SK hynix Approved ₩54 Trillion of New Fabs Twelve Days Before Its ₩40 Trillion Buyback. The Buyback Is Not Capacity Discipline.

On August 19 SK hynix announced the largest buyback in Korean corporate history: ₩40 trillion, about 24.07 million shares — 3.3% of shares issued — repurchased between August 20 and November 19 and then cancelled outright rather than held in treasury. The company simultaneously raised its 2025–2027 shareholder-return target from "within 50%" of cumulative free cash flow to "more than 50%", and said the ₩40 trillion is not the last step.

The reading that spread immediately was the interesting one, and it is the one worth testing. A memory maker at the top of a cycle, printing a 76.3% operating margin, having nearly doubled its net cash in three months, chooses to retire 3.3% of itself instead of building fabs. Every memory cycle in living memory ended when prices rose, everyone built, and the new capacity arrived together. A company declining to build at the peak would be the single most valuable piece of evidence available about whether this cycle breaks that pattern.

It would be. It is not what happened — and the chart above is the whole correction: the capacity commitment is the larger of the two, and it came first.

The other board meeting

Twelve days before the buyback, on August 7, 2026, the same board approved ₩54.3 trillion of new fab investment: ₩35.2 trillion for the Yongin Y2 plant, which will make DRAM and next-generation HBM, and ₩19.1 trillion for Cheongju M17, for NAND. Both sit inside the ₩1,100 trillion mid-to-long-term plan the company laid out in June, of which ₩600 trillion is earmarked for Yongin.

So the two decisions belong to the same fortnight, and the capacity one is the larger:

Capacity got 1.36× what shareholders got. The buyback is not the company choosing returns over supply; it is the smaller half of a fortnight in which SK hynix committed the won equivalent of $61.9 billion to doing both.

The operating record says the same thing. First-half 2026 capital expenditure was ₩18.33 trillion, up 72.6% from ₩10.62 trillion a year earlier. And the company's own Q2 slides guide full-year 2026 capex to the high ₩40 trillion range, with plans to pull forward M15X mass production and accelerate Yongin Fab1 in early 2027. SK hynix is spending roughly a buyback's worth on capacity every year, and it just raised that number.

Why it can do both, which is the boring answer and the correct one

The premise of the discipline reading is that ₩40 trillion is a lot of money to a memory company. At this point in this cycle it is not.

The buyback is 50.4% of one quarter's revenue, 66.1% of one quarter's operating profit, and 57.7% of closing net cash — and net cash roughly doubled inside that same quarter. A company generating ₩60 trillion of operating profit a quarter does not face a trade-off between ₩40 trillion of buyback and ₩54 trillion of fabs. It faces a sequencing question, and it answered it twice in twelve days.

Note also what the company said about the shape of the spend: capacity additions are phased, with execution tied to confirmed customer demand. That is discipline language, and it is real — but it is discipline about timing the ramp, not about declining to build. Those are different claims and only one of them supports the bull case.

There is a second reading of the buyback that has nothing to do with capacity at all. It was announced the morning after the stock fell 9.8%, in a week the KOSPI halted twice, and the release explains that intrinsic value is "not fully reflected in its current stock price." A large repurchase announced into a drawdown, justified in writing by an undervalued share price, is also what a board does when it wants the tape to stop falling. Both readings can be true. Neither is evidence about supply.

What this does to the Micron model

This is where it matters, because the subject was never SK hynix.

The single load-bearing assumption in our Micron model is whether the industry adds capacity into the shortage. The bull case says so in as many words — it is "a claim about the industry's capacity discipline, and the last four cycles all ended when that discipline broke." The bear case is the same sentence with the discipline breaking, and it names a year: "the glut arrives early. Everyone's new capacity lands together in 2028."

The spread between those cases is the widest on this site:

Now place the announcement. Yongin Y2 and Cheongju M17 are reported to come online in 2028 and 2029 at the earliest. The second-largest DRAM maker in the world has just approved ₩54.3 trillion of capacity that arrives in the year the bear case names, while guiding current-year capex up 45% and first-half capex up 73%.

