models" and customer conversations supporting investment "beyond next year." Same structural-demand defense as prior quarters, now aimed at a different threat.
- LTA architecture disclosure — the biggest new information on the call. ~10 customers signed (including key customers); typical term ~5 years; non-uniform pricing mechanisms designed to damp volatility; deposits incorporated to enforce purchase commitments and improve demand visibility. LTA share of total sales deliberately undisclosed ("appropriate level"). This is the evidentiary backbone of the no-oversupply claim.
- HBM4 execution narrative restored. Mass production shipments began in Q2 for key customers; yield and quality "nearing HBM3E levels" (HBM3E described as mature); full ramp in H2. This directly answers last quarter's language-drift concern.
- Capacity expansion as competitive duty. M15X mass production pulled forward; Yongin Fab 1 early-2027 cleanroom followed by rapid capacity build-out; P&T7 packaging and M17 NAND announced; new domestic cluster beyond Yongin — a reversal of last quarter's flat "no plans" for fabs beyond Yongin. Overseas expansion left open ("no further decisions"), with site selection framed around power, water, talent, supply chain, customer access.
- Shareholder returns still deferred. "Various options" under evaluation; specifics blocked by ADR-related regulatory constraints; communication promised "within the year" — same deadline as last quarter, now with a regulatory rationale.
Key numbers and quarter mechanics- P&L: revenue 79.3trn (+51% QoQ, +257% YoY); OP 60.5trn (76% OPM, +5pp QoQ); D&A 4.0trn; EBITDA 64.6trn (81%); non-op +62.2trn (investment-asset sale/valuation +63.3trn, FX +1.1trn); pre-tax 122.7trn; net 93.9trn (118% margin).
- Volumes/ASP: DRAM bits +HSD QoQ (in line); DRAM ASP +~30% QoQ — management attributed the below-expectation print to high-value product shipments pushed into H2 and mix effects, and guided H2 bit growth above H1 with HBM4 and 1cnm mix lifting blended ASP. NAND bits +mid-teens QoQ (in line, off Q1's low base); NAND ASP +mid-50% QoQ.
- Mix detail: sales centered on HBM3E and AI server DRAM; server LPDDR incl. SOCAMM2 "grew significantly"; eSSD revenue 2x QoQ; Solidigm ≥30TB eSSD revenue >3x QoQ.
- Demand outlook: 2026 DRAM demand +mid-20%, NAND +high-10s (bit growth); supply-demand balance unlikely to improve near-term due to process complexity and fab lead times.
- Q3 guide: DRAM bits +~10% QoQ (server-led); NAND bits +LSD QoQ — a notable NAND deceleration vs Q2's mid-teens.
- Balance sheet/capex: cash 88trn; debt 18.6trn; net cash 69.4trn; D/E 7%; 2026 capex high-40trn range.
- PC/mobile: "temporary sales adjustments due to difficulties in securing memory" — the framing shifted from demand destruction (Q1: cost-pressure shipment cuts) to supply starvation; recovery expected as shortage eases.
- Persistent gaps: no DRAM/NAND revenue split, no HBM revenue/mix, no segment margins, no LTA coverage ratio, no HBM pricing detail, no Kioxia gain mechanics.
Product and launch scorecard- HBM4: mass production shipments began Q2 for key customers; yield/quality "nearing HBM3E" levels; full ramp H2 2026. Clears last quarter's "getting ready to ramp" ambiguity — the strongest execution evidence on the call. Still no volume, share-of-HBM, or customer-breadth quantification.
- HBM4E: samples delivered to a major customer in H1 (ahead of the prior "H2 2026 samples" timeline); built on a mature, proven process; volume production targeted 2027.
- iHBM (new disclosure): in development for HBM5-era thermal management; cooling elements integrated in-package; >30% thermal-resistance reduction claimed. Pre-product but a concrete differentiator alongside hybrid bonding.
- SOCAMM2 (1cnm): full supply commenced in Q2; server LPDDR sales "grew significantly"; sample shipments planned to broaden the customer base beyond the initial platform.
- 321-layer NAND: became the largest share of NAND production in Q2; ~50% of domestic capacity by end-2026 reiterated ("as planned").
- NAND AI-storage strategy: portfolio framing across TLC eSSD, high-capacity QLC eSSD (data lake/HDD replacement), and SLC-mode high-performance SSDs; new tiers for KV-cache offload and near-GPU storage in development. Solidigm's ≥30TB momentum (>3x QoQ) is the first hard revenue evidence for the QLC/HDD-replacement thesis.
- 2027 HBM pricing: negotiations underway, "progressing smoothly"; management explicitly linked HBM pricing to sharply higher conventional DRAM prices plus resource/opportunity cost of HBM production — an analyst-readable signal that 2027 HBM pricing is being negotiated up, not down.
- Quiet/dropped threads: LPDDR6 flagship win not mentioned; CXL, HBF, PQC21 not mentioned; 245TB eSSD, GDDR7, LPDDR server module, OpenAI LOI, tariffs, China/VEU all still absent.
Sell-side read-through- Jay Kwon (J.P. Morgan): AI-capex slowdown risk (leasing, efficient models) — the cycle-longevity question has evolved from spot-price softness to hyperscaler capex behavior. Management's answer was emphatic and customer-cited but number-free.
- Rok-ho Kim (Hana): basis for long-term demand and oversupply risk — got the LTA-anchored demand-visibility defense and "phased, flexible" execution pledge. "Will not lead to oversupply right away" is a carefully qualified version of last quarter's stronger claim.
- Sunwoo Kim (Meritz): LTA terms — extracted the call's most valuable disclosures: ~10 customers, ~5-year typical terms, deposits, non-uniform pricing. Coverage ratio refused.
- SK Kim (Daiwa): the only challenging question — DRAM ASP +~30% "fell below market expectations." Management blamed H2-pushed high-value shipments and mix, and guided H2 bit growth above H1 with mix-driven ASP recovery. A plausible but unverifiable explanation; it concedes Q2 blended ASP was mix-depressed, not price-depressed.
- Dong-hee Han (SK Securities): HBM competitive moat vs fast-progressing rivals — produced the yield-nearing-HBM3E claim and the iHBM reveal. Competition anxiety is now explicit in the question set.
- Nicolas Gaudois (UBS): 2027 HBM pricing for HBM3/4/4E — got direction (constructive, conventional-DRAM-linked) but no terms, as expected.
- Sanjeev Rana (CLSA): overseas expansion (US/Japan) — door opened ("no domestic/overseas distinction in principle") but nothing decided.
- Hyung-geun Ryu (Daishin): NAND segmentation (QLC vs SLC-mode) — portfolio answer, no product specifics or customer names.
- Young-ho Ryu (NH): ADR fungibility — one-way conversion initially (ADR→stock free from July 30; stock→ADR limited by filings and the 17.79mn-share cap); no decision on expanding the program.
- Su-rim Lee (DS): capital allocation post-Kioxia/ADR — returns still "within the year," now with an ADR regulatory gag as the stated reason for no specifics.
- Notable absences: no question on the KRW63.3trn one-off gain dominating the print (third straight quarter of unexamined non-op swings); no HBM4 share/volume quantification demanded this time; no tariff, China, or OpenAI questions; no challenge on the NAND Q3 guide decelerating to LSD.
Management credibility- Positive: guidance hit on both axes again. DRAM +HSD, NAND +mid-teens — both as guided. The bit-guidance track record is now four-plus quarters clean.
- Positive: HBM4 language-drift resolved by delivery. Last quarter's "getting ready to ramp" became "mass production shipments began in Q2" with yields nearing HBM3E — the softening was scheduling, not execution. HBM4E samples also landed ahead of the stated H2 window.
- Positive: capex opacity ended. High-40trn 2026 figure after six quarters without a number; M15X pull-forward and Yongin timeline give it structure.
- Positive: LTAs moved from "structural options under review" to ~10 signed deals with disclosed architecture (5-year terms, deposits) — real progress on the prior quarter's key watch item, though coverage and pricing remain undisclosed.
- Mixed: ASP explanation under pressure. The below-consensus DRAM ASP was attributed to mix and shipment timing — consistent with the HBM4 H2 ramp story, but it means Q2's headline ASP strength was partly portfolio optics, and the H2 recovery claim is now falsifiable.
- Negative: headline earnings are now mostly non-operating. Net margin 118% vs OPM 76%; KRW63.3trn of one-off gains in a 93.9trn print. Management presented the net figure without flagging its composition beyond the line-item disclosure, and no analyst probed it.
- Negative: shareholder returns slipped again in substance. Same "within the year" promise as Q1, now with a regulatory constraint attached — the ADR process has become a reason for continued deferral while net cash piles toward 100trn.
- Watch: reversal on fabs beyond Yongin. Last quarter: "no plans." This quarter: new domestic cluster announced plus openness to overseas sites. The change is explained by demand, but it is a genuine strategic U-turn within one quarter.
What changed versus the prior quarter- ASP momentum decelerated hard while volumes re-accelerated: DRAM ASP +~30% (vs +mid-60%), NAND ASP +mid-50% (vs +mid-70%); DRAM bits +HSD and NAND bits +mid-teens (vs flat/-10%). The quarter was more volume-real than Q1 — and Q3 guides DRAM +~10%, the strongest bit guide of the cycle.
- HBM4 crossed from preparation to revenue: Q2 mass-production shipments, yields nearing HBM3E, H2 full ramp; HBM4E sampled early. Last quarter's central credibility question was answered.
- Capex went from "increase significantly" to high-40trn, with M15X pulled forward and three new capacity announcements (P&T7, M17, post-Yongin cluster). The investment supercycle is now matching the price supercycle.
- LTAs became concrete: ~10 customers signed, ~5-year terms, deposits, volatility-dampening pricing — vs "various approaches under review" in Q1.
- ADR completed: Nasdaq listing July 10, largest foreign-company US IPO; fungibility mechanics disclosed.
- Earnings quality deteriorated from caveat to distortion: non-op gains 14trn → 62.2trn; net margin 77% → 118% vs OPM 72% → 76%. The Kioxia sale is new information explaining the jump.
- PC/mobile narrative flipped: from demand destruction (cost-pressure shipment cuts) to supply starvation ("difficulties securing memory") — a friendlier framing that implies deferred, not destroyed, demand.
- New disclosures: iHBM (>30% thermal-resistance reduction, HBM5-era); 321-layer now the largest share of NAND output; 2026 demand growth forecasts (DRAM +mid-20%, NAND +high-10s); 2027 HBM pricing talks underway with conventional-price linkage.
- Dropped/quiet: LPDDR6, CXL, HBF, PQC21, tariffs, OpenAI LOI, China/VEU, 245TB eSSD, GDDR7.
Bull case- The cycle is now volume-led, not just price-led: Q3 DRAM bit guide of +~10% with HBM4 ramping and 1cnm conventional mix rising implies H2 revenue growth even if conventional ASP momentum flattens. Management explicitly guided H2 bit growth above H1 with mix-driven ASP improvement.
- HBM execution risk is de-risked: HBM4 in mass production at near-mature yields, HBM4E sampled early, 2027 pricing talks "progressing smoothly" with conventional-price tailwinds, and iHBM extending the roadmap into HBM5.
- Demand visibility is now contractual: ~10 five-year LTAs with deposits convert the no-oversupply argument from assertion to structure — and give management a rational basis for the high-40trn capex and post-Yongin cluster.
- NAND has a second engine: eSSD revenue doubling QoQ and Solidigm ≥30TB tripling is the first hard evidence that AI-inference storage (QLC HDD replacement, KV-cache offload) is a revenue line, not a slide.
- Balance sheet is a fortress: 69.4trn net cash, 7% D/E, 81% EBITDA margin — the 100trn net-cash target is reachable within quarters, and the ADR broadens the investor base into it.
- PC/mobile weakness reframed as supply starvation implies pent-up demand that returns when supply eases — upside to the +mid-20% DRAM demand forecast rather than downside.
Bear case- The headline print is mostly one-off. Strip the KRW63.3trn investment gains and Q2 net profit is roughly half the reported 93.9trn; net margin of 118% is not an earnings-power data point. Three straight quarters of large unexamined non-op items make reported net income nearly useless for trend analysis.
- DRAM ASP missed expectations, and the explanation (mix, H2-pushed shipments) concedes that blended pricing power is now hostage to HBM4 ramp execution. If the H2 mix recovery doesn't materialize, the ASP deceleration (+mid-60% → +30%) will look like the start of normalization, not a timing artifact.
- NAND momentum is already fading in the guide: Q3 bits +LSD after +mid-teens, with no ASP guide — the eSSD surge may be lumpy.
- Capex is escalating aggressively at peak margins: high-40trn in 2026 plus M15X acceleration, M17, P&T7, and a post-Yongin cluster — a strategic U-turn from "no fabs beyond Yongin" one quarter ago. If the LTA-anchored demand visibility proves softer than claimed, this is exactly how past oversupply cycles were built.
- Shareholder returns remain a promise, not a policy: two consecutive quarters of "within the year," now gated by ADR regulatory constraints, while cash accumulates and reinvestment is prioritized.
- Competitive anxiety is rising (explicit HBM-moat question), and management's answer rested on track record rather than quantified current share, qualification breadth, or 2027 volume commitments.
- Still unexamined: tariffs, China/VEU, OpenAI LOI, customer concentration, LTA pricing floors/caps, and the terms under which deposits could be forfeited.
Next-quarter watchlist- H2 mix-recovery claim: whether HBM4 ramp and 1cnm conventional shipments deliver the promised H2 bit growth above H1 and blended-ASP improvement — the direct test of the Q2 ASP-miss explanation.
- Q3 delivery vs guide: DRAM bits +~10%, NAND bits +LSD; whether NAND's deceleration is timing or eSSD lumpiness; any ASP commentary as 2027 contract talks progress.
- HBM4 quantification: share of HBM revenue, customer count, and whether 2027 HBM pricing concludes with disclosed direction; HBM4E development milestones toward 2027 volume production.
- LTA conversion and coverage: additional signings beyond ~10, any disclosure of sales coverage ratio, deposit magnitudes, or pricing-mechanism structure — and whether deposits show up in the balance sheet.
- Shareholder-return plan: the twice-promised "within the year" framework — size, recurrence, mix — now that the ADR regulatory window is the stated constraint; watch whether the constraint outlives the lockup period.
- Capex trajectory: whether high-40trn holds or rises; M15X ramp contribution; Yongin Fab 1 early-2027 cleanroom on schedule; any overseas-site decision following the opened door.
- Earnings quality: Q3 non-op composition post-Kioxia; whether net margin re-converges toward OPM; any remaining investment-asset stakes that could produce further one-offs.
- SOCAMM2 and Solidigm traction: customer-base expansion via samples; whether ≥30TB eSSD growth sustains after a 3x quarter; 321-layer progress toward ~50% of domestic capacity by year-end.
- PC/mobile: whether "supply starvation" framing holds or demand destruction re-emerges; any LPDDR6 flagship evidence.
- Overdue items: tariff strategy (two quarters past the promised update), OpenAI LOI, China/VEU, LPDDR6/CXL/HBF status, 245TB eSSD, GDDR7, LPDDR server module.
|
| Apr 22, 2026 | -2.13% | — | Read transcript briefingQuarter in one view- Q1 2026: the price supercycle went vertical. Revenue KRW52.6trn (+60% QoQ, +198% YoY) — first quarter above KRW50trn, another record. OP KRW37.6trn (nearly 2x QoQ; OPM 72%, +13pp QoQ, all-time high). D&A KRW3.7trn; EBITDA KRW41.3trn (79% margin); pre-tax KRW51.6trn; net KRW40.3trn (77% margin).
- ASPs did almost everything again, at an even more extreme rate. DRAM bits flat QoQ (as guided); DRAM ASP +mid-60% QoQ. NAND bits -~10% QoQ (high base, lower discrete sales, longer lead times from mix shift); NAND ASP +mid-70% QoQ. Server DRAM and eSSD led pricing.
- Earnings quality caveat persists and grew. Non-op +KRW14trn, including KRW9.9trn investment-asset valuation gains (up from 6.6trn) and KRW1.6trn FX gain. Net margin (77%) again exceeds OPM (72%) — flattered by non-operating items, as in Q3.
- Balance sheet inflected hard: cash incl. short-term investments KRW54.3trn (+19.4trn QoQ); debt KRW19.3trn (-2.9trn); net cash KRW35trn (from ~12.7trn); D/E 12% (from 18%). New explicit target: net cash >KRW100trn.
- Capital allocation pivoted back toward reinvestment: management now calls reinvesting cash "the best use of capital," while promising a shareholder-return implementation plan (dividends, buybacks, cancellations) "within the year." ADR moved from "under study" to a confidential SEC registration submitted March 24, targeting a US listing within the year.
- Q2 2026 guide: DRAM bits +high-single-digit QoQ; NAND bits +mid-10% QoQ (321-layer products and eSSDs).
