Samsung Earned a Record ₩89.5trn and Fell 21.9% in the Same Month (August 2026): How Korea's 2x Single-Stock ETFs Amplified a Memory Selloff

On 30 July 2026 Samsung Electronics reported the best quarter in its history — consolidated revenue of ₩171.5 trillion and operating profit of ₩89.5 trillion, both all-time highs, roughly $119 billion and $62 billion, with operating profit up about nineteen-fold year on year. The stock fell 21.89% over the month that contained that report. Its supplier-cum-rival SK hynix posted operating profit of ₩60.5426 trillion on revenue of ₩79.3187 trillion — a 76% operating margin, up 557% year on year — and fell 34.47% in July. The KOSPI dropped 22.18% for the month, its worst since 2008, then rose more than 17% in a single session on Friday 31 July, then closed Monday 3 August down 5.13% at 6,257.41 with both chipmakers off about 8.8%. Records in the accounts, a crash in the tape, and a 17% up-day in between: that combination is not a story about memory demand. It is a story about who was holding the shares, and with how much borrowed exposure.

The instrument at the centre of it did not exist ten weeks ago. Korea approved single-stock leveraged ETFs — funds targeting twice the daily move of one company — on 27 May 2026, and retail investors bought a net ₩14 trillion of them, about $9.4 billion, against roughly ₩2 trillion from foreign investors. When the memory trade turned over on 23 June, those funds did what daily-reset leverage always does: they decayed faster than the underlying fell, and sold into every decline to restore their multiple.

What actually happened: two sets of numbers that refuse to agree

Take the reported results first, because they are not in dispute and they came from the companies themselves.

Samsung's Q2 2026 release shows consolidated revenue of ₩171.5 trillion, up 28% quarter on quarter, and operating profit of ₩89.5 trillion — both all-time highs. The Device Solutions division, which contains memory, produced ₩127.5 trillion of revenue and ₩89.2 trillion of operating profit. Read those last two figures against the group total and the arithmetic is stark: essentially the entire company's profit came from chips. Fortune put the group operating profit at about $62 billion, a nineteen-fold increase on the year, and revenue at roughly $119 billion.

SK hynix's own release reported revenue of ₩79.3187 trillion and operating profit of ₩60.5426 trillion — up 557% year on year at a 76% operating margin. The company said both DRAM and NAND prices rose significantly quarter on quarter, that HBM4 mass shipments began, and that customer demand exceeds supply capabilities.

Now the tape. Across July, the KOSPI fell 22.18% — its worst month since 2008 — with Samsung down 21.89% and SK hynix down 34.47%. On Friday 31 July the index rose more than 17% in a single session, a move CNBC characterised as a record rebound from a meltdown, helped by renewed confidence in AI infrastructure spending after Microsoft's results. Then on Monday 3 August it closed down 5.13% at 6,257.41, Samsung down about 8.8% at ₩239,500 and SK hynix down about 8.8%, according to Seoul Economic Daily.

That table is the single most useful thing on this page. A −34% month inside a +516% year is not a business collapsing; it is a crowded position being cut. And the direction of the crowd is visible in the flow data: on 3 August foreign investors sold a net ₩1.1104 trillion and institutions ₩247.1 billion, while domestic retail investors bought a net ₩1.3185 trillion.

The amplifier: what a 2x daily-reset fund is obliged to do

A single-stock leveraged ETF promises twice the daily move of one company. That word does all the damage, because holding a constant daily multiple against a net asset value that changes every day forces the fund to rebalance in the same direction as the market — selling exposure after a fall, buying it after a rise.

Two consequences follow, and neither requires anyone to have an opinion about HBM4.

The first is volatility decay, and the arithmetic is worth doing once by hand. A stock that falls 10% and then rises 11.1% is exactly flat. A 2x daily fund tracking it falls 20%, then rises 22.2%, and ends at 0.80 × 1.222 = 0.978 — down 2.2% on a round trip that cost the underlying holder nothing. Chain a month of double-digit daily swings together and the divergence stops looking like a rounding error. That is how the Samsung KODEX SK Hynix Single Stock Leverage fund came to be down more than 80% from its 23 June peak while SK hynix fell about 34% in July; the same fund had already lost roughly 45% between its late-May debut and mid-July.

The second is reflexivity. The rebalancing order is a market-direction trade in a single, heavily weighted stock, so the product does not merely track the move — it contributes to it. In an index where two names carry an outsized weight, a large enough leveraged complex on those two names becomes an index mechanism.

Korean authorities responded by tightening access rather than withdrawing the products. Investors must now post a minimum cash margin of ₩300,000 — about $203, roughly ten times the previous effective threshold near ₩30,000 — and proposals under discussion include cutting permitted leverage from 2x toward 1.5x or 1x alongside stricter margin rules and investment caps. Financial Supervisory Service governor Lee Chan-jin had voiced regret about the original approval as early as 22 June: "Maybe I should have lain down on the floor to block it. I personally regret (I didn't)," he said, in remarks reported by Yahoo Finance. Retail losses across leveraged ETFs have been estimated by Citi at about $38.7 billion, according to reporting by the Korea JoongAng Daily. CNBC documented the retail reaction directly.

