Samsung's profit jumped 1,800%. Its own chip chief says only 30% of that was Samsung's doing
Samsung just reported one of the most profitable quarters any company has ever had — roughly $62 billion of operating profit, up more than 1,800 per cent in a year. Then its own chip chief told employees that only about 30 per cent of it was Samsung's doing. The rest belongs to a memory shortage the company says will last until 2028. An analysis of a windfall, honestly labelled.
Every figure in this article is drawn from Samsung Electronics' earnings releases, or from reporting by CNBC, Fortune, Bloomberg, the Seoul Economic Daily, UPI, TechCrunch and the semiconductor trade press as attributed. Korean won figures are converted at approximately 1,440 won per U.S. dollar and marked as approximate. Sections marked as analysis are identified as such.
For the quarter ended in June, Samsung Electronics reported revenue of 171.5 trillion won — roughly $119 billion, an all-time high — and operating profit of 89.5 trillion won, roughly $62 billion, up more than 1,800 per cent from a year earlier. It was the third consecutive record quarter. The implied operating margin, around 52 per cent, is the kind of number commodity manufacturers are never supposed to see.
The most remarkable statement about those results came from inside the company. At an internal meeting reported by the Seoul Economic Daily, Jun Young-hyun, the vice-chairman who runs Samsung's chip division, told managers: "Looking at it coldly, the portion of these results that our own capabilities contributed is about 30 per cent" — attributing the rest to the memory supercycle, and warning against complacency: "Since when did Samsung Electronics work with the goal of beating SK Hynix?"
It is the rarest artefact in corporate life: an executive discounting his own triumph, accurately.
The shortage that prints money
The mechanics of the windfall are brutally simple. The AI buildout consumed the world's memory supply, and prices went vertical: conventional DRAM contract prices rose 90 to 95 per cent in a single quarter early this year, per TrendForce, and Samsung's own filing disclosed blended memory prices up 146 per cent against last year's average. Nearly all of the quarter's profit — 89.2 of the 89.5 trillion won — came from the semiconductor division. Samsung's first quarter alone out-earned its entire fiscal 2025.
Nor does the company expect relief. Samsung told investors the chip crunch will last until 2028, per CNBC, with customers now seeking multi-year contracts — behaviour unheard of in a business built on spot pricing. OpenAI has gone furthest: an October letter of intent makes Samsung a strategic memory partner for the Stargate project, contemplating up to 900,000 DRAM wafers a month, with affiliates exploring AI data centres — including floating ones.
The context makes the vertigo sharper. Memory has historically been the most punishing commodity business in technology — a cycle of gluts and crashes in which Samsung's scale was its only durable advantage. Fiscal 2025, a perfectly respectable year by old standards, produced 43.6 trillion won of operating profit across twelve months. The division now produces more than that in a single quarter, from largely the same factories.
The catch-up that finally worked
What makes the boom bittersweet in Suwon is that Samsung entered it as the laggard. Through the first half of 2025, SK Hynix — the smaller Korean rival — held roughly 62 per cent of the high-bandwidth-memory shipments feeding Nvidia's accelerators, against Samsung's 17. Samsung's HBM3E spent some eighteen months failing Nvidia's qualification tests before finally passing last September, third in line.
The next generation went better. HBM4 samples reached customers early; mass production began in February; and in June, Nvidia's chief executive confirmed Samsung certified for the Vera Rubin platform alongside its rivals. Samsung says HBM4 revenue will more than triple this quarter and exceed 60 per cent of its high-bandwidth-memory sales in the second half — and that its 2026 supply is already sold out. Catching up in one generation, in the most demanding product in the industry, is the part of the quarter Samsung's own capabilities can fairly claim.
What the boom cannot fix
Two businesses show why Jun's 30 per cent arithmetic is honest rather than modest. The foundry — Samsung's decade-long attempt to challenge TSMC in making chips for others — is still losing money: roughly 6 trillion won last year, with losses projected to narrow but persist through this year and profitability pushed to 2027 or 2028. Yields on its 2-nanometre process have hovered near 55 per cent, per trade reporting — below comfortable mass-production thresholds. The marquee win is real — Tesla's $16.5 billion, eight-year deal to build its AI6 chip in Taylor, Texas, which Elon Musk said was "hard to overstate" — but the mass-production timeline has slipped toward 2027.
