
Samsung Electronics
Vertically integrated IDM: designs + fabricates + sells memory and logic chips (Device Solutions), plus consumer devices (Device eXperience), displays (Samsung Display), and Harman audio/auto; cyclical commodity-memory pricing is the dominant earnings swing factor.
Sources — 7 figures with citations
- Latest print: Q3 2026 (quarter ended 2026-09-30, released 2026-10-29)filed2026-11-12revenue KRW 195.3 trillion; +14% QoQ (the release gives no YoY figure); guide: Q4 2026: continued AI-infrastructure memory investment, foundry double-digit revenue growth; FY2026 capex about KRW 84.4 trillion; 2026 shareholder returns KRW 90-110 trillionnews.samsungsemiconductor.com — Recorded Nov 12 2026; figures as the source states them. The entry's analysis text predates this print unless dated otherwise.
- earnings, 2026-07-30filed2026-07-30Q2 2026: record revenue KRW 171.5 trillion (+28% QoQ) and record operating profit KRW 89.5 trillion; DS division revenue KRW 127.5 trillion and operating profit KRW 89.2 trillion with memory at all-time-high revenue and operating profit; HBM4 sales scaled up and the industry's first HBM4E samples shipped; MX and Networks revenue KRW 33.2 trillion with an operating loss of KRW 0.7 trillion; EPS KRW 10,849 (+52%); foundry cites 2nm HPC design wins and targets double-digit H2 revenue growth.news.samsungsemiconductor.com — CONFIRMED; recorded Nov 12 2026; source dated 2026-07-30
- earnings, 2026-10-29filed2026-10-29Q3 2026: record revenue KRW 195.3 trillion (+14% QoQ) and operating profit KRW 107.4 trillion, which Samsung calls the first time a tech company exceeded KRW 100 trillion in quarterly operating profit; DS division revenue KRW 147.1 trillion (+15% QoQ) and operating profit KRW 106.5 trillion, memory at all-time highs on AI demand amid capacity constraints, HBM4 sales 'significantly scaled up'; foundry improved on HBM base-die and advanced-node sales with a Q4 target of double-digit revenue growth and the Taylor, Texas fab set to begin operations in 2027;news.samsungsemiconductor.com — CONFIRMED; recorded Nov 12 2026; source dated 2026-10-29
- guidance, 2026-10-29filed2026-10-29Earnings-call coverage reports that HBM4 sales more than tripled QoQ to 60% of total HBM sales, that 2027 HBM supply is sold out and that Samsung expects the memory shortage to persist through 2028.sammyfans.com — REPORTED; recorded Nov 12 2026; source dated 2026-10-28
- other, 2026-11-04filed2026-11-04SEDaily (via Sammyfans) reports Samsung will build 2nm and 4nm foundry lines at Giheung to add sub-5nm base-die capacity for HBM4; TrendForce separately reports Taylor Fab 2 (2nm, about 2.7 million square feet, about 1.5x Fab 1) is nearly fully booked before 2030 output, with AMD HBM base-die orders among the drivers, and that full-scale construction was slated to begin in early November 2026.sammyfans.com — REPORTED; recorded Nov 12 2026; source dated 2026-11-04
- Watched signal: ~KRW110T 2026 capex (+128% y/y) industry-wide record across all three makersfiled2026-11-12Samsung's own full-year 2026 capex is now expected at about KRW 84.4 trillion (DS KRW 76.8 trillion), with KRW 29 trillion spent in Q3 and KRW 56.9 trillion in January-September.news.samsungsemiconductor.com — Outcome of a signal this entry said to watch; recorded Nov 12 2026.
- Watched signal: Memory demand-fulfillment rate at record low; customers pulling 2027 demand forwardfiled2026-11-12Samsung's Q3 release expects 2027 AI infrastructure capex at a historic level with AI demand far exceeding industry supply; call coverage (REPORTED) says 2027 HBM supply is sold out and the shortage lasts through 2028.news.samsungsemiconductor.com — Outcome of a signal this entry said to watch; recorded Nov 12 2026.
The thesis on this name
Memory Super-Cycle
DRAM/NAND scale leader racing SK Hynix on HBM3E/HBM4 qualification — the swing supplier whose HBM ramp paces the memory cycle.
