005930
The #1 Memory Maker, Priced for the HBM It Does Not Lead

Snapshot: Executive Summary

Samsung Electronics is the world's #1 memory maker overall (DRAM, HBM and NAND) and a diversified electronics conglomerate, headquartered in Suwon, South Korea. Unlike the pure-play SK hynix, it is a sum-of-the-parts: a memory crown jewel, a distant-#2 and loss-making foundry, the Galaxy mobile franchise, the ~84.8%-owned Samsung Display, wholly-owned Harman, a large net-cash balance sheet, and listed affiliate stakes. It is one of the few AI-memory winners that is also a net-cash conglomerate valued well below its pure-play peers on the sum of its pieces.

The catch is the memory that matters. On HBM, the stacked DRAM bolted to every AI GPU, Samsung is the laggard: it cleared NVIDIA's HBM4 qualification only in early 2026 and takes a small initial share of the Rubin allocation while SK hynix holds the majority, leaving its HBM revenue share at an estimated ~22%, far behind SK hynix's ~57% and roughly level with Micron's ~20%⁴. Its foundry is a distant #2 to TSMC and structurally loss-making into ~2028⁵, and its mobile unit was warned toward a possible first-ever annual loss as AI-driven memory-cost inflation squeezes margins¹⁰.

FY2025 was solid but unspectacular: revenue ₩333.61tn, operating profit ~₩43.6tn, record R&D of ~₩37.7tn¹. Then 2026 inflected violently on the AI-memory boom: Q1 2026 revenue reached ₩133.87tn, and Q2 2026 (reported Jul 30) set records with revenue ₩171.5tn (+28% QoQ) and operating profit ₩89.5tn (+1,814% YoY, ~$59B), briefly making Samsung the world's most profitable listed tech company, topping NVIDIA; the DS chip division alone earned ₩89.2tn, while reported OP absorbed a ~₩15-17tn one-off special-bonus provision (underlying OP topped ₩100tn)². The stock re-rated over sixfold to a ₩374,500 May high, then de-rated ~38% to ₩231,000 (Aug 7) as the market began pricing a memory-cycle peak behind the record print, leaving price-to-book at ~3.2x against a ten-year ~1.4x median³.

We value Samsung sum-of-the-parts: memory + foundry + DX/mobile + the 84.8% Samsung Display stake + Harman on peer multiples, plus net cash and listed affiliate stakes (at a holding-company discount), less the display minority⁶. The base lands at ₩283,000 (bull ₩340,000 / bear ₩190,000), now ~22% above the ₩231,000 spot after the de-rating and far below the ₩471,908 Street mean (36 Buy / 1 Hold of 37 analysts)⁸: the market has sold the stock below our fair-value range, yet we still Hold, no longer because it is expensive but because the very peak-cycle risk that drove the sell-off (SK hynix's own cap fell to ~₩1,036tn), the persistent HBM-laggard gap, and mobile's realized loss keep us from chasing; the ₩190,000 bear is ~18% below spot. Note the library's memory leader SK hynix (000660) is a Hold at ₩2,700,000: Samsung is the cheaper, weaker-HBM way to play the same supercycle.

Tactical: Samsung is trading at ₩231,000, ~22% upside to our ₩283,000 sum-of-the-parts base target after the Korean memory complex de-rated ~40% from its peak despite record Q2 profit. We rate it a Hold: the sell-off dragged the stock below our fair-value range, but the memory arm, though Buy-quality and just briefly more profitable than NVIDIA, is still the HBM laggard behind SK hynix, its foundry loses money into ~2028, and mobile has now booked its first-ever quarterly operating loss. The very cycle-peak risk that drove the sell-off (SK hynix's own cap fell to ~₩1,036tn) keeps us from chasing; the ₩190,000 bear is ~18% below spot. We would need evidence the memory cycle is durable, not peaking, before upgrading. Rating Hold.