What management is focused on- Defending the cycle's duration against peak-out fears. The first two questions were both about price sustainability (spot-price softness, supercycle longevity). Management's answer: spot is a tiny, unrepresentative slice; the imbalance is structural (AI-driven demand + post-downturn underinvestment + cleanroom space constraints); customers "prioritize securing volume over pricing"; the upcycle "is likely to last longer than in the past."
- Reframing efficiency technology as demand-accretive. Pushed back on SRAM/LPU substitution (hybrid architecture: LPUs for fast response, HBM GPUs for heavy compute) and on KV-cache compression (same memory used more efficiently → more services → more memory). This is a pre-emptive defense of the long-term demand thesis.
- HBM4 ramp "in line with the agreed schedule," with competitiveness framed as a bundle (TSV, packaging, yield, quality, supply stability) rather than any single metric. Notably, management said it is optimizing HBM vs general-DRAM allocation "for balanced growth in the AI ecosystem rather than trying to maximize revenue."
- Capacity as the strategic bottleneck: 2026 capex to "increase significantly" YoY, weighted to Yongin infrastructure, M15X ramp, and EUV procurement. Yongin Fab 1 Phase 1 cleanroom opening accelerated three months (May 2027 → February 2027); Phase 1 is DRAM; Phases 2–6 product mix TBD. No fabs beyond Yongin planned.
- LTAs evolving toward multi-year structures with "various approaches and structural options" under review — but supply constraints limit how many requests can be accommodated. Framed as a route to lower industry volatility and better investment efficiency.
- Financial-soundness target quantified for the first time: net cash >KRW100trn, pursued "in parallel" with expanded returns.
Key numbers and quarter mechanics- P&L: revenue 52.6trn (+60% QoQ, +198% YoY); OP 37.6trn (72% OPM, +13pp QoQ); D&A 3.7trn; EBITDA 41.3trn (79%); non-op +14.0trn (FX +1.6trn, valuation gains +9.9trn); pre-tax 51.6trn; net 40.3trn (77%).
- Volumes/ASP: DRAM bits ~flat QoQ (in line with guide); DRAM ASP +mid-60% QoQ (vs +mid-20% in Q4 — acceleration). NAND bits -~10% QoQ (as guided "down somewhat"); NAND ASP +mid-70% QoQ (vs +low-30% — acceleration). Sales concentrated in HBM and ≥128GB high-density server modules.
- Q2 2026 guide: DRAM bits +high-single-digit QoQ; NAND bits +mid-10% QoQ. No ASP guidance given.
- Balance sheet: cash 54.3trn; debt 19.3trn; net cash 35trn; D/E 12%.
- Capex: 2026 to "increase significantly" YoY; majority to Yongin cluster infrastructure, M15X ramp, EUV tools. No won figure; the prior quarter's mid-30% capex/revenue anchor was not repeated on this call.
- PC/mobile: "some signs of demand softening" — set-shipment adjustments and portfolio changes from memory cost pressure — explicitly offset by server demand. Consistent with last quarter's first demand-destruction acknowledgment, now confirmed as occurring.
- Disclosure gaps persist: no DRAM/NAND revenue split, no HBM revenue or mix, no segment margins, no absolute capex figure, no LTA terms, no HBM4 share-of-HBM answer (asked directly, not answered).
Product and launch scorecard- HBM4: "getting ready to ramp up production" of products meeting customer requirements, in line with customer-agreed schedules. Language shifted from last quarter's "mass production underway" to "getting ready to ramp" — worth flagging as a possible softening, though management attributes timing to customer schedules. Customer demand "for the next three years far exceeds our current supply capacity." No qualification breadth, volume, or mix disclosure; the direct question on HBM4's year-end share of HBM went unanswered.
- HBM4E: samples in H2 2026, mass production targeted 2027; core die on 1cnm (mass production since late 2025, yields "mature"); base die under development with customers. Packaging/logic-die detail requested by Citi was not given beyond "technology internalization."
- SOCAMM2 192GB (1cnm LPDDR5X-based): mass production began this month (April 2026); >2x bandwidth and >75% energy efficiency vs conventional RDIMMs; explicitly optimized for NVIDIA's Vera Rubin platform. The most concrete new-product milestone on the call.
- LPDDR6 (1cnm, industry first): development completed in Q1; +33% speed, >20% power efficiency vs LPDDR5X; full-scale supply from H2, starting with a major smartphone customer's next flagship.
- 321-layer QLC NAND: world's first developed in April, customer qualification completed; PQC21 client SSD (CTF-based 321-layer QLC) supply begun; enterprise lineup to span TLC + QLC. Solidigm explicitly cited for high-capacity QLC eSSD synergy — the first Solidigm mention in at least two quarters.
- 321-layer migration target reiterated: >50% of domestic NAND production by end-2026; framed as a two-generation jump from 176-layer with "significant productivity gain."
- CXL: Gen-1 (CXL 2.0) modules completed customer qualification last year; Gen-2 (CXL 3.0) in development; MOU with a cloud provider (unnamed) for CXL/PIM validation. Positioned as a KV-cache offload option.
- HBF: consortium launched February 2026 for standardization — now slightly more concrete than last quarter's "further develop," but still pre-product.
- Dropped/quiet threads: 245TB eSSD and ultra-high-IOPS SSD not mentioned; GDDR7 not mentioned; LPDDR server module still absent (three quarters); OpenAI LOI still unaddressed; 1cnm >50%-of-Korean-DRAM-capacity target not updated.
Sell-side read-through- Rok-ho Kim (Hana) and Sunwoo Kim (Meritz): both opened on price-cycle durability (spot softness, peak-out risk) — the market's dominant anxiety has shifted from demand reality to cycle longevity. Management gave its most emphatic structural-imbalance defense yet, attributing spot softness to distributor inventory release.
- Nicolas Gaudois (UBS): pressed for LTA specifics (scope across DRAM/NAND/HBM, terms, timeline). Got philosophy, not terms: multi-year structures under review, supply constraints limit accommodation, no product-scope or contractual detail.
- SK Kim (Daiwa): memory-efficiency threat — produced the LPU/SRAM hybrid-architecture and KV-cache-compression rebuttals. Strategy-rich, number-free, but a coherent answer.
- Ricky Seo (HSBC): HBM4 qualification, full-scale shipment timing, and year-end HBM4 share of HBM — the first two got schedule-language answers; the share question was ignored entirely. The market is still probing HBM4 execution, and management still won't quantify.
- Yeong-Min Ko (Daol): investment expansion vs oversupply risk — management committed to significant 2026 capex growth while arguing LTAs and customer intelligence prevent a repeat of past oversupply. "No major concerns about oversupply as in the past" is a strong, falsifiable claim.
- Ye-Hyun Kwon (J.P. Morgan): NAND/eSSD plans — yielded the 321-layer QLC qualification milestone, the >50% migration target, and the Solidigm synergy mention.
- Young-ho Ryu (NH): post-HBM next-gen memory — produced SOCAMM2, CXL 3.0, and HBF consortium details.
- Peter Lee (Citi): HBM4E differentiation, packaging, logic die — got sample/MP timing (H2 2026 / 2027) and 1cnm core die, but no packaging or logic-die specifics.
- Jongwook Lee (Samsung): Yongin operations and additional fabs — extracted the three-month cleanroom acceleration (Feb 2027), Phase 1 = DRAM, and a flat "no plans" for fabs beyond Yongin.
- Simon Woo (BofA): commodities/energy exposure (helium, bromine, tungsten, LNG) — a new risk angle; management claims diversified suppliers, sufficient inventories, and LTA-based LNG sourcing; "limited" impact throughout.
- Dong-Hee Han (SK Securities): KRW100trn net-cash goal vs shareholder returns, and ADR — got the "within the year" return-plan commitment and confirmation the SEC review is underway with nothing further disclosable.
- Notable absences: no question on the KRW9.9trn valuation gain (second straight quarter of large unexamined non-op items); no OpenAI LOI; no tariff question (last quarter's deferral unaddressed); no China/VEU; no challenge on the missing absolute capex number; no HBM4 pricing question.
Management credibility- Positive: guidance beaten again on the beatable axis. Q1 DRAM bits flat as guided; NAND bits -~10% as guided ("down somewhat"). The supply-constrained framing from Q4 proved accurate — and ASPs (+mid-60%/+mid-70%) far exceeded anything implied.
- Positive: delivery on flagged milestones. ADR moved from "looked into" to a filed confidential registration in one quarter. SOCAMM2 went from portfolio mention to mass production. 321-layer QLC reached qualification. Yongin was accelerated, not delayed.
- Positive: financial targets now explicit — net cash >KRW100trn and a return plan "within the year" create testable commitments.
- Mixed: HBM4 language drift. Last quarter: mass production of customer-requested volumes underway, yield parity targeted. This quarter: "getting ready to ramp up," no yield update, no share-of-HBM answer despite a direct question. Not necessarily a slip — customer schedules drive timing — but the specificity went backward while the questions got more pointed.
- Negative: earnings quality is now a persistent structural issue. KRW9.9trn of valuation gains in Q1 (after 6.6trn in Q4); net margin above OPM again. No analyst has asked about either quarter's non-op swings.
- Negative: capex opacity continues — "increase significantly" with no won figure, and the mid-30% revenue anchor offered last quarter was not reiterated. Six quarters without an absolute number.
- Watch: the "no oversupply repeat" and "pricing strength continues for the time being" claims are now the central falsifiable commitments. Management is explicitly tying its capex discipline to LTA-based demand visibility it won't disclose.
- Watch: tariff exposure went entirely unaddressed — last quarter's "communicate at a later date" has not been honored on this call.
What changed versus the prior quarter- Price acceleration, not deceleration: DRAM ASP +mid-60% (vs +mid-20%), NAND ASP +mid-70% (vs +low-30%). OPM 58% → 72%. The cycle steepened rather than normalized.
- Capital allocation tone reversed again: Q4 delivered a surprise KRW14.3trn of dividends/cancellation; Q1 reframes reinvestment as "the best use of capital," with returns deferred to a plan "within the year" and a new KRW100trn net-cash hurdle. The direction of the returns story is now explicitly episodic, not recurring.
- ADR moved from study to execution: confidential SEC filing submitted March 24; US listing targeted within the year.
- HBM4 narrative moved from defense to schedule language, but with less specificity: no yield-parity update, no share target, and "getting ready to ramp" replacing "mass production underway." HBM4E timeline (samples H2 2026, MP 2027) disclosed for the first time.
- Demand-destruction acknowledgment firmed up: from "customers revising plans" to observed "adjustments in set shipments and changes in product portfolios" in PC/mobile — still framed as fully offset by server.
- New disclosures: LPDDR6 development complete (H2 supply, flagship smartphone win); SOCAMM2 192GB in mass production for Vera Rubin; 321-layer QLC qualified (world first); PQC21 shipping; CXL 3.0 Gen-2 and cloud-provider MOU; HBF consortium (Feb 2026); Yongin Phase 1 accelerated to Feb 2027; net-cash >100trn target; Solidigm named again.
- Dropped threads: tariff response (promised "later date" never came), OpenAI LOI, 245TB eSSD, GDDR7, LPDDR server module, 1cnm capacity-share target, China/VEU.
Bull case- The supercycle is steepening, not peaking: +mid-60%/+mid-70% ASPs in a seasonally weak quarter, with management attributing spot softness to distributor inventory and asserting customers prioritize volume over price. Q2 bit guidance (+HSD DRAM, +mid-10% NAND) implies volume re-acceleration on top of still-favorable pricing.
- Margins are at levels that reset the earnings base: 72% OPM, 79% EBITDA margin, KRW35trn net cash, D/E 12% — one quarter of Q1 profit roughly equals the entire FY2025 OP (37.6trn vs 47.2trn).
- Demand visibility is structurally improving: multi-year LTAs under active construction, three years of HBM demand exceeding supply capacity, and supply additions physically gated by cleanroom construction lead times (Yongin Phase 1 only opens Feb 2027).
- Product execution is broadening beyond HBM: SOCAMM2 in mass production for Vera Rubin, LPDDR6 with a flagship smartphone design win, world-first 321-layer QLC qualified with Solidigm synergy, CXL 3.0 and HBF optionality — the AI-memory portfolio is wider than a year ago.
- The efficiency-threat bears were answered directly: LPU/SRAM framed as hybrid-complementary, KV-cache compression as demand-expanding. Coherent, architecture-level rebuttals.
- ADR listing within the year could broaden the investor base into the strongest earnings momentum in company history.
Bear case- This is now a pure price-spike earnings stream. Bits were flat/down; the entire +13pp OPM gain came from ASP. When conventional pricing turns, the same leverage works in reverse — and management's "for the time being" phrasing on pricing durability is deliberately unbounded.
- HBM4 specificity regressed. "Getting ready to ramp" vs last quarter's "mass production underway," no yield update, and a refused share-of-HBM question — while HSBC and Citi keep probing. If execution were clean, quantification would be easy.
- Reported earnings are increasingly uninformative: KRW9.9trn valuation gains inside a KRW40.3trn net print; two consecutive quarters of multi-trillion-won non-op swings in both directions, and no analyst scrutiny.
- PC/mobile demand destruction is now observed, not hypothetical — shipment adjustments and portfolio changes are happening; the "server offsets everything" claim holds only while AI capex holds.
- Capex is escalating into the peak with no absolute number, the mid-30% anchor quietly dropped, and a KRW100trn net-cash target that implicitly prioritizes hoarding/reinvestment over the returns the market just tasted.
- The returns story whipsawed: record payout in Q4, "reinvestment is the best use of capital" in Q1, plan promised "within the year." Episodic capital returns at cycle peak earnings is a classic late-cycle pattern.
- Unexamined risks remain unexamined: tariffs (promised update never delivered), OpenAI LOI, China, customer concentration, and the terms of the LTAs on which the entire no-oversupply argument rests.
Next-quarter watchlist- Q2 delivery vs guide: DRAM bits +HSD, NAND bits +mid-10% — and whether ASP momentum persists into H2 contract negotiations or the spot softness migrates to contract pricing.
- HBM4 hard proof: actual ramp volumes, yield-parity status (targeted last quarter, silent this quarter), qualification breadth, and any quantified HBM4 share of HBM revenue by year-end. Watch for further language drift on timing.
- HBM4E milestones: sample shipments in H2 2026; base-die and packaging detail eventually owed to Citi's unanswered question.
- LTA conversion: whether "structural options under review" produce disclosed multi-year agreements with volumes, durations, or prepayments — the linchpin of the no-oversupply thesis.
- Shareholder-return plan: the promised "within the year" framework — size, recurrence, buyback vs dividend mix — and whether the KRW100trn net-cash target caps it.
- ADR execution: SEC review progress, offering size/structure, listing timing.
- Capex specificity: whether a won figure ever arrives; M15X ramp contribution; Yongin Feb 2027 cleanroom on track.
- PC/mobile destruction: whether shipment cuts and spec downgrades broaden, and whether LPDDR6 flagship adoption evidences the on-device-AI offset.
- SOCAMM2/321-layer QLC traction: revenue contribution, customer count beyond NVIDIA's platform, and progress toward >50% 321-layer migration by end-2026.
- Earnings quality: size of valuation gains/losses in Q2 non-op; whether net margin re-converges toward OPM.
- Overdue items: tariff strategy communication, OpenAI LOI status, 245TB eSSD, GDDR7, LPDDR server module (three quarters silent), 1cnm capacity-share target, China/VEU.
|
| Jan 28, 2026 | +6.49% | — | Read transcript briefingQuarter in one view- Q4 2025: another record, driven by price, not volume. Revenue KRW32.8trn (+34% QoQ, +66% YoY, all-time high); OP KRW19.2trn (+68% QoQ, +137% YoY; OPM 58%, +11pp QoQ); D&A KRW3.6trn; EBITDA KRW22.7trn (69% margin); pre-tax KRW17.7trn; net KRW15.2trn (46% margin).
- The mechanics flipped from volume-led to ASP-led. DRAM bits +low-single QoQ (in line with guide) but DRAM ASP +mid-20% QoQ on conventional DRAM price spikes; NAND bits ~+10% QoQ (beat vs +low-single guide, off Q3's low base) with NAND ASP +low-30% QoQ. High-density DDR5 module shipments +~50% QoQ.
- Earnings quality caveat is now extreme. Non-op was a KRW1.5trn *loss* despite KRW6.6trn of investment-asset valuation gains, because of a KRW8.4trn derivatives loss on exchangeable-bond exchange rights. Net margin (46%) is now *below* OPM (58%) — the reverse of last quarter's flattery, but the KRW6.6trn valuation gain inside the P&L remains a quality issue.
- Shareholder returns U-turned. After explicitly ruling out additional returns last quarter, the board approved a KRW1,500/share special dividend (year-end DPS KRW1,875; FY2025 total KRW3,000/share, ~KRW2.1trn payout) plus retirement of 15mn treasury shares (2.1% of shares, ~KRW12.2trn value). ADR issuance is being "looked into," nothing finalized.