Earnings, margins, COGS & capex
Q1 FY26 was a record quarter driven almost entirely by memory: consolidated revenue KRW 133.9T (+43% q/q, all-time high) and operating profit KRW 57.2T (all-time high), with the Device Solutions (chip) division alone posting KRW 81.7T revenue and KRW 53.7T OP vs ~KRW1T a year earlier. Memory ASP rose ~146% vs the FY25 average. The non-memory businesses (mobile/DX, foundry) are comparatively weak — foundry likely still loss-making (~KRW1.6T est. loss) and mobile slumped — so this is a memory-supercycle earnings profile, not broad-based.
Income statement — where each revenue dollar goes
% of revenueOf every $1 of revenue, ~99¢ is cost of goods and ~0¢ operating expense, leaving ~1¢ of operating profit.
Revenue trend
Margins
sharply up from low-teens FY25 on memory ASP surge
record — memory-driven
still unprofitable; targeting improvement on HBM4 base-die + full advanced-node utilization H2'26
COGS structure
Dominated by silicon wafer + fab depreciation + process materials; memory is a fixed-cost-heavy commodity so OP swings violently with ASP (the +146% ASP move drove the ~50x memory OP jump). Not disclosed line-by-line for Q1 FY26.
Capex
~KRW110T (~$73.2B) 2026 semiconductor capex+R&D, +128% y/y — funds HBM4 capacity (target ~250k wafers/mo by end-2026, +~47%), advanced packaging, and closing the foundry yield gap with TSMC.
Latest earnings
Beat — profit surged ~756% y/y (chip OP up ~48x), above estimates
Memory tightness expected to persist into 2027; demand-fulfillment rate at record low with customers pulling 2027 demand forward; HBM revenue >3x y/y in 2026; ~KRW110T 2026 capex; foundry earnings improvement targeted H2'26
- Memory ASP move
- +~146% vs FY25 avg
- DRAM market share (Q1 FY26)
- ~38% (#1)
- HBM market share (Q3'25)
- ~35% (#2, behind SK Hynix ~53-57%)
- 2026 HBM wafer target
- ~250k wafers/mo by year-end
Growth drivers
- HBM4 ramp — mass production began Feb 2026, first to ship HBM4 to NVIDIA + AMD; HBM revenue guided to >3x y/y in 2026, HBM4 >50% of HBM sales from Q3
- Conventional DRAM (DDR5/server) ASP surge from AI datacenter demand pulling supply forward into 2027
- NAND tightening alongside DRAM
- Foundry HBM4 base-die supply + 2nd-gen 2nm mobile ramp H2'26 (path to narrowing foundry losses)
- Advanced packaging for chiplet AI processors
Bull & bear
Samsung is the lowest-priced way to own the AI memory supercycle: a process-leading #1 DRAM maker, now caught up on HBM4 (first to ship to NVIDIA/AMD), trading on a single-digit forward P/E (~6-8x) with record OP and visible multi-year HBM contracts into a 2027 shortage.
- Record Q1 FY26: KRW133.9T revenue / KRW57.2T OP, memory OP up ~50x on +146% ASP — and supply is far short of demand into 2027
- Forward P/E ~6-8x despite record earnings — sell-side (KB) models KRW327T FY26 / KRW488T FY27 OP, which would make it the world's most profitable company
- HBM4 leadership regained: first to ship, top qualification scores, pricing parity with SK Hynix at NVIDIA — share-gain optionality on top of the cycle
- Vertical integration (own 4nm base-die) gives HBM lead-time independence from TSMC; ~KRW110T capex self-funded from net cash
- Conglomerate optionality (DX/Display/Harman) + foundry turnaround call provide upside the pure-plays lack
This is peak-cycle memory earnings priced as if durable. Samsung is the #2 (not #1) HBM supplier behind SK Hynix at NVIDIA, its foundry loses money, and the record industry-wide capex that today's prices are funding is the mechanism that historically flips shortage into glut within a few quarters.
- Commodity-memory earnings are violently mean-reverting — the Q3'18 peak halved profits in two quarters; record 2026 capex across all three makers seeds a 2027-2028 oversupply
- Profit is concentrated in one product line — mobile/DX slumped, foundry lost ~KRW1.6T; ex-memory the franchise is weak
- SK Hynix owns the HBM relationship with NVIDIA (~2/3 of HBM4), the ~90% AI-accelerator share customer — and overtook Samsung in Korea market cap (22 Jun 2026), confirming the market sees SK Hynix as the AI-memory leader
- Foundry remains structurally subscale vs TSMC — capital sink with uncertain ROI
- Single-stock-day KOSPI swings of >12% (23 Jun 2026) show how reflexive and sentiment-driven the supercycle re-rating is
- Geopolitical/tariff exposure and Chinese DRAM entrants (CXMT) pressure legacy memory pricing over time
What it is worth
Comps + reverse-DCF sanity. Comps: SK Hynix and Micron both crossed ~$1T caps in 2026 on the same AI-memory thesis; all three trade at low single-digit forward P/Es (Samsung ~6-8x fwd) and near-zero forward PEGs — the market is pricing a cycle peak, not durable growth. SK Hynix is no longer the larger Korea cap: about $733B against roughly $1.0T for Samsung at the Nov 3 2026 closes (estimate from aggregator share counts).
