SK Hynix closed at ₩1,678,000 on Aug. 25, up 568.5% over twelve months, yet still 43.8% below its 52-week high of ₩2,987,000. On headline numbers, the Korean memory maker trades at just 7.4 times earnings, a figure that has drawn value investors to the stock. But that multiple is built on reported net income of ₩161.97 trillion (approximately $117.1 billion) that includes roughly ₩78 trillion (approximately $56.4 billion) of non-operating income — gains that have nothing to do with making memory chips.

The distortion is visible in a single line of the income statement: SK Hynix’s pre-tax margin is 109.3%. No manufacturing company can earn more before tax than it collects in revenue from operations. The gap arrives below the operating line, where ₩78.05 trillion of non-operating income — equal to 41.3% of revenue — carries reported earnings far above what the actual memory business generated. Strip that out and tax operating profit at Korea’s roughly 24% corporate rate, and normalized earnings fall to about ₩134,000 a share. At ₩1,678,000, that puts the real P/E near 12.5x, not 7.4x.

The operating business itself is extraordinary. Revenue grew 145.0% to ₩189.17 trillion (approximately $136.8 billion), with a 68.0% operating margin and 76.3% gross margin. Return on equity stands at 92.68%. SK Hynix is converting the high-bandwidth memory shortage into profit at a rate almost no manufacturer of physical goods achieves. The balance sheet backs it up: roughly ₩69 trillion (approximately $49.9 billion) of net cash at the end of the second quarter.

On Aug. 19, the company announced a ₩40 trillion buyback (approximately $28.9 billion) — about $28.6 billion — cancelling roughly 24.07 million shares, or 3.3% of the 730.5 million issued. Management described it as the largest treasury-share cancellation ever undertaken by a listed South Korean company. The program came after the stock fell 9.8% in a single session, signaling that management is buying its own equity into weakness rather than at the highs.

The Bull and Bear Cases, Priced

The analyst consensus target sits at ₩3,170,279, implying 88.9% upside. The bull case of ₩3,000,000 roughly matches the 52-week high. The bear case of ₩1,000,000 represents a 40.4% decline. The difference hinges on which earnings figure investors trust.

ScenarioPrice (KRW)Move from ₩1,678,000Headline P/ENormalized P/EAnalyst consensus target₩3,170,279+88.9%13.9×23.6xBull case₩3,000,000+78.8%13.2×22.4xToday₩1,678,000—7.4×12.5xBear case₩1,000,000−40.4%4.4×7.5×52-week low₩253,000−84.9%1.1×1.9x

Note: Normalized P/E uses EPS of roughly ₩134,000, derived from taxing EBIT of ₩128.71 trillion at approximately 24%. Headline P/E uses reported EPS of ₩227,588.

The normalized column is the one that should govern decisions. At 12.5x, today’s multiple is reasonable but unremarkable for a peak-cycle semiconductor business. The bull case at ₩3,000,000 requires paying 22.4x normalized earnings at what may prove to be the top of the memory cycle — precisely the multiple investors were paying in January before the shares halved.

Cyclicality, China, and a 2.41 Beta

Memory is the most violently cyclical business in semiconductors, and a 68% operating margin is not an equilibrium — it is a signal to every competitor and every government with an industrial policy. The mechanism is well understood: extraordinary margins attract capacity, capacity arrives with a two-to-three year lag, and pricing collapses when it does.

China is the specific vector. CXMT’s 466% Shanghai debut hit Micron and SK Hynix, though the Chinese memory maker’s prospectus contains no high-bandwidth memory product. That detail cuts both ways. CXMT arriving with capital and no HBM is not an immediate threat to SK Hynix’s most profitable franchise — but it is a direct threat in commodity DRAM, which still underpins a large share of volume. Competition does not have to reach the crown jewels to compress the average selling price.

The market has already shown how fast this repricing happens. In late July, the stock fell 14.65% in the AI chip selloff that also hit AMD. With a beta of 2.41, SK Hynix moves roughly two and a half times the market. The drawdown from ₩2,987,000 to ₩1,678,000 happened without any deterioration in reported earnings at all. A stock that can lose 44% while its profits are still rising can lose considerably more when they stop.

The concentration risk for Korea itself is not theoretical. In late July, KOSPI trading was halted after an 8% plunge as the SK Hynix ADR fell below $140. A single stock large enough to trip a national index circuit breaker is a stock whose own drawdowns become self-reinforcing: index funds sell the market, the market sells the largest constituent, and the largest constituent is the company you own. At a ₩1,218 trillion market capitalization (approximately $880.6 billion), SK Hynix is not a position that can be exited quietly when sentiment turns.

There is a second-order risk in the non-operating income itself. If a meaningful portion is foreign-exchange or mark-to-market driven, it is not merely non-recurring — it is potentially reversing. A stronger won or a drawdown in affiliate valuations would turn a tailwind into a headwind below the operating line, and reported EPS would fall faster than operating profit. That is the scenario in which a “4x forward earnings” stock re-rates upward on the multiple while the price goes down.

The won has already strengthened 11.90% against the dollar from July 1 to Aug. 21, the largest appreciation among major currencies, driven by semiconductor export flows, corporate dollar sales, and expectations of further Bank of Korea rate hikes after the central bank raised its benchmark rate to 2.75% in July. Thirteen of 20 economists surveyed expect another increase at the Aug. 27 monetary policy meeting.

Samsung Diverges on Capital Returns

The contrast with Samsung Electronics is instructive. Samsung approved a shareholder return plan of ₩90 trillion to ₩110 trillion on Aug. 21, but its shares fell 8.70% to ₩257,000 as investors digested the details. The Financial Services Commission’s 10% rule under the Financial Industry Structural Improvement Act limits how much stock Samsung Life Insurance and Samsung Fire & Marine Insurance can collectively hold, constraining buyback capacity. Analysts expect Samsung’s buyback and cancellation to reach only ₩10 trillion to ₩20 trillion, with the weight of shareholder returns shifting toward dividends — a stark contrast to SK Hynix’s ₩40 trillion buyback.

At Hot Chips 2026 at Stanford University on Aug. 23, both companies presented solutions to the memory bottleneck. Samsung proposed customized HBM designed around customer GPUs and NPUs, while SK Hynix detailed hybrid bonding technology that could enable 20-layer memory stacks, narrowing TSV circuits and expanding core die thickness by 24%.

What to Watch

Three signposts matter most. First, the disclosure of what sits in non-operating income. If the next filing shows the ₩78 trillion is largely equity-method gains from operating affiliates, the earnings are more durable than they appear and the bull case strengthens. If it is predominantly foreign exchange and mark-to-market, expect reported EPS to be far more volatile than operating profit in both directions.

Second, the buyback provides a floor, but a known one. A ₩40 trillion program against a ₩1,218 trillion market capitalization is roughly 3.3% of the company. Buybacks slow declines; they do not reverse them. A company retiring stock at 22x normalized earnings on the way up destroys value as efficiently as it creates it on the way down.

Third, the bear case arrives through DRAM pricing, not HBM. The HBM franchise is defensible for now, and CXMT has no HBM product in its prospectus. Commodity DRAM is where new capacity lands first. Watch contract DRAM pricing and inventory days rather than HBM headlines, because that is where a 68% operating margin starts compressing — and margin compression from this altitude does not need to go far to halve the earnings.

The framing for anyone sizing a position: SK Hynix is a genuinely exceptional operating business at what is plausibly a cyclical peak, trading on a multiple that looks half as expensive as it actually is. That combination rewards discipline about the entry price and punishes anyone who bought because a screen said seven times earnings.