- Balance sheet strengthened further: cash KRW34.9trn (+20.8trn vs end-2024); borrowings KRW22.2trn (-0.4trn); D/E 18% (from 24%); net cash position maintained and enlarged.
- Q1 2026 guide: DRAM bits flat QoQ (supply-constrained, not demand); NAND bits down "somewhat" on Q4 base effect. 2026 demand growth: DRAM >20%, NAND 19%; server shipments +high-teens.
What management is focused on- Price supercycle as the quarter's engine: "broad-based price increases," "severe supply-demand imbalance," memory "selling out as soon as it is produced," and company inventory projected to fall further in H2 2026. The sold-out narrative from last quarter is now showing up as +mid-20%/+low-30% ASPs.
- Inference architecture as the structural story: KV-cache offload and GPU-centric IO are making the SSD "a central part in the compute pipeline" rather than a CPU peripheral — the eSSD thesis now extends to ultra-high-IOPS SSDs, 245TB next-gen products, and HBF (an HBM-extension technology) for the first time.
- HBM4 defense under pressure: the first Q&A question was about "noise" on HBM4 progress; management insists mass production of customer-requested volumes is underway, targets HBM4 yields comparable to 12-high HBM3E via Advanced MR-MUF on 1bnm, and aims for "overwhelming market share" — while conceding it "cannot meet HBM demand 100%, so some competition is expected to enter."
- LTAs as the new market structure: today's LTAs are described as "strong mutual commitments," not loose volume intents; customers want multi-year deals but capacity can't accommodate all requests.
- Capacity acceleration continues: M15X ramping earlier than planned (adding 1bnm capacity for HBM), 1cnm and 321-layer migration accelerating, Yongin Phase 1, Cheongju P&T7 (new mention), and Indiana packaging all proceeding. 2026 capex "increase considerably," with a new soft anchor: capex/revenue around mid-30%.
- New strategic vehicle: a US-based "AI Co." for AI solution investment/commercialization — sized as "not large relative to cash generation," explicitly excluded from capex and FCF.
- Portfolio broadening beyond HBM: SOCAMM2 and GDDR7 named as AI-memory portfolio additions; 256GB DDR5 RDIMM (1bnm 32Gb die) highlighted.
Key numbers and quarter mechanics- P&L: revenue 32.8trn (+34% QoQ, +66% YoY); OP 19.2trn (58% OPM, +11pp QoQ); D&A 3.6trn (flat QoQ); EBITDA 22.7trn (69%); non-op -1.5trn (valuation gains +6.6trn, derivatives loss -8.4trn on EB exchange rights); pre-tax 17.7trn; net 15.2trn (46%).
- FY2025: revenue 97.1trn (+47% YoY); OP 47.2trn (+101% YoY); HBM revenue more than doubled YoY; record annual DRAM and NAND revenue.
- Volumes/ASP: DRAM bits +low-single QoQ; DRAM ASP +mid-20% QoQ; high-density DDR5 modules +~50% QoQ (vs doubling in each of the prior two quarters — growth rate decelerating off a bigger base). NAND bits ~+10% QoQ (beat); NAND ASP +low-30% QoQ (second consecutive double-digit-plus increase).
- Q1 2026 guide: DRAM bits flat QoQ; NAND bits down on base effect. Framed as supply-limited; Q1 seasonality acknowledged but "customer demand remains strong."
- 2026 outlook (management): DRAM demand >20%, NAND 19% (NAND nudged up from "high-teens"); server shipments +high-teens; PC/mobile shipments and content-per-device growth to slow on price increases — a new demand-destruction acknowledgment.
- Balance sheet: cash 34.9trn; borrowings 22.2trn; net cash ~12.7trn (from 3.8trn); D/E 18%.
- Capex: 2026 to "increase considerably" YoY; management endorsed a mid-30% capex/revenue ratio as maintainable — the first quantitative frame after five quarters without a won figure. Still no absolute number.
- Disclosure gaps persist: no DRAM/NAND revenue split, no HBM share of DRAM, no segment margins, no Solidigm mention, no absolute capex figure, no LTA terms.
Product and launch scorecard- HBM4: in mass production of "volumes requested by customers," but the quarter's most-pressed product. Management addressed progress "noise" head-on: on schedule per customer-agreed timeline, 1bnm-based, Advanced MR-MUF targeting HBM3E-12Hi-comparable yields, goal of "overwhelming market share." The explicit concession that unmet demand leaves room for competitors is new and notable. No HBM4 revenue, mix, or customer-count disclosure.
- HBM3E 12-high: still the stated driver of DRAM shipment growth; FY2025 HBM revenue more than doubled.
- Custom HBM / HBM4E: "active" technology discussions with key customers; "one-team" framing with partner companies (unnamed — likely the foundry/packaging partner, but not stated in transcript).
- 1cnm DRAM: "full-scale mass production" of 1cnm DDR5 commenced — consistent with last quarter's "already ongoing smoothly"; migration accelerating; no update on the >50%-of-Korean-capacity-by-end-2026 target trajectory.
- 256GB DDR5 RDIMM: developed on 1bnm 32Gb die — claimed industry-highest density; no shipment/revenue data.
- SOCAMM2 and GDDR7: newly named as AI-memory portfolio expansion; no timing or customer detail.
- 321-layer NAND / 245TB eSSD: 321-layer transition is the competitiveness lever; next-gen 245TB product in development targeting ultra-high-density AI-inference storage; no update on the >50%-of-NAND-bits-by-end-2026 target.
- HBF (high-bandwidth flash): first mention — "further develop" language only; an R&D option, not a product.
- eSSD: demand described as structural (KV-cache offload, GPU-centric IO); ultra-high-IOPS SSD in preparation. No shipment growth figure this quarter (vs +double digits last quarter).
- Dropped again: LPDDR server module commitment (now two quarters overdue), Solidigm, OpenAI LOI conversion status, DDR4 phase-out.
Sell-side read-through- Peter Lee (Citi): opened with the HBM4 "noise" question — the market is actively questioning HBM4 execution. Got a detailed defense (1bnm, MR-MUF yield parity target, overwhelming-share ambition) but also the admission that demand exceeds supply and "some competition is expected to enter." No volumes or qualification breadth disclosed.
- Hyung-Keun Ryu (Daishin): LTA structure — management refused deal specifics but confirmed a qualitative shift from "loose" volume intents to binding mutual commitments, and that customer demand for multi-year contracts exceeds what capacity can accommodate.
- Dong-Hee Han (SK Securities): pull-in vs real demand — management's strongest rejection yet of the pull-in thesis: customer inventories falling across server, PC, and mobile; server customers building sets immediately upon securing volume; company DRAM inventory fell QoQ in Q4 and is projected to fall further in H2 2026; NAND inventory weeks now "nearly matching" DRAM's.
- Simon Woo (BofA): capacity allocation — got the M15X 1bnm-for-HBM detail and a "grow together with customers" framing; no allocation criteria disclosed.
- Sunwoo Kim (Meritz): returns durability and ADRs — management said additional returns will be reviewed based on performance/cash flow (not committed as recurring), and confirmed ADR issuance is under study with nothing finalized.
- Sung Kyu Kim (Daiwa): AI storage — produced the call's most expansive NAND answer: SSD as central to the compute pipeline, ultra-high-IOPS SSDs, 245TB, HBF. Strategy-rich, number-free.
- Rok-ho Kim (Hana): price-driven demand destruction — management conceded PC/mobile customers are revising shipment plans and reviewing spec downgrades on lower-tier products, but argued on-device AI replacement demand prevents broader contraction. The clearest acknowledgment yet that the price spike has a cost.
- Chae Min-sook (Korea Investment): the new AI Co. — got the US location, the "not large" sizing claim, and sequential-investment structure; no mandate size, targets, or P&L treatment beyond the FCF exclusion.
- Ryu Young-ho (NH): capex — extracted the mid-30% capex/revenue anchor, the first quantitative capex frame in five quarters, plus confirmation AI Co. investment sits outside capex/FCF. Still no won figure.
- Soo-rim Lee (DS): 100% US tariff threat — a pure deferral: "monitor the discussions between the government and communicate at a later date." No commitment on US fab expansion beyond Indiana packaging.
- Notable absences: no question on the KRW6.6trn valuation gain or the KRW8.4trn derivatives loss; no HBM4 pricing/2026 contract follow-up; no OpenAI LOI conversion question; no Solidigm; no China/VEU; no challenge on the still-missing absolute capex number.
Management credibility- Positive: the Q4 guide was beaten where it was beatable. NAND bits ~+10% vs +low-single guided; DRAM bits +low-single as guided. The "inventory-normalization-limited, not demand-limited" framing from last quarter proved correct — ASPs did the work.
- Positive: the net-cash and financial-soundness commitments were delivered and exceeded — D/E 18%, net cash ~12.7trn, and the "consider returns before policy expiry if meaningful FCF" clause was invoked exactly as pre-committed. The reversal on additional returns is a kept conditional promise, not a broken one — though it does reverse an explicit "not at this time" from three months ago.
- Positive: HBM4 mass production is confirmed as underway for customer-requested volumes, consistent with the September readiness claim. But the need to rebut "noise" in the very first question, and the concession that competitors will enter unmet demand, tempers the "overwhelming share" rhetoric.
- Negative: earnings quality is now a structural issue. KRW6.6trn of valuation gains in one quarter (double last quarter's 3.3trn), partially masked by an KRW8.4trn derivatives loss. Reported net of 15.2trn contains large non-operating swings in both directions; no analyst asked about either item.
- Negative: capex opacity persists in absolute terms. The mid-30% ratio is a real concession, but with revenue growing 66% YoY, a ratio anchor permits very large absolute increases; five quarters without a won figure.
- Negative: tariff exposure drew a non-answer. A 100% tariff scenario is material for a Korea-based manufacturer with only packaging in the US; "monitor and communicate later" is the weakest response on the call.
- Watch: PC/mobile demand destruction was conceded for the first time — management's "no broader contraction" judgment is now a falsifiable claim.
- Watch: the LPDDR server module promise is two quarters overdue and unmentioned.
What changed versus the prior quarter- From volume-led beat to price-led beat: Q3 was bits + ASP; Q4 was ASP (+mid-20% DRAM, +low-30% NAND) with DRAM bits only +low-single. Pricing is now the entire incremental margin story (OPM 47% → 58%).
- Shareholder returns reversed: from "ruled out in year one" to a KRW2.1trn total dividend plus a ~KRW12.2trn treasury retirement, plus ADR study disclosed. The largest capital-return shift in the company's recent history, three months after a refusal.
- HBM4 narrative shifted from unchallenged leadership to active defense: last quarter's story was contracting completion; this quarter opened with progress "noise," a yield-parity *target* (not achievement), and an admission of competitor entry into unmet demand.
- Demand-destruction language appeared: PC/mobile shipment plan revisions and spec-downgrade reviews were acknowledged for the first time; 2026 PC/mobile content growth guided slower. The super-cycle story now has an admitted edge.
- Capex framing quantified (partially): from "far outpace" with no anchor to "increase considerably" with a mid-30% capex/revenue ratio endorsed.
- Non-op swung from +3.4trn to -1.5trn, but with valuation gains doubling to 6.6trn and a new 8.4trn EB-derivatives loss — earnings quality moved from flattering to noisy.
- New disclosures: AI Co. (US-based investment vehicle); HBF technology; SOCAMM2/GDDR7 portfolio; 245TB eSSD; Cheongju P&T7; NAND inventory weeks near DRAM parity; company inventory to fall further in H2 2026; tariff non-answer.
- Dropped threads: OpenAI LOI status, LPDDR server module, Solidigm, 1c/321-layer 50% target progress, China/VEU — none addressed.
Bull case- Pricing power is proving the sold-out thesis: +mid-20% DRAM and +low-30% NAND ASPs in one quarter, with inventories at customer and company level still falling and management projecting further inventory declines into H2 2026. OPM 58% and 69% EBITDA margin with D&A flat.
- Visibility keeps extending: binding-style LTAs replacing loose volume intents, customers unable to secure volume, server demand +high-teens in 2026, DRAM/NAND demand >20%/19% — and supply additions physically constrained by space and migration.
- HBM4 is in mass production with yield parity to HBM3E 12-high targeted via proven MR-MUF; custom HBM engagement deepens lock-in; FY2025 HBM revenue already doubled.
- NAND has fully inflected: two consecutive quarters of double-digit-plus ASP gains, record FY2025 NAND revenue, and a credible architectural story (KV-cache offload, GPU-centric IO) plus a product roadmap (245TB, ultra-high-IOPS, HBF).
- Capital returns are now real and large: ~KRW14.3trn of combined dividends and cancellation value, net cash ~12.7trn, D/E 18% — and management signaled willingness to do more as cash builds.
- Capex is finally frameable: mid-30% of revenue against 66% revenue growth implies heavy investment funded entirely from operations.
Bear case- The quarter was a price spike, and price spikes cut both ways. DRAM bits grew only low-single; if conventional DRAM pricing normalizes, the +11pp OPM gain reverses just as fast. Management itself conceded PC/mobile customers are cutting shipment plans and downgrading specs.
- HBM4 execution is now openly questioned. Progress "noise" was significant enough to lead the Q&A; yield parity is a target, not a result; and management conceded competitors will enter demand it cannot serve — the first crack in the share-dominance narrative.
- Earnings quality is deteriorating in both directions: KRW6.6trn valuation gains and KRW8.4trn derivatives losses in a single quarter make reported net nearly uninformative; core profitability must be reconstructed, and no analyst pressed on it.
- Tariff risk is unpriced and unanswered: a 100% US semiconductor tariff threat met with "monitor and communicate later," with no US fab commitment beyond packaging.
- Capex is escalating into the tightest part of the cycle — M15X early ramp, Yongin, Cheongju P&T7, Indiana, plus a new AI Co. investment vehicle — with still no absolute won figure and a ratio anchor that permits very large absolute growth.
- The returns reversal, while shareholder-friendly, signals management sees limited better uses for ~KRW14trn — or is responding to external pressure (the ADR study and "inquired disclosure" hint at governance/market-pressure dynamics not explained on the call).
- Unexamined risks remain unexamined: China, Solidigm, customer concentration, and the OpenAI LOI's conversion all went unmentioned for another quarter.
Next-quarter watchlist- Q1 delivery vs guide: DRAM bits flat, NAND bits down — with ASPs the swing factor; watch whether conventional DRAM contract prices hold the Q4 step-up into Q2 negotiations.
- HBM4 proof points: yield-parity achievement vs target, revenue contribution, qualification breadth beyond the lead customer, and any evidence on whether competitor entry is materializing in the unmet-demand gap.
- PC/mobile demand destruction: whether shipment-plan revisions and spec downgrades broaden, and whether the on-device-AI offset management predicted shows up in content-per-device data.
- Capex specificity: whether the mid-30% ratio converts into a won figure, and the M15X early-ramp contribution to HBM capacity.
- LTA conversion: whether "strong mutual commitment" LTAs produce disclosed volumes, durations, or prepayments — and whether the OpenAI LOI converts to a firm agreement.
- Tariff response: the promised "later date" communication on US fab strategy; any acceleration of Indiana or new US commitments.
- AI Co.: mandate size, investment pace, governance, and any related-party dimensions; confirm it stays outside capex/FCF as claimed.
- Earnings quality: size of valuation gains/losses and EB-derivatives swings in Q1 non-op; whether net margin re-converges toward OPM.
- Migration targets: progress markers toward 1c >50% of Korean DRAM capacity and 321-layer >50% of NAND bits by end-2026 — both due for a status update.
- Overdue items: LPDDR server module (two quarters silent), Solidigm, 245TB eSSD and HBF development milestones, SOCAMM2/GDDR7 customer traction.
- Shareholder returns: whether "continue reviewing" produces a recurring framework or remains episodic; ADR decision.
|
| Oct 28, 2025 | +3.54% | — | Read transcript briefingQuarter in one view- Q3 2025: another record, and the "digestion" guide was beaten again. Revenue KRW24.4trn (+10% QoQ, +39% YoY, all-time high; transcript prints "$24.4 trillion" — clearly a KRW transcription error); OP KRW11.4trn (+24% QoQ, +62% YoY, first-ever quarter above KRW10trn; OPM 47%, +5pp QoQ, +7pp YoY); D&A KRW3.6trn; EBITDA KRW14.9trn (61% margin); pre-tax KRW14.8trn; net KRW12.6trn (52% margin, record).
- DRAM beat, NAND guided down as expected: DRAM bits +high-single QoQ vs +low-to-mid-single guided; DRAM ASP +mid-single QoQ with "strong ASP growth for conventional DRAM." NAND bits -mid-single QoQ (vs "limited growth" guide — consistent, off Q2's high base) but NAND ASP +low-teen QoQ on price recovery and eSSD mix — the first NAND ASP increase in at least three quarters.
- The pull-in hangover never came. Management admits it expected "more moderate demand conditions in Q3" from H1 preemptive purchases; instead AI server demand spiked. Customer and company inventories both fell; DDR5 is shipping "straight out of production."