And then there is the strangest line in the report: Samsung's own phone division lost money — about 0.7 trillion won — in the very quarter the company set profit records, because of "elevated component costs." The components in question are memory chips. Samsung's handsets are being squeezed by Samsung's own prices; foldable phones went up $100 across the line. The supercycle giveth to the third floor and taketh from the fifth.
One long shadow did lift: in July 2025, Korea's Supreme Court issued a final acquittal of executive chairman Lee Jae-yong on the merger-related charges that had followed him for nearly nine years — giving the company its first stretch of unencumbered leadership in a decade.
Analysis: honesty as a strategy
The following section is analysis, drawn from the reported facts above.
Every commodity boom produces companies that mistake the cycle for themselves. Samsung's chip chief has pre-emptively refused the confusion, and the distinction he drew is the correct analytical frame: the 146 per cent price inflation is the market's gift and will leave when supply catches up — the company itself dates the party's end at 2028. What Samsung owns is the 30 per cent: the HBM4 catch-up, the sold-out order book, the OpenAI partnership, the Tesla foundry deal.
The unresolved risks map exactly onto that split. If memory prices normalise before the foundry turns profitable and before HBM share gains stick, Samsung reverts to what it was in 2023 — a cyclical giant with a broken second act and phones that make more headlines than margin. If the catch-up holds, the windfall years will have financed the transformation at zero cost. Investors' reaction to the record quarter — volatile rather than euphoric, per conflicting same-day reports — suggests the market is pricing precisely this question and has not decided.
Either way, the honest 30 per cent may prove the most valuable disclosure of the earnings season. Somewhere in every AI-boom beneficiary's results is a split between what the wave did and what the company did. Samsung is the only one that published the ratio.
Jun's question to his managers — since when was beating SK Hynix the goal? — points at the real finish line. The prize in front of Samsung is not a share statistic; it is becoming indispensable to the AI buildout at every layer at once: the memory beside the GPU, the foundry beneath the custom chip, the packaging that joins them. No other company on Earth is positioned to attempt all three. That, not the 1,814 per cent, is what the next three years will actually decide.
Sources
- Samsung Electronics Q2 2026 results announcement, July 30, 2026 — revenue KRW 171.5T, operating profit KRW 89.5T, semiconductor division profit KRW 89.2T, mobile division loss on elevated component costs, HBM4 commentary; Fortune and CNBC same-day coverage, including the crunch-until-2028 remarks.
- Samsung Q1 2026 and Q4/FY2025 announcements — Q1 operating profit KRW 57.2T exceeding all of fiscal 2025 (KRW 43.6T on revenue of KRW 333.6T).
- Seoul Economic Daily, August 16, 2026 — Jun Young-hyun's internal remarks (the ~30% attribution and SK Hynix comment); Seoul Economic Daily, August 2026 — H1 2025 HBM share figures.
- TrendForce, May 18, 2026 — Q1 DRAM/NAND contract-price surges and Samsung's disclosed 146% blended price rise; TrendForce, September 2025 — the HBM3E qualification after an 18-month setback.
- Yahoo Finance/Nvidia remarks, June 5, 2026 — Samsung, SK Hynix and Micron certified for Vera Rubin HBM4; DigiTimes, July 30, 2026 — HBM4 tripling and second-half mix from the earnings call; KED Global, October 2025 — 2026 HBM supply sold out.
- Bloomberg and KED Global, July 28, 2025 — the $16.5B Tesla AI6 foundry agreement and Musk's remarks; Electrek and TrendForce, March 2026 — Taylor timeline slippage; trade reporting on foundry losses, 2nm yields and 2027–28 profitability targets.
- Samsung and OpenAI announcements and TechCrunch, October 1, 2025 — the Stargate memory partnership, up to 900,000 wafers monthly, and affiliated data-centre plans.
- UPI, July 17, 2025 — the Supreme Court's final acquittal of Lee Jae-yong.
Won-to-dollar conversions are approximate. The ~52% operating margin is implied by Samsung's own reported figures. Same-day share-price reactions to the results were reported inconsistently across outlets and are characterised as volatile rather than directional. Analyst HBM market-share splits are estimates, not company disclosures.