Investment Thesis

Rating: Hold. The probability-weighted scenarios (50% base / 25% bull / 25% bear) blend to ~₩274,000, now ~19% above the ₩231,000 spot after the ~40% de-rating of the memory names from their peak. All three anchors except the ₩190,000 bear (~18% downside) sit above spot, so this is no longer a Hold on a stretched price: it is a Hold held because the sell-off is the market's verdict on an approaching memory-cycle peak, mobile has realized its first-ever quarterly loss, and the HBM4 gap persists (HBM4E only now sampling). We stay far below the ₩471,908 Street mean: we credit Samsung's record memory profit but will not underwrite the euphoric Street into a cycle that both Samsung and SK hynix have already de-rated. Upgrade trigger: evidence the memory cycle is durable AND the HBM4 ramp proving in at NVIDIA at scale. Downgrade to Sell: the memory cycle clearly rolling while the HBM gap stays permanent.

Business Overview

Samsung Electronics runs several businesses under one listing. Device Solutions (DS) is the chip division: Memory (DRAM, HBM, NAND, the profit core and crown jewel) plus Foundry and System LSI (contract chipmaking and Samsung's own Exynos/image-sensor silicon, the loss-making leg). Device eXperience (DX) is the consumer arm: Mobile (MX), the Galaxy phone/tablet/wearable franchise and the DX profit core, plus Visual Display and Digital Appliances (TVs and home appliances, near breakeven). Samsung Display (SDC), ~84.8% owned, is the dominant small and medium OLED panel maker. Harman, wholly owned, is audio and automotive electronics. On top sits a large net-cash balance sheet and stakes in listed affiliates (Samsung Biologics, Samsung SDS, Samsung Electro-Mechanics, Samsung SDI)⁶. Because the pieces diverge so sharply, a single P/E obscures the value: this report values them sum-of-the-parts.

Where Samsung sits

• Memory (the crown jewel): #1 globally in DRAM and NAND combined; the profit core, now driven by the AI-memory up-cycle. But the HBM laggard behind SK hynix.
• Foundry + System LSI (the drag): a distant #2 to TSMC, structurally loss-making, subsidized by memory profits; an option on the 2nm ramp, not a cash engine today.
• DX / Mobile (the consumer core): the Galaxy franchise is the DX profit engine; TVs and appliances are near breakeven and squeezed by memory-cost inflation.
• Samsung Display + Harman + net cash + affiliates: the balance-sheet and portfolio floor that keeps this a Hold rather than a Sell.

Memory & HBM: The Crown Jewel and the Crux

The single most important section in this report. Samsung is the world's #1 memory maker by DRAM and NAND bit volume, with a larger commodity-memory base than SK hynix. In a normal cycle that scale is the advantage. In this cycle it is not enough, because the value has migrated to HBM, the stacked DRAM that gates every AI accelerator, and on HBM Samsung is behind. It cleared NVIDIA's HBM4 qualification in early 2026 and shipped HBM4E samples to major customers by mid-2026, but still takes a small share of the Rubin allocation, while SK hynix holds the majority and Micron runs roughly level with Samsung⁴. Estimates of Samsung's HBM revenue share vary widely (roughly ~22%, Counterpoint end-2025, matching the sister SK hynix dashboard), which is why we keep the HBM share flagged as an estimate.

Why the laggard status matters to the valuation

HBM is where the pricing power and the margin are. SK hynix, the pure HBM leader, has itself de-rated to a ~₩1,036tn market cap (down ~40% from the peak) as the market prices a memory-cycle top, even though it stays the allocation leader³. Samsung's memory arm has ~2x the bits and, in Q2, out-earned SK hynix (DS operating profit ₩89.2tn), so a like-for-like through-cycle read-across would screen it above SK hynix's crushed spot cap; we value it at ~₩1,600tn, a deliberate haircut to that read-across for the HBM-laggard gap and the heavier, lower-margin NAND mix. The bull case is that HBM4/HBM4E scales at NVIDIA and the arm re-rates on ~2x the bit base (memory ~₩1,900tn); the bear case is that the gap is structural and the cycle rolls (memory ~₩1,050tn). This one judgment (the memory segment is ~80% of the sum-of-the-parts equity value) swings the whole rating, which is why the interactive model puts the memory value on a live slider.

Foundry & System LSI: The Loss-Making Option

Samsung Foundry is the world's #2 pure-play foundry, but a distant one: roughly ~7% share against TSMC's ~70%, a gap that widened rather than narrowed through 2025 as AI compute concentrated at TSMC. The division (Foundry plus System LSI, Samsung's own Exynos and image-sensor silicon) is structurally loss-making, with analyst estimates clustering around ~₩5-7tn of operating loss in FY2025, subsidized entirely by memory profits. Management guides annual foundry profitability only to around 2028⁵.