- Balance sheet flipped to net cash: cash KRW27.9trn (+10.9trn QoQ); debt KRW24.1trn (+2.2trn); net cash KRW3.8trn (from net debt KRW4.9trn); D/E 24% (-1pp).
- Earnings quality caveat returned: non-op +KRW3.4trn, of which KRW3.3trn is investment-asset valuation gains (Kioxia-style flattery is back) plus KRW0.21trn FX gains. Net margin (52%) now exceeds OPM (47%) — reported net overstates operations again.
- The headline strategic disclosure: 2026 is sold out. HBM 2026 supply discussions with key customers completed at pricing that "sustains current profitability"; DRAM and NAND 2026 capacity "essentially sold out"; HBM tight "into 2027"; customers issuing pre-purchase POs and seeking multiyear LTAs for conventional memory.
- Q4 guide: DRAM and NAND bits both +low-single QoQ, framed as inventory-normalization-limited, not demand-limited.
What management is focused on- The "super cycle is different this time" argument: broader application base (AI layered on existing demand), inference spreading workloads to general servers and edge, and a structural supply constraint — HBM's wafer intensity caps industry DRAM supply growth. Management explicitly predicts a "long drawn-out memory super cycle" and server set shipments +high-teens % in 2026.
- Inference as the demand unifier: KV-cache offload (HBM → conventional DRAM → SSD) and RAG/vector databases are the stated mechanisms linking AI inference to DDR5 and eSSD demand. This is the prior quarter's 2–3-year NAND thesis pulled forward into a present-tense demand driver.
- 2026 contracting as the proof point: completed HBM supply plans, sold-out DRAM/NAND capacity, pre-purchase POs, and LTA interest are presented as evidence the cycle has shifted to "order-first, produce-later" with greater predictability.
- Capacity acceleration: M15X cleanroom opened ahead of schedule, equipment installation begun, HBM ramp contribution from 2026; Yongin Fab 1 schedule being pulled up from the original May 2027 cleanroom readiness; Indiana packaging plant in preparation. 2026 capex "will far outpace" 2025 — still no won figure.
- HBM4 execution: development and mass-production preparation completed in September (meeting the "within 2025" commitment); shipments start Q4 2025; expansion through 2026; "highest speed in the industry" claim with 2,048 I/O (2x HBM3E).
- Defending the HBM-centric capacity mix: pressed on conventional DRAM margins approaching HBM's, management said HBM profitability "remains high" and it will not shift mix on "short-lived" price moves — long-term volume agreements take precedence.
- Shareholder returns deferred: net cash achieved, but management explicitly ruled out additional returns in year one of the three-year policy, arguing reinvestment is the best use of cash given rising capex needs.
Key numbers and quarter mechanics- P&L: revenue 24.4trn (+10% QoQ, +39% YoY); OP 11.4trn (47% OPM, +5pp QoQ); D&A 3.6trn; EBITDA 14.9trn (61%); non-op +3.4trn (FX +0.21trn, valuation gains +3.3trn); pre-tax 14.8trn; net 12.6trn (52%).
- Volumes/ASP: DRAM bits +high-single QoQ (beat vs +low-to-mid-single); DRAM ASP +mid-single QoQ (conventional DRAM strength — a mix shift versus Q2's dilution). NAND bits -mid-single QoQ; NAND ASP +low-teen QoQ (price recovery + eSSD mix premium). eSSD shipments +double digits QoQ.
- Mix datapoint: >128GB DDR5 module shipments doubled QoQ for the second consecutive quarter — cited as evidence of HPC-driven conventional DRAM demand.
- Q4 2025 guide: DRAM and NAND bits both +low-single QoQ; constraint is normalized inventory, not demand.
- 2026 market outlook (management): DRAM demand growth >20% (from high-teens in 2025); NAND bit growth high-teens (from mid-teens); HBM market CAGR >30% for five years "even by a conservative estimate"; server set shipments +high-teens in 2026.
- Balance sheet: cash 27.9trn (+10.9trn); debt 24.1trn (+2.2trn); net cash 3.8trn; D/E 24%. Debt rose even as cash surged — worth checking what the +2.2trn funded.
- Capex: 2026 "will far outpace" 2025; M15X equipment install underway; Yongin and Indiana infrastructure spending growing. Fourth consecutive quarter without a won figure.
- Disclosure gaps persist: no DRAM/NAND revenue split, no HBM share of DRAM, no segment margins, no Solidigm mention, no capex number.
Product and launch scorecard- HBM4: the dated commitment landed. Development and mass-production prep completed September 2025 (vs "within 2025" promised); Q4 shipments starting; meets "highest speed in the industry" customer requirements above JEDEC spec; 2026 pricing set at levels sustaining current profitability — the pass-through commitment from last quarter is claimed as delivered in contracting. Volume/mix details withheld.
- HBM3E 12-high: still the volume driver of the DRAM beat; no new issues disclosed.
- 2026 HBM: contracted. Supply plans finalized with major customers after delays from changed performance requirements; HBM "selling out since 2023," tight into 2027. Custom HBM to grow from HBM4E, co-designed with customers' GPU/ASIC programs.
- OpenAI LOI: disclosed a letter of intent for "large-scale DRAM supply" — new, unquantified, and an LOI rather than a contract; treat as directional.
- 1c-nano DRAM: ahead of the prior timeline. Mass production "already ongoing smoothly" (prior guide: conversion starts H2 2025, full-scale 2026); full lineup across server/mobile/graphics; >50% of conventional DRAM capacity in Korea by end-2026 — a new, falsifiable mix target.
- 321-layer NAND: TLC and QLC expansion planned for 2026; 321-layer to exceed 50% of NAND bit production by end-2026 — another new dated target. NAND strategy remains profitability-first via migration, not capacity adds.
- eSSD: double-digit shipment growth; demand thesis now rests on HDD shortages pushing hyperscalers to QLC eSSD plus RAG/vector-DB and KV-cache offload — more concrete than last quarter's 2–3-year framing.
- Conventional DRAM: DDR5 effectively sold through (ship-direct-from-production); LPDDR5 seasonal recovery; no update on the LPDDR server module "within this year" commitment — now due and unaddressed.
Sell-side read-through- Sunwoo Kim (Meritz): 2026 HBM contract details — got the completion confirmation, the profitability-sustaining pricing claim, and "tight into 2027," but no volumes, customer count, or terms. Also revealed negotiations took longer than expected due to changed HBM4 performance requirements — a mild negative buried in a positive answer.
- Kitae Kim (Ssangyong; operator misattributed to Samsung Securities' Jongwook Lee): above-JEDEC HBM4/HBM4E requirements — management claims it sampled upgraded-spec products "faster than anyone" and is in mass production; implicitly confirms spec escalation is a recurring feature (HBM4E requirements will also rise).
- Simon Woo (BofA): cycle analogy and inventory vs the cloud-era boom — drew the call's most quotable claim: a structurally supply-constrained "long drawn-out super cycle." This is the assumption most exposed to an AI capex pause.
- Ryu Young-ho (NH): eSSD structural-demand rationale — got the fullest articulation yet (HDD substitution, RAG/vector DB, KV-cache offload). Notably, last quarter's inventory-risk questioner did not re-press the pull-in issue; the Q2 concern looks resolved in management's favor.
- Chae Min-sook (Korea Investment): order-first market structure — extracted the "2026 DRAM and NAND essentially sold out" and pre-purchase PO/LTA disclosures, the quarter's most important demand evidence.
- Ricky Seo (HSBC): capex magnitude — got "far outpace" for 2026 but no figure again; fourth straight quarter of capex opacity.
- Bo-Young Choi (Kyobo): migration timelines — extracted two hard targets: 1c >50% of Korean conventional DRAM capacity and 321-layer >50% of NAND bits, both by end-2026.
- Dong-Hee Han (SK Securities): inventory — got "extremely low" DRAM inventory and DDR5 ship-direct disclosure; no quantification.
- Nicolas Gaudois (UBS): M15X full-equipment by end-2026 and Yongin pull-in — management confirmed Yongin schedule is being pulled up but gave no revised date and did not commit to the M15X end-2026 completion framing.
- SK Kim (Daiwa): HBM demand and customer broadening — got the >30% five-year HBM CAGR, the OpenAI LOI, and "primary supplier of not only GPU but ASIC" positioning.
- Jay Kwon (J.P. Morgan): conventional DRAM margin vs HBM — management conceded the gap has narrowed and DRAM margin "could rise closer to HBM" in 2026, but ruled out mix shifts. A direct answer that also implies HBM's margin premium is compressing at the edges.
- Peter Lee (Citi): shareholder returns — a clear "not at this time," justified by rising capex; the net-cash milestone did not loosen the payout stance.
- Notable absences: no question on the KRW3.3trn valuation gains inflating net income; no Solidigm question; no China/VEU or China-GPU follow-up (last quarter's H20-type upside went unaddressed); no tariff question despite it driving the H1 narrative; no challenge on the missing capex figure beyond HSBC's attempt.
Management credibility- Positive: the Q3 "digestion" guide was beaten, and management admitted its own caution was wrong. It expected moderation from H1 pull-ins; demand spiked instead. The prior quarter's contested "no significant inventory build" claim is now supported by falling customer and company inventories — the bears' Q2 thesis did not materialize.
- Positive: HBM4 mass-production prep completed in September, on the "within 2025" commitment, with Q4 shipments starting — the dated, falsifiable promise landed again.
- Positive: 1c mass production is underway — consistent with, and arguably ahead of, the "H2 2025 start" disclosure from last quarter.
- Positive: 2026 HBM contracting completed — resolves last quarter's "visibility secured vs negotiations proceeding" tension in the direction of the stronger scripted claim, and pricing is asserted to sustain current profitability.
- Negative: earnings quality regressed. KRW3.3trn of valuation gains drove non-op to +3.4trn; net margin (52%) exceeds OPM (47%). After last quarter's clean P&L, the Kioxia-style flattery is back, and no analyst asked about it.
- Negative: capex opacity is now four quarters old. "Far outpace" is directionally informative but, with M15X, Yongin, and Indiana all ramping, the absence of any figure is increasingly a choice, not an oversight.
- Negative: NAND ASP recovery (+low-teen) is real but follows -20% and -high-single declines — cumulative realized pricing is still well below early-2025 levels; management did not frame it that way.
- Watch: the "sold out through 2026" and "super cycle" language is the strongest cyclical claim management has ever made on an earnings call. It is now the benchmark against which any 2026 demand wobble will be judged.
- Watch: LPDDR server module supply "within this year" — due this quarter and not mentioned.
What changed versus the prior quarter- From "deliberate deceleration" to another beat: Q3 was guided as digestion (DRAM low-to-mid-single, NAND limited); DRAM delivered +high-single and ASPs rose in both segments. The pull-in payback thesis is dead for now.
- NAND inflected: ASP swung from -high-single to +low-teen — the first increase in three quarters — and the eSSD thesis moved from a 2–3-year option to a present demand driver with named mechanisms (HDD substitution, RAG, KV-cache offload).
- 2026 HBM: from "visibility secured, negotiations open" to "contracted at profitability-sustaining prices." The central open item from last quarter closed favorably.
- Scope of sold-out expanded: last quarter only HBM had 2026 visibility; now all DRAM and NAND 2026 capacity is "essentially sold out," with pre-purchase POs and LTA requests for conventional memory — a structural change in customer behavior.
- Capex escalated again: from "2025 above plan, HBM-majority" to "2026 will far outpace 2025," with M15X opened early and Yongin being pulled up. The discipline language remains, but the trajectory is now two consecutive years of increases.
- Balance sheet: net debt 4.9trn → net cash 3.8trn — an ~8.7trn swing in one quarter; deleveraging complete, and the debate has moved to shareholder returns (rejected for now).
- Earnings quality reversed: non-op swung from -0.49trn to +3.4trn on valuation gains; net again flatters operations.
- HBM4: from samples-in-optimization to production-ready with Q4 shipments — and the margin pass-through claim moved from commitment to contracted-reality (per management).
- New disclosures: OpenAI LOI; 1c >50% of Korean conventional DRAM capacity by end-2026; 321-layer >50% of NAND bits by end-2026; HBM tight into 2027; five-year HBM CAGR >30%; server sets +high-teens in 2026; Yongin schedule pull-up.
- Dropped threads: no China GPU export update, no Solidigm, no LPDDR server module status, no DDR4 phase-out commentary.
Bull case- Two consecutive all-time-high quarters with accelerating margins — OPM 47% (+5pp QoQ) on record revenue, driven by HBM, high-end DRAM, and eSSD mix.
- 2026 is de-risked in a way no prior cycle has been: HBM contracted at profitability-sustaining prices, all DRAM/NAND capacity essentially sold out, customers issuing pre-purchase POs and seeking multiyear LTAs, HBM tight into 2027. Revenue visibility now extends 5+ quarters.
- The demand story broadened exactly as bulls hoped: inference is pulling conventional DRAM (>128GB DDR5 modules doubling QoQ twice) and NAND (eSSD +double digits, ASP +low-teen) — the AI trade is no longer HBM-only.
- Supply constraint is structural: HBM wafer intensity caps industry DRAM bit growth, supporting the "long super cycle" claim and conventional DRAM pricing (spot rising "almost daily," DDR4/DDR5 shortage premia).
- HBM4 leadership maintained: first to mass-production readiness, above-JEDEC speeds, Q4 shipments, custom HBM from HBM4E deepening customer lock-in; OpenAI LOI signals customer-base broadening into ASIC/AI-lab demand.
- Net cash balance sheet with 61% EBITDA margins funds M15X, Yongin, and Indiana without leverage; capex acceleration is affordable.
- New falsifiable targets set: 1c >50% of Korean DRAM capacity and 321-layer >50% of NAND bits by end-2026 — execution markers that, if hit, confirm cost leadership.
Bear case- Management has fully committed to the super-cycle narrative at the top of its own P&L. "Sold out through 2026," "tight into 2027," "long drawn-out super cycle" — if AI capex pauses or monetization doubts resurface, there is no hedged language left to retreat to, and 2026 capex "far outpacing" 2025 will already be committed.
- Reported earnings overstate operations again: KRW3.3trn of valuation gains in a KRW12.6trn net income; core net is closer to ~KRW9–10trn. The market may re-rate the quality of the beat.
- Capex is escalating on unquantified terms: two straight years of increases, three fabs in motion, no won figure for four quarters. If 2026 HBM terms or conventional pricing disappoint, the fixed-cost base is already being built.
- HBM4 economics remain assertion, not evidence: "pricing at a level that can sustain current profitability" is management's characterization; no margin, mix, or volume data was disclosed, and negotiations ran long because customers escalated performance requirements — a reminder that customers hold leverage too.
- NAND recovery is one quarter old and partly HDD-shortage-driven (a substitution effect that can reverse); cumulative ASP erosion from -20% and -high-single declines is far from recovered.
- The OpenAI LOI is not a contract — large-scale LOIs in AI infrastructure have a wide range of conversion outcomes; sizing it is impossible from the transcript.
- Conventional DRAM margins approaching HBM's (management's own concession) cuts both ways: it validates the super cycle but implies HBM's premium is compressing as competitors qualify HBM4.
- Unexamined risks went quiet, not away: China/VEU, China GPU export policy, Solidigm, and customer concentration received zero airtime this quarter.
Next-quarter watchlist- Q4 delivery vs guide: DRAM/NAND bits +low-single QoQ — with inventories "extremely low" and DDR5 shipping direct from production, any miss would signal supply problems, not demand; any big beat re-tests the pull-in question.
- HBM4 ramp evidence: first HBM4 revenue contribution, qualification breadth beyond the lead customer, and any margin commentary confirming or contradicting the profitability-sustaining pricing claim.
- 2026 contract specifics: volumes, customer count, ASIC vs GPU mix, LTA structures for conventional memory, and whether the OpenAI LOI converts to a firm supply agreement.
- Capex figure: whether the 2026 plan finally gets a won number now that HBM contracting is complete — the stated precondition from prior quarters.
- M15X and Yongin: M15X equipment-install pace and 2026 HBM ramp contribution; a revised Yongin Fab 1 cleanroom date (originally May 2027); Indiana packaging progress.
- Migration targets: 1c share trajectory toward >50% of Korean conventional DRAM capacity by end-2026; 321-layer toward >50% of NAND bits; NAND profitability trend as QLC/TLC eSSD mix rises.
- NAND durability: whether ASP holds after the +low-teen rebound; eSSD shipment growth persistence; any sign HDD substitution is normalizing.
- Earnings quality: size of valuation gains in Q4 non-op; FX swings; whether net margin re-converges below OPM.
- Dropped items due for update: LPDDR server module supply (promised "within this year"), China GPU export-related HBM orders, Solidigm contribution, DDR4 phase-out progress.