So the evidence runs the other way from how the buyback was read. It does not confirm the bull case's central assumption; it is the first hard, dated, board-approved evidence against it. One company's fab approval does not settle an industry's discipline — Samsung is not tracked here and Micron's own capex guidance is a separate question — but it is a fact with a number and a date attached, which is more than the discipline argument has had on either side.

The model does not change on this. The bear case already assumes exactly this: new capacity approved now, landing together in 2028. What changes is which case the evidence supports, and a reader holding the bull case because "nobody is building" should know that on August 7 somebody approved ₩54.3 trillion of building. The fair values above are the model's, unchanged; it is the weight on them that moves.

The pattern reading also deserves correcting. Sandisk's $14B repurchase authorisation, which we covered in its Q4 analysis, sat alongside our framing of it as "the kind of capital allocation that is very hard to reverse." Two large memory buybacks in a month is a fact about capital returns, not a fact about capacity — and treating a buyback as a proxy for restraint is precisely the inference this piece exists to break.

A ₩2.3 billion difference from nothing but arithmetic

One smaller thing, because it recurs in every headline about this announcement.

Every wire is quoting the buyback at $28.6 billion, at spot. Our SK hynix coverage converts at ₩1,523.5 per US$1.00, the rate the company's own prospectus states it uses for all conversions, applied throughout so that the series is constant-currency and the growth shown is operating growth rather than a mix of operations and the won. At that rate ₩40 trillion is $26.26 billion.

The gap is $2.34 billion, from nothing but the choice of rate, on a number being reported to three significant figures as the largest buyback in Korean history. Every dollar figure in this article uses the issuer rate. When you see this buyback sized against a US company's, check which won you are being sold.

What to watch

1. Micron's own capex guidance at its next report. SK hynix has now put a number and a date on its capacity. Micron's answer is the other half of the discipline question, and it is the input our model is most exposed to.
2. Whether the Yongin Y2 and Cheongju M17 schedules hold. Approved in August 2026, online 2028–2029. A slip pushes supply past the bear case's date; an acceleration pulls it forward. The fab schedule is the bull-bear switch.
3. The Q3 return announcement. SK hynix said the ₩40 trillion is not the final step and that more comes with third-quarter earnings. The number to watch is not its size but whether capex guidance moves with it.
4. The cumulative free-cash-flow base for 2025–2027. "More than 50%" is a direction until the denominator is published. Without it the policy change cannot be sized.
5. What a Korean treasury-share cancellation does for ADS holders. The repurchase is of Korean common shares and the US listing is a month old. Anyone holding the ADS needs to know whether 3.3% fewer shares is 3.3% more of the company for them.

SK hynix's ₩40 trillion buyback — 24.07 million shares, about 3.3% of shares issued, repurchase window August 20 to November 19, cancellation on completion, and the 2025–2027 return target raised from "within 50%" to "more than 50%" of cumulative free cash flow — is from the company's August 19, 2026 announcement as carried by Bloomberg, the Korea Herald, the Korea Times and Korea JoongAng Daily; we have not read the Korean regulatory filing directly. The August 7, 2026 board approvals of ₩35.2 trillion for Yongin Y2 and ₩19.1 trillion for Cheongju M17, the ₩1,100 trillion June long-term plan, and first-half 2026 capex of ₩18.33 trillion against ₩10.62 trillion a year earlier are from press reporting of the board resolution and the half-year figures, not from a filing we hold. Full-year 2026 capex guidance of "the high ₩40 trillion range", the M15X and Yongin Fab1 timing, and quarter-end cash of ₩87.96 trillion against ₩18.59 trillion of debt are from SK hynix's own Q2 2026 results presentation. The 2028–2029 come-online dates for Yongin and Cheongju are press characterisations rather than company guidance and are the softest figures here. June-quarter revenue of ₩79.3 trillion and operating profit of ₩60.5 trillion are as stored on this site from the July 29 preliminary-results 6-K, per our SK hynix coverage. Every dollar figure converts at ₩1,523.5 per US$1.00, the issuer's own prospectus rate, which is why they differ from wire figures quoted at spot. Micron scenario fair values are computed from our model as committed, against a stored price of $971.66 — a build-time snapshot as of August 19, not the live price. Sandisk's $14B authorisation is from its Q4 FY2026 analysis. Samsung is not tracked on this site, which bounds how far any industry-wide claim here can go.