The option, not the engine

There is a real bull kernel: Samsung reached 2nm (SF2) GAA mass production in late 2025 (ahead of Intel, roughly a generation behind TSMC's ramp), anchored a marquee ~$16.5bn Tesla AI6 win for its Taylor, Texas fab, and validates the node internally with the Exynos 2600. That is genuine option value: if yields hold and more anchor customers land, the loss turns toward breakeven and the segment re-rates. But it is an option, not a cash engine, so we value it near zero in the bear case and modestly in the base. The integrated IDM model cuts both ways: captive volume and memory co-optimization, but fabless rivals stay wary of handing designs to a company that also competes in end products.

DX, Display & Harman: The Rest of the Parts

Outside the chip division, three businesses round out the sum-of-the-parts. DX / Mobile (MX) is the Galaxy phone/tablet/wearable franchise, the consumer profit core, running a double-digit annual margin in a good year; but in Q2 2026 it booked its first-ever quarterly operating loss (₩0.7tn on ₩33.2tn revenue) as AI-driven memory-cost inflation squeezed margins, putting a first annual loss in play¹⁰. Visual Display and Digital Appliances (TVs and home appliances) are near breakeven and dilutive to the blend. Samsung Display (SDC), ~84.8% owned, is the dominant small and medium OLED supplier (an estimated ~40-48% of the OLED revenue market), a real, cash-generative asset we value net of the ~15.2% minority. Harman (acquired 2016 for ~$8bn, wholly owned) is audio and automotive electronics, a steady mid-teens-multiple business.

Financial Health & Trends

Solid, then inflecting

FY2025 was a solid but unspectacular year: revenue ₩333.61tn (WSJ, +10.9% YoY), operating profit ~₩43.6tn (a ~13% margin), and record R&D of ~₩37.7tn as memory only inflected late in the year¹. 2026 then turned violently higher on the AI-memory boom: Q1 2026 revenue was ₩133.87tn, and Q2 2026 set records with revenue ₩171.5tn (+28% QoQ) and operating profit ₩89.5tn (+1,814% YoY), the DS chip division alone earning ₩89.2tn and the group briefly out-earning NVIDIA; reported OP absorbed a ~₩15-17tn one-off special-bonus provision, so underlying OP topped ₩100tn². Q2 EPS was ₩10,849, lifting trailing-twelve-month EPS toward ~₩27,000, while a forward P/E on the H1 run-rate is only ~5.5x. One caveat we keep flagged: below the confirmed division level (DS / MX / SDC), the memory-versus-foundry-versus-DX profit breakdown remains an author estimate.

Note: Q2 2026 revenue, operating profit and EPS are from Samsung Global Newsroom (005930.KS, Jul 30 2026); FY2025 and Q1 2026 revenue from WSJ; book value per share and the FY2025 operating-profit split are author estimates pending Samsung DART filings. Values are ₩ (won). The separate listed preferred line (005935) is not shown.

Capital Allocation & Returns

Samsung's balance sheet is a genuine floor under the thesis: an estimated ~₩100tn net cash after decades of memory capex, a shareholder-return policy of roughly 50% of free cash flow, a ~₩367/quarter dividend (~0.5% yield), and, most notably, aggressive treasury-share cancellation (~₩16tn / ~87M shares carried out through H1 2026), which is per-share accretive⁷. Q2 also funded a one-off ~₩15-17tn special performance bonus (10.5% of business profit), a real but non-recurring call on the record cash flow. Korea's Value-Up reforms (mandatory cancellations, larger buybacks) are a structural tailwind to a multiple that has spent most of the past decade below 1.5x book.

The reinvestment question is different from SK hynix's. Samsung is funding a memory-capacity build alongside a foundry that still loses money, so the capital-allocation risk is that the loss-making foundry keeps absorbing memory cash flow while the Value-Up returns are real but small against a ~₩1,351tn cap. The net cash is a floor on the valuation, not a driver of it.