- Shareholder returns: any softening of the "not at this time" stance as net cash builds through Q4.
|
| Jul 23, 2025 | +1.02% | — | Read transcript briefingQuarter in one view- Q2 2025: record quarter, both segments beat guidance. Revenue KRW22.2trn (+26% QoQ, +35% YoY, all-time high); OP KRW9.2trn (+24% QoQ, +68% YoY, record; OPM 41%, -1pp QoQ); EBITDA KRW12.6trn (57% margin); D&A KRW3.4trn; pre-tax KRW8.7trn; net KRW7.0trn (31% margin).
- Both volume guides beaten: DRAM bits +mid-20% QoQ vs +low-teen guided; NAND bits +>70% QoQ vs +>20% guided (off Q1's low base). DRAM ASP +low-single QoQ (mix: more lower-priced conventional DRAM); NAND ASP -high-single QoQ (solution-product pricing still weak despite spot recovery).
- Demand driver: AI plus tariff pull-ins. Management attributes the beat to big-tech AI capex plus "preemptive purchasing" — customers shifted from conservative inventory to "securing appropriate levels" amid tariff uncertainty; legacy EOL announcements tightened supply for smaller customers/module makers.
- Balance sheet inflected to strong deleveraging: cash KRW17trn (+2.7trn QoQ); debt KRW21.8trn (-1.5trn); net debt KRW4.9trn (-4.1trn); D/E 25%, net D/E 6% (-4pp/-5pp QoQ).
- Capex discipline broken — deliberately: 2025 total investment "will increase compared to the previous plan," majority of the increment to HBM equipment, justified by newly secured 2026 HBM demand visibility. No won figure given.
- Q3 guide is a deliberate deceleration: DRAM bits +low-to-mid-single QoQ; NAND bit growth "rather limited" — framed as digestion after the Q2 surge, not demand weakness.
What management is focused on- HBM as the organizing principle of the whole company: 2025 HBM revenue doubling reaffirmed; 2026 demand visibility "secured" via customer discussions; capex increase, M15X ramp, and capacity reallocation from conventional DRAM all justified by HBM. Management now explicitly frames HBM demand as "sticky" and leading suppliers as having "greater leverage."
- Managing the pull-in narrative: Q2's beat is candidly attributed partly to tariff-driven preemptive buying, but management argues system-build demand also rose, so customer inventories "did not rise to significant levels" and a sharp H2 correction is unlikely. This is the call's central contested claim.
- Legacy DRAM as an opportunity, not an exit: DDR4/LPDDR4 shortages from supplier EOLs are being served from inventory and China fabs; DDR4 revenue share already down to single-digit (from double-digit last year); mass-market phase-out planned but long-term supply for select customers retained.
- HBM4 profitability defense: cost increases (higher I/O, logic base die) to be reflected in pricing, with "current profitability levels" maintained — a direct answer to margin-dilution concerns.
- NAND still managed for profit, but with a new demand story: prudent investment maintained; the new framing is eSSD as AI inference data cache ("part of the compute cache"), with initial demand expected "within the next two to three years."
- China operations defended: VEU status (October 2023) unchanged; China fabs to keep operating per plan and serve legacy DRAM demand.
Key numbers and quarter mechanics- P&L: revenue 22.2trn (+26% QoQ, +35% YoY); OP 9.2trn (41% OPM, -1pp QoQ); D&A 3.4trn; EBITDA 12.6trn (57%); non-op -0.49trn (FX loss -0.61trn on USD/KRW, other income +0.12trn incl. investment valuation gains); pre-tax 8.7trn; net 7.0trn (31%).
- Volumes/ASP: DRAM bits +mid-20% QoQ (beat vs +low-teen); DRAM ASP +low-single QoQ (mix-driven — higher share of lower-priced conventional DRAM). NAND bits +>70% QoQ (beat vs +>20%); NAND ASP -high-single QoQ (solution-product price declines offsetting spot recovery).
- Q3 2025 guide: DRAM bits +low-to-mid-single QoQ; NAND bits "limited" growth; HBM 2025 doubling unchanged.
- Balance sheet: cash 17.0trn (+2.7trn); debt 21.8trn (-1.5trn); net debt 4.9trn (-4.1trn); D/E 25%; net D/E 6%.
- Capex: 2025 total investment to exceed the prior plan; majority of increment to HBM equipment; final scale contingent on concluding 2026 customer negotiations. No won figure.
- Mix/disclosure notes: no DRAM/NAND revenue split, no HBM share of DRAM, no segment margins disclosed on the call. DDR4 now single-digit % of revenue (from double-digit last year).
- Earnings quality: non-op swung to -0.49trn (FX-driven); the Kioxia-type valuation gains shrank to within +0.12trn "other" — net margin (31%) is now *below* OPM (41%), a cleaner quarter than Q1.
Product and launch scorecard- HBM3E 12-high: delivered the crossover. "Expanded in full scale as planned" — the >50%-of-HBM3E Q2 commitment from two quarters ago was met; "high marks from all customers for performance and supply reliability."
- HBM4: on track, pricing strategy now explicit. March samples (industry-first) now in system-level optimization with partners; "on time for customers' needs"; cost increase to be passed through in pricing to hold profitability. No qualification or ramp date given beyond prior "mass-production prep within 2025."
- 2026 HBM: visibility claimed, negotiations open. "Secured visibility on next year's demand figure" but negotiations on mix/pricing still proceeding; supply details withheld until concluded — the H1-2025 visibility commitment is claimed as met, but only partially evidenced.
- eSSD: >120TB QLC products began active sales expansion in Q2 — the 122TB transition is now commercial. New demand thesis: eSSD as AI inference cache, initial demand in 2–3 years.
- 321-layer NAND: new milestones. UFS 4.1 product developed in May (random read/write +15%/+40% vs prior gen); client and enterprise SSDs on 321-layer to be developed by year-end.
- Conventional DRAM portfolio broadened: >8,000Mbps DDR5, >128GB server modules, LPDDR5X, LPDDR4X for China; LPDDR-based server modules to begin supply "within this year" (new timing commitment); GDDR7 16Gb in supply, 24Gb-based products in preparation.
- 1c-nano DRAM: conversion starts H2 2025, full-scale in 2026 — first timing disclosure; details deferred. No answer on the 15–20% ramp scale HSBC asked about.
- Next-gen DRAM (vertical gate / 3D DRAM): exploratory, sub-10nm; explicitly not near mass production — miniaturization continues in parallel.
- M15X on schedule: opens Q4 2025, full-scale mass production 2026, primarily next-gen HBM; ramp paced to customer volume visibility. Yongin Phase 1 Q2 2027 reaffirmed; Indiana packaging fab referenced.
Sell-side read-through- Young Ho Ryu (NH): pull-in magnitude and H2 inventory risk — got the call's most important admission: Q2's beat was "indeed driven by purchasing demand stronger than expected," customers shifted to inventory-securing, and legacy EOLs tightened supply. Management countered with system-build strength and low supplier inventories. Partially testable against Q3/Q4.
- Jay Kwon (JP Morgan): NAND beat composition — answered: hyperscaler eSSD demand, component pull-ins, and China mobile promotional demand. Notably, the +70% was partly one-off-ish drivers.
- Sun-Woo Kim (Meritz): HBM4 margin dilution — got a clear pass-through pricing commitment ("maintaining current profitability levels"). Falsifiable when HBM4 ramps.
- Jongwook Lee (Samsung Securities): DDR4 — extracted the single-digit revenue share disclosure and the mass-market phase-out plan; management called the price spike "temporary demand concentration," not structural.
- Peter Lee (Citi): capex increase size and split — confirmed the increase and HBM-equipment majority, but no won figure again; final scale tied to 2026 negotiations. Third consecutive quarter without a capex number.
- Ricky Seo (HSBC): M15X utilization and 1c ramp — got "no occasions where we cannot supply due to fab space" (an assertion, not a number) and 1c conversion timing (H2 start, 2026 full-scale); the 15–20% ramp question was deferred.
- Rok-ho Kim (Hana): NAND market stagnation duration — answered with the inference-cache thesis and a 2–3 year demand horizon; no near-term profitability bridge offered.
- Simon Woo (BofA): next-gen DRAM — got a technology roadmap discussion but no mass-production timing; implicitly concedes conventional scaling is nearing limits.
- Unidentified (Dale/대신? participant): China GPU export resumption (H20-type) — management confirmed it was a primary HBM supplier for the product pre-restriction and is "well positioned to respond swiftly" — a potential unpriced HBM upside, still being assessed.
- Min-sook Chae (Korea Investment): HBM competition — answered with culture/customer-engagement argument and custom HBM/PIM roadmap; no share or pricing specifics.
- Notable absences: no question on NAND ASP's continued decline (-high-single after Q1's -20%); no HBM share-of-DRAM or HBM margin question; no Solidigm question; no tariff-quantification follow-up despite pull-ins being central to the beat; no question on the OPM decline QoQ.
Management credibility- Positive: two consecutive guidance beats with clean mechanics. Q1 DRAM beat, then Q2 both-segment beat (+mid-20% vs +low-teen; +70% vs +>20%) — guidance is proving conservative, and management disclosed the pull-in component rather than hiding it.
- Positive: the 12-high crossover commitment landed. The dated, falsifiable Q2 promise from prior calls was met "as planned" — the HBM execution streak continues.
- Positive: earnings quality improved. Non-op turned negative (-0.49trn) and net margin (31%) sits below OPM (41%) — no Kioxia-style flattery this quarter; the cleanest P&L read in three quarters.
- Positive: capex opacity partially resolved. After two quarters of deflecting UBS/Citi, management confirmed the 2025 increase and its HBM-equipment focus, with a coherent justification (2026 visibility). Still no won figure, but the direction question is answered.
- Negative: "2026 visibility secured" vs "negotiations proceeding" is tension. The presentation claims secured demand visibility; the Q&A says mix/pricing negotiations are ongoing and supply details can't be given. Both can be true (volumes visible, terms open), but the stronger framing was in the scripted remarks.
- Negative: NAND ASP fell again (-high-single) after Q1's -20%. The "spot prices rebounding" narrative from last quarter did not translate into solution-product pricing; no analyst pressed on cumulative realized-price erosion.
- Negative: the H2 no-correction claim rests on assertion. "Customer inventories did not rise to significant levels" is given without data, one quarter after the same claim preceded a pull-in-driven beat. The Q3 guide (low-single DRAM, flat NAND) is consistent with digestion — which cuts both ways.
- Watch: HBM4 pricing pass-through is now an explicit commitment; any HBM4 margin dilution at ramp would be a direct credibility hit.
What changed versus the prior quarter- From beat-on-DRAM-only to beat-everything: Q1 was a DRAM beat with a NAND ASP collapse; Q2 beat on both segments' volumes, though NAND ASP fell again (-high-single vs -20% — the decline moderated).
- Tariff pull-ins moved from "limited" to material: Q1's "limited pull-in" became Q2's admitted primary beat driver — the prior quarter's downplaying now looks understated in hindsight.
- Capex stance flipped: from "slight increase" discipline language to a confirmed above-plan increase, majority HBM equipment — the UBS question from last quarter is now answered affirmatively.
- Balance sheet pivoted from pause to aggressive deleveraging: net debt -4.1trn to 4.9trn (net D/E 6%) after Q1's Solidigm-driven uptick.
- 2026 HBM visibility: claimed secured (the H1 2025 deadline met, per management) — but contracting details pushed out, tied to capex finalization.
- Legacy DRAM reframed: from exit-progress story to shortage/opportunity story — DDR4 at single-digit revenue share, China fabs positioned as legacy supply assets, EOL-driven tightness acknowledged as beneficial.
- NAND narrative upgraded from supply-discipline to AI-attached demand: the eSSD-as-inference-cache thesis (2–3 year horizon) is new; >120TB QLC eSSD moved from transition to active sales expansion.
- Earnings quality normalized: non-op swung from +1.86trn (Kioxia-flattered) to -0.49trn (FX loss) — reported net now understates rather than overstates operations.
- New disclosures: 1c conversion timing (H2 2025 start, 2026 full-scale); LPDDR server modules within 2025; GDDR7 24Gb in preparation; 321-layer UFS 4.1 specs; 3D DRAM/vertical-gate exploration.
Bull case- Record revenue and OP with both segments beating — the "mild correction" thesis is now fully vindicated; the trough lasted one quarter.
- HBM flywheel strengthening: 12-high crossover delivered, HBM4 samples in customer optimization, 2026 demand visibility claimed, capex being proactively deployed, and management articulating why HBM demand is sticky and supplier leverage is rising. China GPU export resumption is potential unplanned upside where SK hynix was the incumbent supplier.
- HBM4 margin defense is explicit: cost pass-through committed — if delivered, the main structural bear argument on HBM4 is neutralized.
- Balance sheet is now a weapon: net D/E 6% with net debt down 4.1trn in one quarter — the company can fund the capex increase, M15X, Yongin, and Indiana from operations.
- Legacy tightness is a free option: DDR4/LPDDR4 shortages from competitor EOLs lift conventional DRAM pricing and utilize China fabs — mix headwind to ASP but volume/margin support.
- Demand broadening: AI agents/reasoning raising memory intensity, sovereign AI as a new driver, general-server replacement cycle, AI PC/smartphone content growth, and a credible 2–3 year NAND inference-cache option.
- Cleanest earnings quality in three quarters — net income now backed by operations, not valuation marks.
Bear case- The beat was pull-in-flattered, by management's own admission. Tariff-driven preemptive buying and EOL-driven legacy scrambles are non-recurring demand; the Q3 guide (DRAM low-single, NAND flat) already embeds the hangover. If tariffs resolve, H2 could see the correction management keeps dismissing.
- NAND pricing remains broken: -high-single ASP after -20% in Q1; the +70% volume was partly promotional/pull-in driven; the recovery story is now pushed to a 2–3 year inference-cache horizon — an admission the near-term NAND market stays weak.
- Capex discipline has cracked: above-plan 2025 spending with the final figure contingent on customer negotiations — if 2026 HBM terms disappoint, the equipment is already ordered. M15X, Yongin, and Indiana are fixed commitments regardless.
- HBM4 economics unproven: pass-through pricing is a commitment, not a fact; logic base die and higher I/O raise costs, and competitors qualifying HBM4 could limit pricing power exactly when SK hynix needs it most.
- "Secured 2026 visibility" is unverifiable and coexists with open mix/pricing negotiations — the gap between the scripted claim and Q&A hedging leaves room for negative surprise when terms finalize.
- Concentration and policy risk persist: the prior quarter's ~60% US-customer exposure was not revisited; China fab operations depend on VEU status holding; China GPU export rules can re-tighten as fast as they loosened.
- OPM slipped 1pp QoQ despite record revenue — mix (more conventional DRAM) diluted margins; if HBM4 ramp costs land before pass-through pricing does, further compression is possible.
Next-quarter watchlist- Q3 delivery vs guide: DRAM bits +low-to-mid-single, NAND "limited" — whether the deceleration is orderly digestion or the start of the pull-in payback; watch customer inventory commentary for cracks in the "no significant build" claim.
- 2026 HBM contracting: conclusion of mix/pricing negotiations with major customers; any disclosure of volumes, customer count, or ASIC exposure; whether the finalized capex figure is disclosed alongside.
- HBM4: mass-production preparation completion "within 2025"; qualification progress from March samples; first evidence on whether pass-through pricing holds (HBM4 pricing/margin commentary).
- NAND ASP trajectory: after -20% then -high-single, whether solution-product pricing stabilizes; >120TB QLC eSSD revenue traction; 321-layer client/enterprise SSD development by year-end.
- Capex: the final 2025 won figure once negotiations conclude; M15X Q4 opening on schedule and 2026 ramp pace; 1c conversion start in H2.
- China GPU export resumption: whether the reported AI-chip shipments to China convert into incremental HBM orders — management flagged swift-response capability as the incumbent supplier.
- Legacy DRAM: DDR4 price-spike durability (management calls it temporary); phase-out execution for mass market; China fab utilization.
- LPDDR server modules: supply start "within this year" — a dated commitment now due.
- Earnings quality: FX swings in non-op; any reappearance of large valuation gains; Solidigm contribution (still undiscussed).
|
| Apr 23, 2025 | +2.05% | — | Read transcript briefingQuarter in one view- Q1 2025: a guided-down quarter that beat on DRAM and held margins. Revenue KRW17.6trn (-11% QoQ, +42% YoY) — second-highest quarter ever after Q4's record; OP KRW7.44trn (OPM 42%, +1pp QoQ despite lower revenue); EBITDA KRW10.77trn (61% margin); D&A KRW3.33trn (up slightly again); pre-tax KRW9.3trn; net KRW8.11trn (46% margin).
- DRAM volume beat, NAND volume as guided: DRAM bits -high-single-digit QoQ vs -low-teen guided (beat, on HBM3E/DDR5 strength and better PC/smartphone sales); DRAM ASP flat QoQ (mix offset conventional price declines). NAND bits -high-teen QoQ (in line with guide); NAND ASP -~20% QoQ — a severe drop, blamed on mix (higher discrete-product sales into restocking), not purely price.