2027-2028 oversupply flips ASP down (Q3'18 analog: profit halved in two quarters); SK Hynix retains NVIDIA HBM lead; foundry keeps losing money → earnings and multiple both compress.
Record 2026 memory profits moderate through 2027 as capacity lands; HBM share stabilizes at #2; foundry losses narrow slowly. Earnings high but past peak — value roughly fair at current multiple.
Tightness extends into 2027 + HBM4 share gains + foundry inflection → KB-style KRW327T FY26 / KRW488T FY27 OP; even a modest re-rate off a single-digit P/E implies large upside.
Reverse-DCF read: at ~$1.4T cap on ~$255B TTM revenue (~5.5x sales) but record peak-cycle OP, the price implies the market is NOT extrapolating the +146% ASP / KRW57T-quarterly OP run-rate — the single-digit forward P/E means consensus already discounts a sharp normalization. To justify the current price as cheap, memory tightness must persist into 2027 (management's base case) AND foundry must stop bleeding. The bear math: capitalize a mid-cycle (not peak) OP and the multiple looks fairer, not cheap. ANALYSIS ONLY — non-US name, not a buy/own recommendation.
SWOT
Strengths
- World's #1 DRAM maker (~38% share) and a top-2 NAND maker — scale + process leadership in the supply-constrained memory market
- Only true vertically integrated memory + foundry + advanced-packaging IDM; procures own 4nm logic base-die for HBM4, giving lead-time independence from TSMC
- First to ship HBM4 to NVIDIA + AMD (Feb 2026), highest qualification scores; HBM4 at 11.7 Gb/s exceeds the 10 Gb/s NVIDIA/AMD spec
- Record balance-sheet strength and historically large net cash to self-fund the largest-ever semiconductor capex
- Diversified conglomerate optionality — DX (smartphones/TVs #1 globally), Samsung Display, Harman cushion pure-memory cyclicality
Weaknesses
- Foundry still loss-making (~KRW1.6T est. Q1 loss) — years behind TSMC on leading-edge yield/customer trust
- Lost the HBM3E generation lead to SK Hynix, which is NVIDIA's dominant (~2/3) HBM supplier and overtook Samsung in Korea market cap on 22 Jun 2026
- Mobile/DX earnings slumped in Q1 FY26 — group profit is dangerously concentrated in one cyclical commodity (memory)
- Extreme earnings cyclicality — prior supercycle (Q3'18 peak) halved profits within two quarters
- Group margins/FCF not transparently disclosed at segment granularity; heavy 2026 capex compresses near-term FCF
Opportunities
- AI/HBM supercycle — memory shortage customers are pre-booking into 2027, multi-year HBM4 contracts; KB Securities est. KRW327T FY26 / KRW488T FY27 OP
- Catching the HBM4 generation — comparable pricing to SK Hynix at NVIDIA signals restored pricing power and a path to retake share
- Foundry inflection if HBM4 base-die + 2nm mobile lift utilization toward profitability
- Advanced packaging / chiplet AI demand as a new high-margin leg
Threats
- Memory oversupply risk ~2027-2028 as Samsung + SK Hynix + Micron all add record capacity into a price-elevated market — the classic cycle reversal
- SK Hynix's HBM lead with NVIDIA (the ~90%+ AI-accelerator share customer) caps Samsung's HBM share upside
- TSMC's foundry dominance keeps Samsung's logic business structurally unprofitable / subscale
- Geopolitics — US/China export controls, Korea-US tariff exposure, Chinese DRAM entrants (CXMT) commoditizing legacy DRAM/NAND
- Labor disruption (a May 2026 strike was flagged as a potential ~4% DRAM-output risk)
Moats, dependencies & bottlenecks
Moats
Oligopoly with SK Hynix + Micron; node leadership erodes if a generation is missed (as HBM3E showed).
Lead-time independence from TSMC for HBM — a genuine structural edge vs Micron/SK Hynix on logic-die sourcing.