Sum-of-the-Parts Valuation

A conglomerate cannot be valued on one multiple, because the pieces diverge: a Buy-quality memory arm, a loss-making foundry, a commodity-hardware consumer business, a cash-generative OLED subsidiary, and a fortress balance sheet. We value each on a peer multiple, add net cash at face and the market value of listed affiliate stakes (at a holding-company discount), and deduct the Samsung Display minority⁶. The swing factor is memory, ~80% of the equity value, which we value on through-cycle earnings and haircut for the HBM-laggard, NAND-heavier mix; SK hynix's own pure-play cap has itself de-rated to ~₩1,036tn (down ~40% from the peak) as the market prices a memory-cycle top.

Why the base now sits above spot

On these inputs the base sum-of-the-parts is ~₩283,000, versus a ₩231,000 spot and a ₩471,908 Street mean (range ₩210k-₩725k)⁸. After the ~38% pullback our base now sits above spot, an inversion of the July picture: the market has sold the stock below our fair-value range. We still credit Samsung's record memory profit (memory ~₩1,600tn) without underwriting SK-hynix-plus for the HBM laggard, and we mark the money-losing foundry as option value. The gap to the Street stays wide (roughly 40% below the ₩472k mean), a contrarian stance held now on peak-cycle risk rather than on price. The analyst target distribution below shows how far consensus still sits above both spot and our base.

Scenario Model: Sum-of-the-Parts

The model is the heart of the report. Each scenario values Samsung by adding up the pieces: memory + other operating segments + net cash + affiliate stakes = equity, divided by ~6.67bn economic (common + preferred) shares. Toggle Base / Bull / Bear to load the anchor inputs; drag the two sliders (memory enterprise value, and the aggregate of the other operating segments) to see the implied price target update. The base (memory ~₩1,600tn) reconciles to ₩283,000: now above the ₩231,000 spot after the de-rating and far below the ₩472k Street mean. The memory slider is the swing factor, because memory is ~80% of the equity value.

Net cash is held at ~₩100tn across scenarios (a balance-sheet fact, not a scenario variable); affiliate stakes load per scenario (₩32-46tn, net of a holding-company discount). The implied P/B cross-check reads book value per share ~₩71,850 (est). The segment enterprise values are author estimates; the Q2 2026 figures are confirmed (Samsung, Jul 30 2026) and market data is as of Aug 7 2026.

Sensitivity grid: memory EV × other-segments EV (PT in ₩)

Risk / Reward calculator

Note: The assistant reasons from the dashboard's data snapshot and thesis sections: it does not browse the web or access real-time fundamentals beyond what's in data.js. Treat its responses as scenario-modeling support, not primary research. Author judgments on rating, PT, and probabilities remain with the analyst; the segment enterprise values are author estimates and the Q2 2026 figures are confirmed (Samsung, Jul 30 2026; market data Aug 7 2026).

Upcoming Catalysts

Risk Factors

• HBM laggard status persists. The pre-eminent risk to the bull case. If Samsung stays behind SK hynix (and Micron) on HBM4 allocation at NVIDIA, the memory discount is permanent and the market is paying for a win Samsung never earns.
• Foundry losses into 2028 (or later). Management guides annual foundry profitability only to ~2028, with 2nm yields still questioned. Persistent losses keep absorbing memory cash flow and cap the sum-of-the-parts.
• Memory cyclicality and out-year oversupply. The same nearly $600bn Korea capacity build that Samsung is helping fund seeds future oversupply; a memory price rollover reverses the record margins fast, because operating leverage cuts both ways⁹.
• Mobile margin squeeze (now realized). AI-driven memory-cost inflation pushed the Galaxy franchise to its first-ever quarterly operating loss (₩0.7tn in Q2 2026), hitting the DX profit core just as the chip division peaks; a first annual loss is now in play¹⁰.
• The de-rating is a warning. Both Samsung and SK hynix fell ~40% from their peak despite record Q2 profit, and SK hynix's cap is down to ~₩1,036tn: the memory names are already trading as if the cycle is topping, so multiple compression is a live downside if the inflection fades. At ~3.2x book Samsung is still well above its ~1.4x ten-year median³.
• Korea discount and chaebol governance. The Lee family controls Samsung through circular ownership despite a small direct stake; governance overhangs have long weighed on the multiple, and Value-Up reforms may narrow the discount only slowly.
• China competition (CXMT). CXMT is gaining fast in commodity DRAM; the threat is to Samsung's lower-margin bit base more than to HBM near-term, but it pressures the NAND-heavy mix that already discounts the memory arm.
• Frothy, leverage-heavy Kospi. Samsung is central to a record, retail-driven Kospi with record volatility and heavy margin debt; a market-wide drawdown could amplify a de-rating independent of fundamentals⁹.
• Upside risk (to our Hold). The symmetric risk: if HBM4 scales at NVIDIA faster than we assume, the memory arm re-rates toward SK hynix economics on ~2x the bit base and our base proves too conservative.