- Non-operating +KRW1.86trn, again one-off-heavy: +KRW0.12trn FX gain and +KRW1.74trn "other" including Kioxia valuation gains — i.e., the Kioxia mark swung from -0.2trn in Q4 to a large positive. Net margin (46%) is again flattered; OPM is the clean read.
- Balance sheet steady despite Solidigm payment: second closing completed, remaining $2bn paid; cash KRW14.3trn (transcript says "billion" — almost certainly a transcription error, +0.2trn QoQ); debt KRW23.3trn (+0.6trn); net debt KRW9.0trn (+0.5trn); D/E 29%, net D/E 11% (both improved slightly QoQ on equity growth).
- Q2 2025 guide is a strong rebound: DRAM bits +low-teen QoQ; NAND bits +>20% QoQ; HBM3E 12-high to exceed half of HBM3E shipments in Q2 (the prior "by H1 2025" commitment, now pinned to Q2).
- New macro overhang: tariffs. Management repeatedly declined to quantify tariff impact; disclosed ~60% of revenue from US-headquartered customers but stressed actual direct US-bound shipments are much lower; some pull-in demand acknowledged but called "limited."
What management is focused on- Tariff uncertainty as the new dominant risk variable: H2 demand visibility explicitly downgraded — "heightened global uncertainties such as tariff policies have increased volatility in the second half demand projections." This replaces last quarter's confident H2-recovery framing with a hedge.
- HBM plan immovability: 2025 HBM revenue "more than double" YoY reaffirmed; sales plans for major customers "unchanged from the levels agreed upon in existing contracts" despite GPU export-restriction speculation; 12-high to be the majority of HBM3E revenue this year and >50% of HBM3E sales in Q2.
- DeepSeek reframed as demand-positive: MLA efficiency lowers AI entry barriers → more participants → more demand for HBM *and* high-capacity server DRAM (96GB DDR5 modules called out with rising Q1 demand); reasoning models (OpenAI O3, DeepSeek R1) extend inference time → more memory. The inference-shift question is now answered with a specific mechanism, not just rhetoric.
- New long-term HBM anchor: ~50% average annual HBM market growth 2024–2028 — a new multi-year quantification.
- NAND as a supply-side story: recovery attributed to industry production cuts and conservative capex rather than demand; "profit-oriented operational decisions" and "prudent approach to capex" maintained; QLC eSSD (61TB→122TB transition, 244TB in development on 321-layer) as the mid-term growth driver, with HDD replacement by big tech flagged as a potential inflection.
- Capex discipline under pressure: "slight increase" vs 2024 reiterated; UBS's question about a possible capex *increase* and M15X equipment pull-in was not confirmed — management restated discipline language and gave no won figure again.
Key numbers and quarter mechanics- P&L: revenue 17.6trn (-11% QoQ, +42% YoY); OP 7.44trn (42%, +1pp QoQ); D&A 3.33trn; EBITDA 10.77trn (61%); non-op +1.86trn (FX +0.12trn, other +1.74trn incl. Kioxia valuation); pre-tax 9.3trn; net 8.11trn (46%).
- Volumes/ASP: DRAM bits -high-single QoQ (beat vs -low-teen guide); DRAM ASP flat QoQ. NAND bits -high-teen QoQ (as guided); NAND ASP -~20% QoQ (mix-driven: higher discrete share into customer restocking).
- Mix: DRAM share of revenue rose 74% → 80% QoQ; DRAM OPM "improved slightly" QoQ; NAND remained profitable ("in the black") despite the ASP drop. No HBM share-of-DRAM figure given this quarter (vs >40% disclosed in Q4).
- Q2 2025 guide: DRAM bits +low-teen QoQ; NAND bits +>20% QoQ; 12-high >50% of HBM3E shipments in Q2.
- Balance sheet/cash: cash ~14.3trn; debt 23.3trn; net debt 9.0trn; D/E 29%; net D/E 11%; $2bn Solidigm second-closing payment made in the quarter.
- Tariff exposure disclosure: ~60% of revenue from US-headquartered customers (audit-report basis); direct US-bound shipment share "not that high"; impact assessment refused until tariff criteria/methods are final.
- HBM market outlook: ~50% CAGR 2024–2028 (new); 2026 supply visibility with major customers to be secured within H1 2025 (reaffirmed, now with "finalize agreements with other customers thereafter").
Product and launch scorecard- HBM4: milestone delivered early. Samples shipped to major customers in March — "first time in the world"; mass-production preparation for HBM4 12-high to complete within 2025 (consistent with prior H2 2025 ramp-readiness). This is the quarter's strongest execution proof point.
- HBM3E 12-high: on plan. Transition "progressing smoothly"; >50% of HBM3E shipments in Q2 reaffirmed; "customer demand remains robust." No platform-timing friction disclosed.
- LPCAMM2: moved from preparation to revenue. Began shipping to certain PC customers in Q1 — last quarter it was "being prepared for 2025." First named on-device AI product to ship.
- SOCAMM (low-power DRAM module for AI servers): new product name on the call; "preparing for timely supply through close cooperation with customers" — no timing or volume specifics.
- eSSD: capacity transition underway. Demand shifting from 61TB (2024) to 122TB (2025); 244TB product in development leveraging 321-layer NAND (introduced last November) — a new forward product disclosure. FY2024 eSSD +300% YoY reiterated.
- High-capacity server DRAM: 96GB DDR5 module demand rising (DeepSeek/inference-driven); 24GB/32GB high-density DRAM transition cited for cost/power competitiveness.
- HBM4E/1c, 16-high HBM4, M15X/Yongin: not re-detailed this quarter; M15X opening Q4 2025 and Yongin Q2 2027 completion reaffirmed (Yongin broke ground in Q1).
Sell-side read-through- Sun-Woo Kim (Meritz): tariff demand impact — got a refusal to quantify plus useful color: customers maintaining planned demand, some short-term pull-in requests, possible pre-tariff consumer rush purchases, AI server impact "relatively limited." Partial answer.
- Simon Woo (BofA): HBM demand/supply changes from customer roadmap shifts and China GPU export restrictions — got the call's most important defense: 2025 contracted volumes unchanged, >100% HBM revenue growth intact, full lineup + flexible capacity. Directly answered, but contract-based (i.e., only as good as the contracts).
- Peter Lee (Citi): DeepSeek and 96GB server DRAM — extracted the most substantive demand-mechanism answer of the call (MLA, reasoning models, inference-time memory intensity). Fully answered.
- Ryu Young-ho (NH): US export exposure — extracted the ~60% US-customer revenue disclosure, the quarter's hardest new number, plus the caveat that direct US shipments are much lower. Impact sizing still refused.
- Giuni Lee (Goldman): pull-in demand and H2 inventory-correction risk — management said Q1 pull-in was limited, concentrated in PC/mobile, customer inventories "did not show significant changes," and a pandemic-style H2 adjustment is unlikely. Testable claims; the H2 risk question was answered with supplier-discipline logic rather than inventory data.
- Nicolas Gaudois (UBS): capex increase/M15X equipment pull-in — not confirmed; management restated "slight increase" and discipline language, again refusing a won figure or 2026 direction. Second consecutive quarter of capex opacity.
- Korea Investment (Mintbook Choi): NAND spot-price rebound sustainability — attributed to preemptive buying + supplier cuts; short-term price recovery "likely to continue," but demand recovery pace explicitly tied to macro uncertainty. Hedged.
- Ricky Seo (HSBC): 2026 HBM outlook — extracted the new ~50% CAGR (2024–28) and reaffirmed H1 2025 visibility on 2026 volumes with major customers. Dated, falsifiable.
- Dong-Hee Han (SK Securities): margin bridge — got the mix explanation (DRAM 74%→80% of revenue, DRAM OPM up slightly, NAND still profitable). No product-level margin numbers given.
- Loho Kim (HANA): eSSD — got the 61TB→122TB transition and 244TB development disclosure.
- SK Kim (Daiwa): DRAM strategy — elicited the claim that SK hynix ranked #1 in DRAM revenue and profitability in Q1 2025 (attributed to a market research report, not company disclosure).
- Notable absences: no question on the NAND ASP -20% collapse's composition or recovery math; no HBM share-of-DRAM update requested; no legacy (DDR4/LPDDR4) exit progress question despite last quarter's single-digit target; no HBM pricing/margin question; no Solidigm integration question despite the $2bn payment.
Management credibility- Positive: DRAM beat the guide they set. -high-single delivered vs -low-teen guided — the Q1 "sharp step-down" proved too conservative on DRAM; the correction-so-far is milder than guided, supporting (not yet proving) the "mild correction" thesis.
- Positive: HBM milestones keep landing. HBM4 samples shipped in Q1 (a milestone not even promised for Q1); 12-high crossover reaffirmed for Q2; >100% 2025 HBM revenue growth reaffirmed under direct questioning about export restrictions.
- Positive: new disclosures under pressure. The ~60% US-customer revenue figure and the 50% HBM CAGR were volunteered in answers — management is still giving anchorable numbers when pressed.
- Negative: NAND ASP -20% is a large adverse surprise with a thin explanation. "Increased sales of discrete products" explains mix, but a 20% sequential ASP drop one quarter after the first ASP decline of the recovery suggests NAND pricing is worse than the "spot prices rebounding" narrative implies. No analyst challenged it.
- Negative: H2 conviction walked back. Last quarter's "H2 recovery" framing is now "increased volatility in second-half demand projections" — the mild-correction thesis's second half is now explicitly conditional on tariffs.
- Negative: capex opacity continues. UBS asked directly about a 2025/2026 capex increase and M15X pull-in; management neither confirmed nor denied and again gave no won figure — third-party speculation left unresolved.
- Mixed: earnings quality again flattered by Kioxia. +1.74trn of non-operating "other" income (largely Kioxia valuation) reverses last quarter's -0.2trn loss — this line is now a material, volatile, non-cash swing factor in reported net income.
- Watch: the pull-in denial is falsifiable. "Limited pull-in, no significant customer inventory change" can be checked against Q2/Q3 bit trends and the +>20% NAND Q2 guide's durability.
What changed versus the prior quarter- The correction proved shallower than guided on DRAM, deeper on NAND price: DRAM bits beat (-high-single vs -low-teen); NAND bits met guide but ASP fell ~20% — far worse than Q4's -mid-single.
- The risk narrative rotated from inventory to tariffs: Q4's concern was supply-chain inventory and seasonality; Q1's is tariff policy, with H2 demand visibility explicitly downgraded and pull-in dynamics acknowledged for the first time.
- HBM4 accelerated from plan to physical samples: March sample shipments (world-first claim) vs Q4's "development completion by H2 2025" — execution running ahead of the stated timeline.
- New long-term demand anchor: ~50% HBM CAGR 2024–2028 — the first multi-year market growth quantification management has given.
- DeepSeek absorbed into the bull thesis: last quarter's generic "inference is demand-positive" is now a specific mechanism (MLA, reasoning models, 96GB DDR5 demand) — the inference-threat debate was not re-litigated.
- On-device AI products started shipping: LPCAMM2 moved from "preparing" to shipping; SOCAMM named for AI servers — the AI PC/server module portfolio is now generating revenue.
- Mix shifted further to DRAM: 80% of revenue (from 74%) — the company is increasingly a DRAM/HBM story with NAND as a managed-for-profit appendage.
- Balance-sheet trajectory paused: net debt rose 0.5trn (Solidigm $2bn payment) after Q4's big deleveraging; ratios still improved slightly on equity growth.
- eSSD roadmap extended: 244TB QLC development disclosed; demand transition 61TB→122TB named — the NAND growth story is now explicitly capacity-migration-led.
Bull case- Margin resilience is the headline: 42% OPM in a revenue-down quarter, with DRAM OPM *improving* — management's "structural transformation" claim has one more quarter of supporting evidence; the P&L is behaving differently from prior downturns.
- HBM moat widening on schedule: world-first HBM4 samples in March, 12-high crossover in Q2, 2025 revenue doubling reaffirmed under export-restriction questioning, 2026 visibility on track for H1 — every HBM commitment continues to be met or beaten.
- Demand drivers multiplying: DeepSeek-style efficient models expand the AI participant base; reasoning models raise inference memory intensity; 96GB DDR5, LPCAMM2, SOCAMM, and 122TB/244TB eSSD broaden the AI-attached portfolio beyond HBM alone.
- Q2 guide implies the trough is already in: DRAM +low-teen and NAND +>20% sequential bit growth would recover most of the Q1 decline in one quarter — consistent with "mild correction."
- NAND supply discipline is real this time: industry cuts are lifting spot prices; SK hynix stayed profitable in NAND through a -20% ASP quarter; the QLC eSSD/HDD-replacement optionality is intact.
- Balance sheet absorbed a $2bn acquisition payment with ratios still improving — self-funding through the cycle remains credible.
Bear case- NAND ASP -20% is a red flag the call glossed over: even mix-driven, that magnitude of realized-price decline signals severe conventional NAND weakness; the Q2 +>20% bit guide into a tariff-uncertain market risks another miss if restocking was pull-in-flattered.
- H2 is now explicitly unknowable: management downgraded its own H2 visibility due to tariffs; if pull-ins did occur (some acknowledged), the H2 inventory-correction risk Goldman's question raised is being dismissed with assertion, not data.
- ~60% US-customer revenue concentration is a large disclosed exposure to US trade/AI policy at the moment policy is most volatile; the "direct shipments are low" defense is unquantified.
- Earnings quality deteriorating in composition: two consecutive quarters of >1.5trn non-operating gains (Wuxi, then Kioxia) mean reported net income is running well ahead of operating reality; Kioxia marks can reverse.
- HBM growth is contract-dependent: "unchanged from contracted levels" cuts both ways — if major customers' GPU roadmaps shift (the premise of BofA's question), the rigidity of annual contracts delays but does not eliminate the adjustment.
- Capex creep risk unresolved: UBS's specific question about a 2025/2026 capex increase and M15X equipment pull-in was deflected; fixed commitments (M15X Q4 2025, Yongin, Indiana) continue regardless of demand.
- No HBM mix disclosure this quarter: the >40%-of-DRAM metric was not updated — a small but notable disclosure regression at a time when HBM is the entire thesis.
Next-quarter watchlist- Q2 delivery vs guide: DRAM bits +low-teen, NAND bits +>20% QoQ — especially whether the NAND rebound materializes or repeats the recent pattern of NAND volume misses; watch NAND ASP direction after the -20% quarter.
- HBM3E 12-high crossover: >50% of HBM3E shipments in Q2 — the dated commitment, now due; any slippage would be the first HBM execution miss.
- 2026 HBM contracting: "visibility with major customers within H1 2025" is due by the next call — confirmation, customer count, and any ASIC-customer detail; watch whether the ~50% CAGR framing survives contact with actual 2026 negotiations.
- Tariff resolution: final scope/criteria for semiconductor tariffs; whether the ~60% US-customer exposure translates into real shipment impact; evidence for or against the "limited pull-in" claim in customer inventory data.
- HBM4 progression: mass-production preparation completion "within the year"; customer qualification feedback on the March samples; TSMC base-die collaboration detail (not mentioned this quarter).
- NAND quality of recovery: spot vs contract price behavior; discrete vs eSSD mix; any update on 122TB ramp and 244TB development; whether HDD-replacement discussions with big tech become orders.
- Capex: any won figure for 2025; confirmation or denial of the M15X equipment pull-in UBS raised; M15X Q4 opening on track.
- Earnings quality: Kioxia valuation swings in non-operating; Solidigm contribution now that the second closing is complete and $2bn is paid.
- Disclosure follow-ups: HBM share of DRAM revenue (absent this quarter); legacy DDR4/LPDDR4 exit pace toward the single-digit target; DRAM #1 market-share claim verification.
|
| Jan 22, 2025 | +0.73% | — | Read transcript briefingQuarter in one view- Q4 2024: another record, with the volume guides delivered this time. Revenue KRW19.77trn (+12% QoQ, +75% YoY) — new record; OP KRW8.08trn (+1.05trn QoQ, OPM 41%, +1pp) — record again; EBITDA KRW11.25trn (57% margin); pre-tax KRW9.58trn; net KRW8.01trn (41% margin); D&A KRW3.17trn (up slightly QoQ, first increase after a declining trend).
- Both volume guides met: DRAM bits +mid-single-digit QoQ (vs +mid-single guide — delivered, HBM3E/server DDR5-led); NAND bits -mid-single-digit QoQ (vs +low-teen guide — a clear miss, blamed on PC/mobile customer inventory adjustment). NAND ASP also fell mid-single-digit QoQ — the first down quarter for NAND ASP in this recovery.
- DRAM ASP +~10% QoQ on mix despite DDR4/LPDDR4 price declines — the bifurcation framework is working as management described.
- HBM crossed the 40% threshold: >40% of DRAM sales in Q4 (vs ~40% guided, 30% in Q3); HBM3E 12-high shipments began as planned. FY2024 HBM sales +4.5x YoY.