~KRW110T self-funded capex; few competitors can match the absolute investment.
HBM4 qualified at NVIDIA/AMD, but SK Hynix holds the dominant NVIDIA position — Samsung is the share-challenger, not the incumbent.
#1 smartphones/TVs gives diversification but thin margins vs memory.
Dependencies
The gatekeeper of AI-memory demand; controls ~90% of AI accelerators. HBM allocation to Samsung vs SK Hynix sets the upside.
Second major HBM4 customer; MI-series accelerators diversify Samsung's AI-memory demand.
Sole EUV lithography supplier for leading-edge DRAM (1c) and foundry nodes — a hard upstream bottleneck.
Deposition/etch WFE for the capacity ramp; tool availability gates the +47% HBM wafer expansion.
Competitor/partner Foundry rival; some HBM logic-die work in the industry routes through TSMC — Samsung's own base-die reduces this dependency.
~70%+ of group OP is memory ASP-driven; the single largest swing factor, outside management control.
Advantages
- Largest-ever self-funded semiconductor capex into a shortage
- Only fully vertically integrated memory+foundry+packaging IDM
- First-mover on HBM4 shipments with top NVIDIA/AMD qualification scores
- #1 DRAM share + diversified conglomerate cash flows
Weaknesses
- Lost HBM3E lead to SK Hynix; #2 at NVIDIA
- Chronically unprofitable foundry
- Earnings concentrated in one cyclical commodity
- Weak mobile/DX contribution in the current quarter
Bottlenecks
- EUV tool / WFE availability gating the HBM capacity ramp
- Advanced-packaging (TSV/hybrid-bonding) capacity for HBM4
- Foundry yield gap vs TSMC limiting logic-die and external-customer revenue
- HBM allocation decisions by NVIDIA capping share-gain pace
- Labor relations risk (May 2026 strike flagged ~4% DRAM-output impact)
Top signals & trends
Top signals
7 sourced items since this entry's vintage are in the evidence rows (latest print, guidance, deals, resolved watch items); the analysis text is as of the entry's stated date and predates them unless dated otherwise.
Confirms a genuine shortage, not channel-stuffing — pricing power persists near-term.
The classic supercycle-ending mechanism — capacity added at peak prices seeds 2027-2028 glut risk.
Market verdict that SK Hynix is the AI-memory leader; Samsung is the discounted challenger.
Restored pricing power vs the HBM3E generation where Samsung lagged.
Either deep value or the market pricing a cycle peak — the central debate.
Trends
Primary driver of the +146% ASP and record OP; multi-year contracts into 2027.
Conventional DRAM ASP +90-95% q/q as wafers shift to HBM.
Negative (medium-term) · Oversupply risk ~2027-2028 if AI demand normalizes as capacity lands.
Keeps Samsung foundry subscale/loss-making; capital sink.
Negative (long-term) · Commoditizes legacy memory; geopolitics adds export-control overhang.
Ecosystem & competitor graph
Suppliers feed the company; customers pull from it. Line thickness shows the strength of each tie (supply-chain dependency, customer earnings contribution). Hover to isolate a tie.
Sole EUV lithography supplier for 1c DRAM + leading-edge foundry — the hardest upstream dependency.
Deposition/etch/CMP WFE for the memory + packaging ramp.
Etch + deposition tools critical to 3D NAND and advanced DRAM.
Process control/inspection — yield-critical for HBM and leading-edge nodes.
Advanced materials, filtration, and packaging chemistries.
The pivotal AI-accelerator buyer of HBM4 (~90% accelerator share); Samsung is its #2 HBM supplier behind SK Hynix.
HBM4 customer for MI-series AI accelerators; diversifies Samsung's AI-memory demand.
Major buyer of Samsung DRAM/NAND/displays for iPhone/Mac (and a device competitor).
Server OEMs and cloud buyers driving the DDR5/server-DRAM demand surge.
Custom-ASIC AI accelerators (hyperscaler XPUs) are an additional HBM demand channel.
Direct memory rival and HBM leader (~53-57% HBM share, ~2/3 of NVIDIA HBM4); overtook Samsung in Korea market cap Jun 2026. The key share competitor.
US #3 memory maker, ~11-21% HBM share, sold out 2026 HBM; crossed $1T market cap May 2026. The US-listed pure-play comp.
Foundry rival Samsung is years behind; the benchmark for Samsung's loss-making logic business.
NAND competitors; relevant to Samsung's storage segment.
Competes with Samsung DX in premium smartphones while also being a major component customer.