Scenario Stress Tests

All stress-test PTs are derived from the same sum-of-the-parts framework used in the interactive tool: each shock changes only the memory and/or other-segment enterprise value noted, with net cash held constant. Deltas are vs the ₩283,000 base PT. After the de-rating the base now sits above the ₩231,000 spot, so the market has already priced toward the adverse end of the range; only the ₩190,000 full bear (~18% below spot) sits under it, and the probability-weighted blend (~₩274,000) is ~19% above spot: a Hold held on cycle-peak risk, not on price.

Bull vs Bear Debate

Technical Analysis

Trader's view

• Price is off its ₩374,500 May high (~38%) after a sharp post-peak, post-earnings de-rating, but still well above the pre-rally base: a stock that re-rated hard and has given much of it back.
• Key support: the ₩225,000 area (recent range low). Our ₩283,000 sum-of-the-parts base is now overhead, not underfoot; a close below ₩210k would open the ₩190k bear.
• Key resistance: the ₩374,500 high, far above. A decisive break likely needs a genuine HBM4 allocation win at NVIDIA, not just qualification.
• Momentum: rolling over off the highs as the broader Kospi swings on record volatility and the memory names de-rate, consistent with a Hold after the market sold below our base.

Glossary & Methodology Notes

Sum-of-the-parts (SOTP)

Valuing a conglomerate by pricing each business separately on a peer multiple, then adding net cash and the market value of listed stakes (at a holding-company discount) and deducting minority interests. Used here because Samsung's segments diverge too sharply for one P/E.

HBM (High-Bandwidth Memory)

Vertically stacked DRAM bolted beside every AI GPU. The highest-value memory, where SK hynix leads and Samsung is the laggard. Samsung cleared NVIDIA HBM4 qualification only in early 2026.

DRAM vs NAND

DRAM is fast working memory (the profit core, now HBM-led); NAND is slower flash storage (more commoditized). Samsung is #1 in both combined, but its NAND-heavy mix is lower-margin than SK hynix's DRAM/HBM concentration.

Foundry vs IDM

A foundry makes chips for other companies (TSMC is the leader). Samsung is an IDM (integrated device maker) that also runs a foundry, which gives captive volume but makes fabless rivals wary of a competitor holding their designs.

The Korea discount / Value-Up

Korean large-caps have long traded at low multiples (governance, chaebol ownership). "Value-Up" is the government reform push (buybacks, mandatory treasury-share cancellation, disclosure) aimed at narrowing that discount.

Samsung Display minority

Samsung owns ~84.8% of Samsung Display (SDC). In the SOTP we value 100% of SDC on a panel-peer multiple, then take Samsung's 84.8% economic share (which is the minority adjustment).

Economic share count

The SOTP per-share divides equity by ~6.67bn shares (common 5.85bn + the separately listed preferred ~0.816bn), the right basis for total equity value. The ribbon market cap uses common shares only (₩1,351T at ₩231,000); a broader headline ~₩1,471T includes the preferred.

Methodology

• Snapshot anchor: August 7, 2026 (post Q2 2026 results, reported Jul 30 2026). Live price patches via the Cloudflare-Worker quote proxy (Yahoo 005930.KS, in KRW) on page load.
• KRW-native: all market data and operating financials are in Korean won (the KRX listing); no FX conversion is applied.
• Hybrid sourcing: Q2 2026 results (revenue, operating profit, division splits) are confirmed from Samsung Global Newsroom (Jul 30 2026); price, 52-week range, share count and consensus are Aug 7 2026 market data; FY2025 and Q1 2026 revenue from WSJ. The sum-of-the-parts segment enterprise values, foundry loss, HBM share, net cash and book value per share are author estimates flagged accordingly.
• The scenario model values Samsung sum-of-the-parts (each segment on a peer multiple, plus net cash and affiliate stakes, less the display minority). The base reconciles to ₩283,000.
• Conclusions are the author's view. Illustrative, not investment advice.