- FY2024: revenue KRW66.2trn (above the 2022 record); OP KRW23.5trn (above the 2018 super-cycle); FCF KRW13.9trn; NAND returned to full-year profitability with record annual revenue/OP; eSSD sales +300%+ YoY.
- Non-operating +KRW1.5trn, mostly one-off: ~KRW1.3trn gain on System IC Wuxi JV conversion/stake sale, ~KRW0.6trn FX gain, partially offset by ~KRW0.2trn Kioxia IPO revaluation loss. Net margin (41%) is flattered by these items; operating margin is the clean read.
- Balance sheet inflected to near-net-cash: cash KRW14.2trn (+5.2trn YoY); debt KRW22.7trn (-6.8trn YoY); D/E 31%, net D/E 12% — resolves last quarter's "70%" transcript anomaly decisively.
- Q1 2025 guide is a sharp sequential step-down: DRAM bits -low-teen QoQ, NAND bits -high-teen QoQ — attributed to seasonality and elevated supply-chain inventory in some end markets.
- New shareholder return policy: fixed dividend +25% to KRW1,500/share (~KRW1trn/yr); the 5%-of-FCF variable dividend is dropped in favor of balance-sheet repair; 50% of cumulative 3-year FCF remains the total-return resource pool. Q4 dividend KRW1,305/share (300 fixed + 1,005 variable under the old policy).
What management is focused on- Reframing the downturn as a "mild correction," not 2023: commodity weakness is real (DDR4/LPDDR4, China supply, H1 seasonality) but HBM-focused capex discipline across suppliers and H2 demand recovery (AI PC/phone content, server replacement) mean "this cycle will be different." NAND correction also framed as milder, contingent on general-purpose demand recovery.
- HBM as a contracted, multi-year business: 2025 HBM revenue guided to grow >100% YoY; 2026 volume discussions already underway with visibility on "most of next year's volumes by H1 2025"; long-term contracts explicitly sought given high investment cost and long TAT. ASIC customers called out as a meaningfully expanding demand source — new customer-base language.
- Inference as HBM demand driver, not threat: pushed back directly on the training-to-inference slowdown concern — advanced inference and AGI ambitions require more bandwidth/capacity; national-level AI investment programs cited as incremental demand.
- Legacy exit now quantified: DDR4/LPDDR4 revenue share to fall from ~20% in 2024 to single-digit in 2025 — the first hard number on the accelerated exit.
- Infrastructure buildout with a discipline caveat: M15X opens Q4 2025; first Yongin fab construction starts 2025, opening Q2 2027; Indiana advanced packaging backed by a signed ~KRW660bn CHIPS Act funding agreement. Infrastructure investment rises "significantly," but total capex only "slight increase" vs 2024, with repeated "profitability-first, flexible" language.
- Financial soundness as an explicit policy goal: positive net cash and "appropriate cash levels" are now stated corporate financial targets, and the dividend policy was rewritten around them.
Key numbers and quarter mechanics- P&L: revenue 19.77trn (+12% QoQ, +75% YoY); OP 8.08trn (41%); D&A 3.17trn; EBITDA 11.25trn (57%); non-op +1.5trn; pre-tax 9.58trn; net 8.01trn (41%).
- Volumes/ASP: DRAM bits +mid-single QoQ (guide met); DRAM ASP +~10% QoQ (mix-led; legacy prices fell). NAND bits -mid-single QoQ (miss vs +low-teen guide); NAND ASP -mid-single QoQ.
- HBM: >40% of DRAM revenue in Q4; FY2024 HBM sales +4.5x YoY; 12-high shipments began in Q4 as planned; 12-high to exceed half of HBM3E shipments by H1 2025 (reaffirmed); 2025 HBM revenue guided >100% YoY.
- eSSD: FY2024 sales +300%+ YoY; 61TB and 122TB QLC eSSDs launched at end-2024 (122TB now described as launched, not merely in validation).
- Non-operating detail (from Kyobo Q&A): +~1.3trn System IC Wuxi JV conversion gain; +~0.6trn net FX; -~0.2trn Kioxia mark-to-market; total ~+1.5trn. Roughly 16% of pre-tax profit is one-off/revaluation in nature.
- Balance sheet: cash 14.2trn; debt 22.7trn; net debt ~8.5trn; D/E 31%; net D/E 12%; FY2024 FCF 13.9trn.
- Q1 2025 guide: DRAM bits -low-teen QoQ; NAND bits -high-teen QoQ.
- 2025 demand outlook: PC +low-to-mid-single (AI PC penetration 30–40%, driving 16GB minimum and 24/32GB adoption); smartphones +low-to-mid-single (AI phone penetration ~30%; China consumer subsidies now include smartphones/tablets — new positive); server market +high-single-digit (2017–18 vintage replacement cycle + new DDR5 CPUs); DRAM bit demand +mid-to-high-teen; NAND +low-teen (NAND demand growth revised down from mid-teen).
- Capex: 2025 "slight increase" vs 2024; majority HBM + infrastructure; infrastructure up "significantly"; no aggregate won figure given.
Product and launch scorecard- HBM3E 12-high: delivered on schedule. Q4 shipments began as planned; volume ramp "proceeding smoothly"; >50% of HBM3E shipments by H1 2025 reaffirmed. Last quarter's UBS platform-timing question was not re-raised; no direct demand evidence offered beyond the ramp statement.
- HBM3E 16-high: industry's first developed in November 2024 on advanced MR-MUF — a new milestone disclosed.
- HBM4: timeline and partners now concrete. Development completion and ramp-readiness by H2 2025; supply starts with 12-high; 16-high follows on customer demand with shipments expected H2 2026; TSMC confirmed as the logic base-die foundry ("one team" collaboration) — answering UBS's direct question affirmatively. HBM4 mainstreams from 2026.
- HBM4E process pre-announced: 1c nm to be applied to HBM4E — new forward disclosure.
- 1c nm DRAM: development completed H2 2024, mass-production-ready. 1c DDR5: 9.2Gbps max speed (~28% faster than prior gen), >9% power-efficiency gain; initial ramp target yield already exceeded at development stage; conventional DRAM ramp from H2 2025, paced against demand.
- eSSD: 61TB and 122TB QLC products launched end-2024 — the 122TB moved from "validation for H1 2025 supply" to launched; FY eSSD +300%+ YoY drove NAND's return to profit.
- On-device AI portfolio: LPCAMM2 and ZUFS being prepared for 2025 customer demand — named but without timing/volume specifics.
- Legacy DRAM: exit quantified (~20% → single-digit revenue share); inventory "streamline and clear" language suggests the carried-inventory overhang is being actively worked down, but no size or progress metrics disclosed.
Sell-side read-through- Sun-Woo Kim (Meritz): is this 2023 again? Got the full "different cycle" thesis — HBM-prioritized capex across suppliers, H2 recovery, mild correction for both DRAM and NAND. Confident but assertion-based; no supply math offered.
- Nicolas Gaudois (UBS): HBM4 specifics — extracted the call's hardest disclosures: H2 2025 ramp-readiness, 12-high first, 16-high shipments H2 2026, and TSMC confirmed for the base die. Substantive, fully answered.
- Jay Kwon (JPMorgan): capex flexibility and breakdown — got "slight increase," HBM/infrastructure-weighted, infrastructure up significantly; the request for a won figure or quantified split was not answered. Partial evasion on numbers.
- Dong-Won Kim (KB): legacy reduction — got the quarter's most quotable number: DDR4/LPDDR4 from ~20% of revenue to single-digit in 2025.
- Sung Kyu Kim (Daiwa): inference-threat question — management flatly rejected HBM demand slowdown from the inference shift; AGI, national AI programs, CES "Physical AI" cited. Rhetorical rather than quantified.
- Ryu Young-ho (NH): server DDR5 slowdown risk and hyperscaler inventory — got "customers holding low DDR5 inventory," new-CPU ramp, 2017–18 replacement cycle; even a temporary DDR5 price dip framed as a customer restocking opportunity. Testable claims.
- Peter Lee (Citi): 1c status — got development complete, yield ahead of target, H2 2025 conventional ramp, and the new HBM4E-on-1c disclosure.
- Dong-Hee Han (SK Securities): NAND cuts — no new cut quantum; "maintain current approach, flexibly adjust" until demand improves. The direct "to what extent and for how long" went unanswered.
- Young-Gun Kim (Mirae): 2026 HBM contracting — discussions started; visibility on most 2026 volumes expected by H1 2025. A dated, falsifiable commitment.
- Bo-Young Choi (Kyobo): non-op decomposition — full itemization obtained (Wuxi +1.3trn, FX +0.6trn, Kioxia -0.2trn). Clean answer.
- Giuni Lee (Goldman): Chinese DRAM capability — management acknowledged Chinese DDR5 reports but asserted a "clear wide gap" in quality/performance and cited export restrictions as a constraint on advanced-node progress. No quantification.
- Simon Woo (BofA): conventional DRAM supply — wafer production for conventional DRAM to grow "slightly" in 2025, bit growth aligned with market demand via 1a/1c migration; HBM maximized within capacity.
- Notable absences: no question on the carried legacy inventory's size or clearing progress; no question on the Q1 bit-decline guide's composition (how much is HBM vs conventional); no HBM margin/pricing question despite the 2025 ">100% revenue growth" guide; no NAND ASP question despite the first down quarter.
Management credibility- Positive: Q4 delivery matched the key promises. DRAM bit guide met; HBM >40% of DRAM sales as guided; 12-high shipments began in Q4 as scheduled; 122TB eSSD launched. The Q3 double-miss did not repeat in DRAM.
- Positive: balance-sheet claims are now verified numbers. Net D/E 12% and D/E 31% replace last quarter's garbled "70%" disclosure; debt -6.8trn YoY; FCF 13.9trn supports the self-funding narrative.
- Positive: non-operating items fully itemized under questioning — the Wuxi gain, FX, and Kioxia loss were disclosed with amounts when asked, allowing a clean quality-of-earnings adjustment.
- Negative: NAND guide missed again. +low-teen guided, -mid-single delivered — the second consecutive NAND volume miss, and Q1's -high-teen guide implies the correction deepens before it improves. The Q3 "eSSD-led +low-teen" claim now looks like a clear forecasting failure.
- Negative: capex opacity partially returned. After giving a 2024 range last quarter, management declined to size 2025 capex in won or split infrastructure vs fab quantitatively when directly asked — only "slight increase" and directional language.
- Mixed: the "mild correction" call is the quarter's biggest unfalsifiable claim. It rests on supplier discipline and H2 demand recovery; management has now guided a steep Q1 volume decline while asserting H2 improvement — both halves will be testable within two quarters.
- Mixed: carried-inventory follow-through is vague. "Streamline and clear inventory" and the ~20%→single-digit legacy revenue plan imply progress, but no inventory size, writedown, or clearing-pace disclosure was offered or requested.
- Watch: 2025 HBM revenue >100% growth and 2026 volume visibility by H1 2025 are now on the record — specific, dated, and anchorable for estimates.
What changed versus the prior quarter- Volume execution split: DRAM returned to guide delivery (+mid-single as guided) after Q3's miss; NAND missed badly again (+low-teen guided, -mid-single delivered) and NAND ASP turned negative (-mid-single) for the first time in the recovery.
- The near-term outlook darkened sharply: Q1 2025 guides of DRAM -low-teen and NAND -high-teen QoQ are the weakest sequential guides of the AI upcycle; management now explicitly frames H1 2025 as a correction period with H2 recovery — a new cadence versus prior "gradual improvement" language.
- HBM mix milestone hit: >40% of DRAM revenue (vs 30% in Q3); 2025 HBM revenue guided >100% growth — a new, aggressive quantification beyond last quarter's "price up YoY."
- HBM4 firmed up: TSMC confirmed by name for the base die; 16-high HBM4 shipments dated to H2 2026; HBM4E process (1c) pre-announced. Last quarter's foundry-partner structure is now a named partner.
- Legacy exit quantified and accelerated: ~20% → single-digit revenue share in 2025 — from strategy language to a hard target.
- NAND demand growth revised down: 2025 NAND bit demand now +low-teen (from mid-teen); DRAM demand trimmed to mid-to-high-teen (from high-teen).
- Shareholder returns restructured: fixed dividend +25% to KRW1,500/share, but the 5%-of-FCF variable component eliminated — total near-term payout likely lower despite the fixed increase; net-cash target formalized.
- China subsidy angle added: smartphones/tablets included in China's consumer subsidies — a new demand positive absent from prior commentary.
- US footprint advanced: Indiana packaging agreement completed and ~KRW660bn CHIPS Act funding signed — from plan to executed agreement.
- One-off-heavy quarter: ~KRW1.3trn Wuxi gain means Q4's 41% net margin overstates run-rate profitability; Q3's margin was cleaner.
Bull case- Record OP with the mix engine intact: 41% OPM on +10% DRAM ASP despite falling legacy prices; HBM >40% of DRAM revenue and guided to >100% revenue growth in 2025 — the P&L is increasingly decoupled from the commodity cycle.
- Contracted visibility extending: 2025 HBM volumes/pricing done; 2026 discussions underway with most volumes expected visible by H1 2025; ASIC customers expanding the base beyond GPU platforms.
- Technology leadership compounding on schedule: 12-high ramping, 16-high HBM3E first-to-develop, HBM4 ramp-ready H2 2025 with TSMC base die, 1c DRAM yield ahead of target — every dated milestone in the last two quarters has been met.
- Balance sheet now a weapon: net D/E 12%, FCF 13.9trn, net-cash target — capacity to fund M15X/Yongin/Indiana through a correction without financial stress, unlike prior downturns.
- Correction framed as supply-disciplined: industry capex concentrated on HBM, legacy capacity exiting (own legacy revenue to single-digit), Chinese entrants gated by technology gap and export controls — supports the "mild correction, different cycle" thesis.
- H2 2025 catalysts stacked: AI PC/phone content step-up (16GB→24/32GB), Windows 10 replacement, server DDR5 replacement cycle with low customer inventories, 1c ramp cost savings, M15X opening.
Bear case- The Q1 guide is a hard landing, not a soft patch: DRAM -low-teen and NAND -high-teen sequential bit declines imply the correction is deeper than the "mild" framing; if H2 recovery slips, 2025 becomes a two-half story with a weak first half.
- NAND is deteriorating, not stabilizing: second consecutive volume-guide miss, ASP now falling, demand growth revised down, and management offered no new production-cut quantum when asked directly — the profit-focused posture has not stopped the price slide.
- Q4 earnings quality is flattered: ~KRW1.3trn of the KRW1.5trn non-operating gain is a one-off stake-sale; underlying net profitability is materially below the 41% headline.
- Legacy collapse is now a quantified revenue headwind: ~20% → single-digit revenue share means a large chunk of 2024 revenue disappears by design in 2025; HBM must grow >100% just to hold the DRAM line if conventional pricing also corrects.
- Carried-inventory risk unresolved: legacy inventory is being "cleared" into a falling-price market with Chinese supply expanding; no size, writedown, or margin-impact disclosure, and no analyst pressed for one.
- Capex discipline is verbal, not numeric: 2025 capex refused in won terms; infrastructure spending rises "significantly" regardless of the correction; M15X/Yongin/Indiana commitments are fixed while demand visibility beyond HBM is poor.
- Dividend cut in disguise: dropping the 5%-of-FCF variable payout (KRW1,005/share this quarter) while raising the fixed by KRW300 reduces near-term total distributions — a signal management prioritizes cash preservation over returns at the cycle's most profitable point.
Next-quarter watchlist- Q1 2025 delivery vs guide: DRAM bits -low-teen, NAND bits -high-teen QoQ — whether the correction is contained to guidance or worse; watch the HBM vs conventional split within the DRAM decline (not disclosed this call).
- HBM 2025/2026 contracting: evidence for the >100% 2025 revenue growth guide; whether "most of 2026 volumes visible by H1 2025" is confirmed at the Q1/Q2 calls; ASIC customer names or mix.
- HBM3E 12-high crossover: >50% of HBM3E shipments by H1 2025 — the dated claim; any platform-timing friction (last quarter's UBS question remains unanswered in substance).
- HBM4 execution: development completion and ramp-readiness by H2 2025; TSMC base-die collaboration detail; 16-high HBM4 on track for H2 2026 shipments.
- Legacy exit and inventory: pace of the ~20%→single-digit revenue reduction; size and clearing of carried legacy inventory; any writedown as DDR4/LPDDR4 prices fall; Chinese DDR5 market entry evidence.
- NAND: whether ASP decline moderates; any actual production-cut action beyond "flexible adjustment"; eSSD momentum after the 61TB/122TB launches; general-purpose demand recovery timing.
- H2 recovery signposts: AI PC penetration toward 30–40%, Windows 10 replacement run-rate, server DDR5 restocking, China subsidy impact on smartphone demand — the "mild correction" thesis depends on these materializing.
- Capex and cash: final 2025 capex figure if disclosed; infrastructure vs fab split; progress toward the positive-net-cash goal; M15X Q4 2025 opening and Yongin construction start.
