5930 — Samsung Electronics Co., Ltd.

005930 — Samsung Electronics Co., Ltd. (KRX/KOSPI) · deep-value analysis (REFRESH)

Analysis date: September 11, 2026 · Reference price: … (005930.KS, close 09/11/2026; … on the day, on a KOSPI …) · Preferred 005935.KS: … · Market cap (separate share classes): … · Prior GBL score from tracker: 47.5% — SPECULATIVE (16.0/30, scored 09/06/2026)

Regime: REFRESH over the thesis from 08/19/2026 (verdict then: NO at …). The delta package delta-005930-20260909.md triggered the redo on the grounds of “6 unverifiable falsifiers”.

New primary source, which did not exist at the reference analysis: 2026 Half-year Business Report (Samsung Electronics, 323 pages, revised interim consolidated financial statements for the half-year ended 06/30/2026). The August report worked off the Q2’26 earnings presentation (unaudited, no notes) and explicitly declared five data gaps. Four of them are now closed at the source. All figures below are in billion KRW (bn) unless stated otherwise; 1,000 bn = 1 trillion KRW.

Executive summary (1 page: thesis, estimated value, verdict)

Thesis, unchanged in substance: Samsung is an excellent company at the peak of the most violent memory cycle in history, and the price already prices in the assumption that this state holds for several years. What has changed in three weeks is that I can now verify, at the primary source, almost everything the August report had to estimate — and the verification does not overturn the thesis, it sharpens it.

Why the analysis was redone, honestly. The “6 unverifiable falsifiers” from the triage were not a signal about the company: they were a tool artifact. The triage queried the symbol 005930, on which Yahoo returns 404 — Quote not found; with the correct symbol 005930.KS all six metrics exist. All six falsifiers are now evaluated and none of them triggers (table in the verdict chapter). The thesis survives its own falsification tests.

What I inherit unchanged, because the filing did not move it: the business structure and the industry moat diagnosis (DRAM oligopoly, capital barrier above USD 20 bn/year); the normalized mid-cycle base built bottom-up (consolidated revenue 440,000 bn, operating margin 17.7%, EBIT 78,000, owner earnings 52,000 bn/year); the five valuation models and their intrinsic values; the Monte Carlo assumptions (oe …, g1 5%, r 10%, gt 2.5%, nd −…); the balance sheet as of 06/30/2026, including net cash of 167,590 bn, since there is no new filing (next reporting: 10/28/2026). Assumptions were rerun identically, at the new price.

The three new facts that matter, in order of importance.

1. Samsung’s HBM share jumped from 21% to 33% in one quarter (Counterpoint Research, Q2’26; SK hynix 58%→50%, Micron 21%→18%), on mass production of HBM4 on 1c DRAM for the Nvidia Vera Rubin platform. The August report worked with “Samsung 17%, third place” — a figure that had gone stale. The single structural weakness in the old thesis — moat erosion exactly where the profit pool is fattest — has reversed. In total DRAM, Samsung remains #1 with 38% (SK hynix 25%, Micron 24%, CXMT 10%).

2. The board approved on 08/21/2026 — two days after the reference analysis — the 2026 shareholder return: 90,000-…, five times the prior record (20,300 bn in 2020), cumulative 2024-2026 of 120,000-140,000 bn. The catalyst the August report called “the most important event to watch on the 2-3 month horizon” has triggered — and the stock fell. Two details change the reading: the 15,000 bn tranche (53.29 mil shares, 08/24-11/21/2026) is NOT canceled — it is held as treasury shares to compensate employees and executives; and the company’s own arithmetic (50% of cumulative free cash flow) implies a 2026 FCF of 187,600-227,600 bn, below the 240,000 assumed in August.

3. The revised half-year report closes gap #1 from August — the inventory breakdown — and closes it in the company’s favor. Over the six months, finished goods grew …, work in progress …, raw materials …: finished goods are the slowest-growing component, and their share of inventory fell from 23.3% to 21.0%. The only O’Glove test that was flashing red in August is resolved benignly.

The new finding that cuts the other way, and it is the most important number in this refresh. Reported free cash flow is correct as a definition, but a third of it is settlement timing lag, not distributable cash. In H1’26 the company recorded tax expense of 34,418 bn and paid 6,127 (a gap of 28,291), while the share-based payment reserve rose from 366 to 15,853 bn (+15,487) — the record year’s performance bonus is being settled in shares, not cash, and the August-November buyback of 15,000 bn is exactly what funds it. Cleaned of these two gaps, TTM free cash flow falls from 140,486 to 97,495 bn, and the yield from 8.51% to 5.91%.

Estimated value — the range, not the point. Five models, inherited intrinsic values (only EPV recalibrated on the company-guided effective tax rate of 22.5%):

Monte Carlo, 20,000 scenarios, assumptions identical to August: median … (was …), P10 …, P90 …, probability of undervaluation 7.0% (was 9.9%). Median intrinsic value is unchanged at … — all the deterioration comes from price.

Verdict: NO at … on the ordinary share — but the preferred 005935.KS at … has entered the accumulation band. This is the new operational conclusion. EPS is identical across the two classes, now confirmed in note 23 of the half-year report (… in Q2’26, 17,950 in H1’26, for both ordinary AND preferred), the dividend per share is identical (… in Q2’26), so intrinsic value per share is the same. At a 25.5% discount to the ordinary, the base model gives the preferred a margin of safety of … instead of …, and the median of the five models … instead of …. The thresholds remain: zero position above …; half-unit between 160,000 and 200,000; full position below 140,000. The preferred at 193,300 is the first instrument to enter the “half-unit” band, with the caveat that historical 30-day volatility is 96.6% annualized — sizing must respect that.