What genuinely changed in the fundamentals: the second derivative

There is a real change underneath the mechanical one, and conflating the two is the main analytical error available here.

Memory is a commodity business, and commodity equities trade the rate of change of prices rather than the level of profits. On that measure, the peak of the acceleration is now visible in published forecasts. TrendForce expects conventional DRAM contract prices to rise 13–18% quarter on quarter in 3Q26 and NAND flash 10–15% — still large increases, but a marked slowdown from earlier in the year, and the stated reason is demand-side: record-high contract prices mean PC and smartphone customers "are reaching their affordability limit", compounded by a higher comparison base.

The supply side has an equally interesting wrinkle. HBM is what pulled capacity away from ordinary memory in the first place — it consumes roughly three times the wafer capacity of DDR5 per gigabyte, so each wafer redirected to AI removes a multiple of that in commodity supply. But TrendForce also notes that HBM wafer revenue was overtaken by DDR5 64GB RDIMM in 1Q26, pushing HBM profitability below conventional DRAM and giving suppliers a reason to reallocate capacity back. Its allocation estimates still trend the other way over time — HBM taking roughly 18% of DRAM wafer input at end-2025, 22% in 2026 and 30% in 2027 — but the per-wafer economics that drive quarterly decisions have stopped pointing in one direction only. A shortage that becomes profitable to relieve is a shortage with a clock on it.

None of that is the same as the shortage ending. Samsung's own outlook says supply constraints are expected to continue and that it expects the market to remain undersupplied through year-end, and Micron — the US-listed member of the group — reported revenue of $41.46 billion with DRAM at $31.3 billion, 76% of the total, and guided the following quarter to $50.0 billion ± $1.0 billion at roughly 86% gross margin.

Where the cost lands: the same shortage, the other side of the ledger

The most instructive line in Samsung's release is not the record. It is the loss.

MX & Networks — the division that sells Galaxy phones — reported ₩33.2 trillion of revenue and an operating loss of ₩0.7 trillion, with the company attributing the decline to "elevated component cost pressures across the industry". Visual Display and Digital Appliances, on ₩14.5 trillion of revenue, also recorded a slight operating loss. One company, one quarter, one input: memory made ₩89.2 trillion on the sell side and destroyed the margin of its own device businesses on the buy side.

That is the cleanest available illustration of a transfer happening across the hardware industry, and it is the same mechanism already visible at Apple, where guidance rather than reported margin absorbed the shock. Gartner estimated in February that surging memory costs would reduce global PC and smartphone shipments in 2026, with a combined cost increase near 130% by year-end. The chain runs: AI capacity demand → memory reallocated to HBM → commodity DRAM and NAND repriced → device bills of materials rise → either margins compress or retail prices rise, which is where a goods-price impulse eventually meets the inflation data central banks read.

The channel to what you actually trade

The direct index channel is US-listed memory and its customers, not Seoul: Micron sits inside the US indices, and the AI-capex complex that buys HBM sits inside NAS100 and US500. That is why the 30–31 July rebound in memory names traced back to a hyperscaler's results rather than to anything Korean, and why the capital-expenditure line in cloud earnings is the variable to watch rather than the KOSPI chart.

The currency channel is thin and should be labelled as such. The won is not among the eight currencies the meter scores, and nothing here moves an interest-rate differential, a growth read or a terms-of-trade balance. The one live link is risk sentiment — one of the five factors — the conduit that carries equity shocks toward the yen and the Swiss franc and away from the Australian and New Zealand dollars. But apply the lesson from the mechanism: an index move sourced from a margin requirement is a weaker risk-sentiment input than an identically sized move sourced from a demand shock. The size of the candle is not the size of the information.

What would change the picture

Four observable things, in rough order of how much they would tell you:

1. Contract prices, not spot chatter. If quarterly contract price increases decelerate further from the 13–18% DRAM range, the cycle argument strengthens on its own terms. If they re-accelerate, the July drawdown was positioning and nothing else.
2. Hyperscaler capital expenditure guidance. Memory demand is a derivative of AI build-out budgets. Committed spending is the demand curve; anything else is inference.
3. Capacity reallocation between HBM and commodity DRAM. Watch whether suppliers move wafers back toward conventional DRAM now that its per-wafer economics have improved relative to HBM. That is the supply-side release valve.
4. The leveraged ETF complex itself. Assets, permitted leverage and margin thresholds determine how much mechanical amplification remains in the next move. A smaller complex means a quieter tape for the same news.

What none of those tell you is where any of these securities go next, and that is the point of separating the layers. A record quarter and a 22% monthly decline can both be true, in the same month, in the same company, for reasons that never meet: one is accounting, the other is positioning, and only the first one was reported. Read more about how this site approaches that separation on the about page.

Educational macro context only — not investment advice.