- Shareholder returns: first quarter under the new policy — fixed-only dividend mechanics and any update on the 50%-of-3-year-FCF framework.
|
| Oct 23, 2024 | +5.44% | — | Read transcript briefingQuarter in one view- Q3 2024: another record, now price/mix-led with volumes shrinking. Revenue KRW17.57trn (+7% QoQ, +94% YoY) — new record; OP KRW7.03trn (+1.56trn QoQ, OPM 40%, +7pp) — surpassing the 2018 super-cycle peak; EBITDA KRW10.1trn (57% margin); pre-tax KRW6.88trn; net KRW5.75trn (33% margin); D&A KRW3.07trn (still declining).
- Both volume guides missed. DRAM bits -low-single-digit QoQ vs +low-single-digit guide (PC/DDR4 weakness, partly offset by DDR5); NAND bits -mid-teen QoQ vs -mid-single-digit guide (discrete and client SSD, PC/mobile inventory correction). Management concedes it is deliberately carrying inventory on weak-price products rather than selling into falling prices.
- ASP did all the work: DRAM blended ASP +mid-teen QoQ (fourth consecutive up quarter); NAND ASP +mid-teen QoQ; prices rose across all products except discrete NAND and MCP.
- HBM and eSSD are now the P&L: HBM revenue +70%+ QoQ, +330%+ YoY, reaching 30% of DRAM sales (guided ~40% in Q4); eSSD revenue +~20% QoQ, +430%+ YoY, now >60% of NAND sales.
- Balance sheet: cash KRW10.9trn (+1.2trn); debt KRW21.8trn (-3.4trn); net debt KRW11trn (-4.6trn); D/E 33%. Transcript states net D/E of "70%" — almost certainly a transcription error (net debt/equity math implies high-teens); flag for verification against the release.
- Q4 guide: DRAM bits +mid-single-digit QoQ (HBM/server-led); NAND bits +low-teen QoQ (eSSD-led); HBM3E 12-high shipments begin Q4 as scheduled.
- Capex finally sized: 2024 "mid-to-high KRW10 trillion range," slightly above beginning-of-year plan; 2025 "slightly higher than this year" — the first aggregate disclosure after four quarters of refusal.
What management is focused on- Defending the HBM franchise against 2025 oversupply fears: 2025 volume and pricing negotiations "already completed with most customers"; 2025 average HBM price expected to rise YoY on mix shift to HBM3E; demand "likely higher than currently anticipated," supply constrained by yield difficulty and qualification — "premature to talk about a slowdown." Blackwell-delay concerns addressed head-on.
- Bifurcated DRAM market as the new framework: legacy (DDR4/LPDDR4) prices falling on Chinese entrants and weak PC/mobile; premium (HBM/DDR5/LPDDR5) tight with low inventories. Blended ASP expected to keep rising even as some product prices fall, because HBM mix rises to ~40% of DRAM revenue.
- Accelerated legacy exit: cutting DDR4/LPDDR4 production "earlier than initially planned" to free capacity for HBM3E, DDR5, LPDDR5; legacy industry inventories expected to normalize H1 2025.
- Profitability-over-volume in NAND, now explicit about the cost: volume market share will decline this year; revenue share will rise. Conservative capacity posture until industry inventories normalize.
- HBM capacity is the binding constraint: admitted production capacity "is facing limitations in meeting all of the increased demand in excess of our original plan" — HBM3E demand accelerating beyond expectations; response is faster legacy conversion, not new wafer adds.
- HBM4 de-risked by process conservatism: 1b nm + proven MR-MUF (not a new node), logic base die via foundry partner in a "one team" structure; H2 2025 shipments reaffirmed.
Key numbers and quarter mechanics- P&L: revenue 17.57trn (+7% QoQ, +94% YoY); OP 7.03trn (40%); D&A 3.07trn; EBITDA 10.1trn (57%); non-op -0.15trn (net interest -0.23trn, FX gain +18bn); pre-tax 6.88trn; net 5.75trn (33%).
- Volumes/ASP: DRAM bits -low-single-digit QoQ (miss vs +low-single guide); DRAM ASP +mid-teen QoQ. NAND bits -mid-teen QoQ (worse than -mid-single guide); NAND ASP +mid-teen QoQ.
- HBM: revenue +70%+ QoQ, +330%+ YoY; 30% of DRAM sales in Q3, ~40% guided Q4; HBM3E volume surpassed HBM3 in Q3 (as guided); TSV capacity more than doubled YoY on track; 2025 volumes/pricing largely contracted.
- eSSD: +~20% QoQ, +430%+ YoY; >60% of NAND sales in Q3 (vs "half of NAND revenue this year" guided last quarter — exceeded).
- Reversal: "minimal" impact this quarter — the tailwind is confirmed done, as guided; no other significant one-offs.
- Inventory: company deliberately carrying inventory on weak-price legacy products, to be worked down via production cuts; PC/mobile customer inventories flat QoQ; server customer inventories healthy; DDR5/LPDDR5 inventories tight.
- Capex: 2024 mid-to-high KRW10trn range (slightly above plan — HBM upside + M15X); 2025 slightly higher than 2024 (HBM/TSV, 1b transition, DDR5/LPDDR5 conversion, M15X + Yongin infrastructure); most of the increase is infrastructure/R&D/back-end, so near-term wafer output impact limited. M15X DRAM contribution from 2026.
- Demand outlook: 2024 PC shipments flat YoY, 2025 +low-to-mid-single-digit (Windows 10 EOL replacement + AI PCs); smartphones +low-to-mid-single-digit 2025 (AI features need +3–4GB DRAM per device); server market +mid-to-high-single-digit 2025; DRAM bit demand growth high-teen in 2025 (from mid-to-high-teen 2024); NAND mid-teen both years.
Product and launch scorecard- HBM3E 8-high: crossover delivered — HBM3E exceeded HBM3 volume in Q3 as promised.
- HBM3E 12-high: on schedule, now with commercial-production proof point. Commercial production started September; Q4 shipments reaffirmed; 12-high to surpass 8-high sales in H1 2025 and be the majority of HBM sales in H2 2025. Management claims 12-high will "take the lead in market share" as 8-high did. UBS challenged the timing as ahead of customer platform needs (NVIDIA not needing 12-high before ~Q2 2025, AMD MI325 on 8-high) — management did not directly reconcile this, answering with ramp/share confidence instead.
- HBM4: process decision locked and disclosed. 1b nm + advanced MR-MUF, H2 2025 shipments reaffirmed; logic die via foundry partner under "one team" development; rationale stated explicitly — HBM is not cost-per-bit-driven, so proven-process stability beats node aggression. Wafer conversion ratio "more than triple" conventional DRAM at HBM4 (vs 2–3x for HBM3E) — structurally tightening conventional supply further.
- 1c nm DDR5: industry's first 16Gb DDR5 on 1c developed end-August; sampling for validation; volume response planned for 2025 DDR5 demand.
- eSSD: 60TB QLC still "only company in the industry providing in mass"; 122TB (note: spec now stated as 122TB, not 128TB) in validation for H1 2025 supply — resolves last quarter's timing inconsistency toward the later date; PEB110 PCIe Gen5 eSSD (238-layer) developed in September.
- Legacy DRAM/NAND: the weak spot — DDR4/LPDDR4 prices falling, Chinese competition cited by name for the first time as a market factor; company response is accelerated exit, not defense.
Sell-side read-through- Dong-Won Kim (KB): legacy DRAM price outlook and blended-ASP decoupling — got the bifurcation framework: legacy down (China, weak demand), premium flat-to-up, blended ASP rising on HBM mix; 2025 HBM ASP guided up YoY. Substantive answer.
- Simon Woo (BofA): the volume-miss question — extracted the quarter's most important concession: both guides missed, and the company is carrying inventory on weak products rather than clearing at falling prices. This creates a disclosed inventory overhang to watch.
- Sung Kyu Kim (Daiwa): 2025 HBM oversupply (Blackwell delays, competitor supply) — got the full bull defense: contracts done, demand upside, supply downside from yields/qualification. No numbers on competitor capacity; the answer is confident but asserts rather than evidences the supply-side constraint for rivals.
- Giuni Lee (Goldman): capacity to respond to HBM upside — got the admission that capacity cannot meet all incremental demand; response is legacy conversion acceleration. Confirms HBM revenue upside is capacity-capped near term.
- Peter Lee (Citi): inventory normalization — PC/mobile customer inventories flat QoQ (not improving), server healthy, legacy industry inventories normalize H1 2025. A concrete, testable timeline.
- Nicolas Gaudois (UBS): the sharpest challenge — 12-high shipping ahead of apparent customer platform need; is the ramp early? Management's answer did not address the platform-timing gap directly; share-leadership claim repeated. Evasion noted.
- Sun-woo Kim (Meritz): NAND investment conservatism — confirmed deliberate: volume share loss accepted, revenue share gain claimed, conservative until inventories normalize.
- Jay Kwon (J.P. Morgan): HBM4 wafer conversion ratio — got a new quantified disclosure: HBM3E at 2–3x DDR wafer intensity, 12-high worse on yields, HBM4 "more than triple." Supports the multi-year conventional-tightness thesis.
- Min-sook Chae (Korea Investment): China response — legacy retreat, premium focus, newcomers "far behind" in technology. No quantification of Chinese supply impact.
- Dong-Hee Han (SK Securities): capex — finally extracted the number: mid-to-high KRW10trn 2024, 2025 slightly higher, with scenario-based flexibility. The four-quarter opacity streak ends.
- Ricky Seo (HSBC): one-offs — confirmed minimal reversal impact and no other significant one-offs; Q3 margin is clean.
- YoungHo Ryu (NH): HBM4/logic die — got the 1b nm + MR-MUF confirmation and foundry "one team" structure; H2 2025 reaffirmed.
- Notable absences: no dividend question, no Kioxia question (fourth quarter), no HBM margin-durability question, no question on the carried-inventory size or its margin risk.
Management credibility- Positive: capex disclosure delivered after four quarters of refusal — a real number with composition detail (infrastructure/back-end weighted) and a flexibility framework. Materially improves the transparency record.
- Positive: reversal and one-off claims verified clean — "minimal" reversal confirmed, no other one-offs; the 40% OPM is operational. Last quarter's quality-of-earnings read holds.
- Positive: HBM milestones keep landing on or ahead of schedule — HBM3E > HBM3 crossover in Q3 as guided; 12-high commercial production September, Q4 shipments reaffirmed; HBM4 H2 2025 reaffirmed with process now specified.
- Negative: first double volume-guide miss of the recovery. DRAM guided +low-single, delivered -low-single; NAND guided -mid-single, delivered -mid-teen. Both misses attributed to PC/mobile, but the NAND miss is the third consecutive soft/down volume quarter, and the Q4 +low-teen NAND bit guide is now a high-bar testable claim.
- Negative: the carried-inventory disclosure cuts both ways — candid, but it means reported revenue understates weak-end demand and creates a future clearing risk; no size or margin-impact quantification offered, and no analyst pressed for one.
- Mixed: 12-high timing question evaded — UBS's platform-mismatch challenge was answered with ramp confidence, not demand evidence. The H1 2025 8/12 crossover claim now carries this open question.
- Mixed: net D/E "70%" is internally inconsistent with net debt of 11trn and D/E 33% — near-certain transcript error, but must be verified against the release before use.
- Watch: "2025 HBM price up YoY" is now on the record alongside "volume and pricing negotiations completed with most customers" — a specific, falsifiable claim that anchors next year's HBM margin debate.
What changed versus the prior quarter- Growth decelerated sharply but profitability inflected up: revenue +7% QoQ (vs +32% in Q2); OPM 33% → 40%; net margin 25% → 33%. The model is now fully price/mix-driven — volumes fell in both segments.
- Volume guides went from beat to double miss: Q2 DRAM beat (+low-20% vs +mid-teen); Q3 both segments missed. The PC/mobile weakness management flagged for two quarters is now hitting its own shipment lines, not just the market narrative.
- Reversal tailwind confirmed finished: "minimal" impact in Q3, as guided — the 40% OPM is clean.
- Capex opacity ended: from four quarters of no figure to 2024 mid-to-high KRW10trn and 2025 slightly higher. M15X contribution dated to 2026.
- HBM mix quantified and steep: 30% of DRAM sales in Q3 → ~40% guided Q4; 2025 pricing locked and guided up YoY — new disclosure.
- eSSD overshot its own target: >60% of NAND sales in Q3 vs "half of NAND revenue this year" guided last quarter; 122TB (revised spec language) H1 2025.
- China entered the script: Chinese legacy DRAM supply cited by management as a price-pressure factor for the first time; response is accelerated DDR4/LPDDR4 exit "earlier than initially planned" — a strategy change, not just commentary.
- Inventory posture changed: from customer-inventory commentary to the company itself carrying inventory on weak products — a new balance-sheet/P&L risk that did not exist in prior quarters' disclosures.
- Demand recovery deferred again: PC 2024 now flat YoY; recovery pushed to 2025 on Windows 10 EOL and AI-device content gains (+3–4GB DRAM per AI smartphone quantified).
Bull case- Clean 40% OPM with the reversal gone — margins are now purely price/mix/D&A-driven, and D&A is still falling; OP already exceeds the 2018 super-cycle peak with HBM mix still rising toward 40% of DRAM revenue.
- 2025 HBM is contracted and priced up: negotiations complete with most customers, ASP guided up YoY, demand described as exceeding expectations with capacity the constraint — revenue visibility through 2025 is as high as the industry offers.
- 12-high leadership compounding: commercial production September, Q4 shipments, H1 2025 crossover, majority of H2 2025 sales — if share leadership holds as claimed, the mix shift alone supports blended ASP regardless of legacy pricing.
- Conventional tightness thesis strengthened, not weakened: HBM4 at >3x wafer intensity, legacy capacity being cut faster than planned, DDR5/LPDDR5 inventories tight, server demand growing — the non-HBM pricing floor looks protected even as legacy falls.
- eSSD is a proven second engine: >60% of NAND sales, +430% YoY, sole-source 60TB, 122TB coming — NAND is profitable and growing revenue while deliberately shedding volume share.
- Self-funded buildout with a shrinking balance sheet: net debt -4.6trn in the quarter, D/E 33%, capex sized and framed within cash generation.
Bear case- The volume engine has stalled and the misses are widening: both segments missed Q3 guides; NAND bits down mid-teen; Q4's +low-teen NAND guide requires an immediate sharp reversal of a three-quarter trend.
- Inventory is now on SK Hynix's own books: carrying unsold legacy product into a falling-price segment (Chinese competition intensifying) is a margin and writedown risk that reverses the clean-inventory story of the past year. Size undisclosed.
- Legacy price erosion is structural, not cyclical: Chinese entrants are named, the company's answer is exit — meaning legacy revenue shrinks by design while premium must carry all growth. Any HBM or eSSD stumble leaves no fallback volume base.
- 12-high may be ramping ahead of platform demand: UBS's point stands unanswered — if NVIDIA/AMD platforms don't absorb 12-high until mid-2025, early shipments risk inventory build or pricing concession at the exact product where expectations are highest.
- PC/mobile recovery has been deferred three times: 2024 PC now flat, recovery resting on 2025 Windows 10 replacement and AI-content gains that are still unproven at the consumer level.
- Capex is rising into a bifurcating market: 2025 above 2024's mid-to-high-teens trillions, with M15X/Yongin infrastructure committed regardless of cycle; the "multiple scenarios" flexibility language is untested.
Next-quarter watchlist- Q4 delivery vs guide: DRAM bits +mid-single-digit, NAND bits +low-teen — the NAND number is the key credibility test after a mid-teen decline; also whether HBM hits ~40% of DRAM sales as guided.
- Carried inventory: size, composition, and clearing pace of the deliberately held legacy inventory; any writedown as DDR4/LPDDR4 prices fall; whether production cuts proceed "faster than planned" as stated.
- HBM3E 12-high ramp vs platform timing: Q4 shipment evidence, customer qualification commentary, and any reconciliation of the UBS platform-gap question; H1 2025 8/12 crossover remains the dated claim.
- 2025 HBM contract detail: any quantification of the completed 2025 volume/pricing agreements; the "ASP up YoY" claim is on the record and will anchor estimates.
- Legacy price trajectory and China: DDR4/LPDDR4 price declines, Chinese supply pace, and whether the accelerated legacy exit shows up in Q4 bit mix; legacy industry inventory normalization by H1 2025 is the dated claim.
- eSSD: 122TB validation progress toward H1 2025 supply; whether >60% NAND revenue share holds as client/discrete recovers or shrinks.
- Capex and balance sheet: final 2024 capex print vs the mid-to-high KRW10trn range; 2025 plan finalization; debt trajectory from 21.8trn; verify the net D/E figure (transcript's "70%" appears erroneous).
- HBM4: foundry partner confirmation, base-die detail, customer commitments, and any update on 16-high process selection ahead of H2 2025 shipments.
|
Corbanu