Three Days, $107 Billion: What Samsung and SK Hynix Just Told the Market

Korea's Memory Duopoly · Part 1 of 3

Three Days, $107 Billion: What Samsung and SK Hynix Just Told the Market

Samsung and SK Hynix announce $107bn in shareholder returns

Nearly every AI accelerator shipping today contains memory made by one of two Korean companies. Last week both of them announced the largest shareholder returns in their history — within 72 hours of each other. The size is not the interesting part.

Key points

  • SK Hynix and Samsung Electronics announced roughly $107 billion in combined shareholder returns between 19 and 21 August.
  • Samsung's figure alone is more than five times its previous record, set in 2020.
  • The trigger for Korea's index rebound was not earnings. Both companies were already posting record profits while their shares fell.
  • On the day the KOSPI recovered, the KOSDAQ small-cap index fell 4.63% — a divergence worth understanding.

If you follow AI infrastructure, you already know the supply chain narrows to a very small number of firms. High-bandwidth memory — the stacked DRAM that sits beside the GPU die and feeds it data — is made at scale by three companies. Two of them are Korean, and together they dominate the category.

That concentration has a second-order effect most foreign investors miss: it makes Korea's benchmark index behave like a two-stock portfolio.

Which is why the last three months have looked the way they have.

A benchmark that is really two companies

From mid-2025, AI datacenter demand pulled DRAM capacity toward HBM production. General-purpose DRAM supply tightened, memory prices climbed, and the two largest DRAM makers re-rated violently. Over roughly a year, Samsung Electronics gained more than 400% and SK Hynix approached 1,000%.

Both are also the first- and second-largest constituents of the KOSPI by market value. So when they move, the index moves — and in 2026 they moved a great deal.

KOSPI 2026 · KEY INFLECTION POINTS 9,000 8,000 7,000 6,000 Approached 9,000 June Broke below 6,000 July 31 Jul rally 3 Aug −5.12% Back above 6,900 21 Aug Schematic of major inflection points, not a daily close chart. Green denotes advances, red declines (Western convention).
Figure 1. Note the gradients. These were not orderly corrections — the index moved 4–5% in single sessions repeatedly. On 3 August the KOSPI fell 5.12% to close at 6,257.45.

July's selloff was not an earnings problem

Over July, Samsung Electronics fell 22.2% and SK Hynix fell 34.9%.

During the same stretch, SK Hynix reported first-half revenue of ₩131.9tn (roughly $94bn) and operating profit of ₩98.2tn (roughly $70bn). That is an operating margin of 74% for a six-month period.

Record profitability, and the stock lost a third of its value. So the explanation has to sit somewhere other than fundamentals.

The leverage unwind

A significant piece was regulatory. On 31 July, Korean financial authorities tightened rules on single-stock leveraged and inverse ETFs. Daily turnover in that segment collapsed from around ₩12tn to the ₩1tn range. A large pool of leveraged money positioned in semiconductor large caps had to unwind, and it unwound quickly.

For foreign investors, the takeaway is structural rather than fundamental: this was a positioning unwind, not a demand shock. Separating the two is the entire exercise in this cycle.

The three days that changed it

Then last week, the direction flipped — again without an earnings catalyst.

19 August — SK Hynix commits to a ₩40tn buyback and cancellation

SK Hynix's board approved the repurchase of 24.07 million common shares for outright cancellation, worth roughly ₩40tn ($29bn). It is the largest share cancellation by a listed Korean company on record. The buying is scheduled to complete within about three months, by November. The company also committed to returning at least 50% of cumulative free cash flow across its 2025–2027 policy window.

The stock rose 12.7% the following session.

21 August — Samsung answers with up to ₩110tn

Two days later Samsung's board disclosed returns of up to ₩110tn ($79bn) for the year, under a principle of returning 50% of cumulative free cash flow across 2024–2026. Roughly ₩30tn of that arrives as cash dividends in the third quarter.

5.4× Samsung's 2026 shareholder return programme relative to its previous record of ₩20.3tn, set in 2020. The order of magnitude has changed, not just the number.
SHAREHOLDER RETURNS COMPARED · USD BN Samsung 2026 maximum disclosed $79bn SK Hynix 2026 buyback + cancellation $29bn Samsung 2020 previous record $14.5bn Converted at approximately ₩1,400/USD. The two 2026 figures are defined differently — Samsung discloses an annual maximum, SK Hynix a repurchase-and-cancellation amount — so treat this as scale, not a like-for-like comparison.
Figure 2. Combined, the two programmes approach $107bn. For context, that is comparable to the annual buyback budgets of the very largest US technology companies.

Why cancellation reads differently than a dividend

Anglo-American investors are used to buybacks. What is worth flagging here is that a Korean cancellation is a stronger commitment than the buyback authorisations many US firms announce and then partially execute.

MECHANISM Free cash flow generated by operations Open-market repurchase company becomes a buyer Cancellation shares retired permanently Share count falls EPS accretion Unlike a dividend, this cannot be quietly reduced next year. And during the repurchase window, the issuer is an active bid in its own stock — SK Hynix compresses that window into roughly three months.
Figure 3. Three distinct effects: permanent share-count reduction, a mechanical bid during execution, and a management signal about future cash generation.

The signalling matters most. Committing tens of billions to retiring stock is a statement that the board expects the cash to keep arriving. SK Hynix's balance sheet supports the claim: net cash stood near ₩69.4tn ($50bn) at the end of June, up more than ₩34tn in a single quarter.

The counter-argument deserves equal weight. Capital returned is capital not deployed. Samsung spent over ₩55tn ($39bn) on R&D and capex in the first half alone, in a market where capacity is the constraint. Returning cash during a capacity-limited boom is a choice with a cost.

The detail almost every headline skipped

On 21 August, the day the KOSPI reclaimed 6,900, Korea's small-cap KOSDAQ index fell 4.63%.

The two indices moved in opposite directions on the same session. This is not noise. When two constituents carry that much index weight, their recovery lifts the benchmark regardless of what the other 900-plus listed companies are doing — and money rotating back into semiconductors has to come from somewhere.

A rising index and a rising market are not the same thing. In a benchmark this concentrated, the headline number can tell you almost nothing about breadth. Part 3 of this series takes that apart.

What to watch from here

SignalWhy it matters
Execution paceSK Hynix intends to complete repurchases by November. Actual daily volumes will show whether the bid is as supportive as expected.
Samsung's Q3 detailThe ₩110tn figure is a maximum. Composition between dividends and cancellation is still to be specified.
Index breadthIf the KOSPI advances while the KOSDAQ falls, concentration is deepening rather than resolving.
Memory pricingEverything above rests on the DRAM cycle. Contract pricing is the upstream variable.

Disclosure — This article summarises publicly available corporate filings and press reporting for informational purposes. It is not investment advice and does not recommend the purchase or sale of any security. Figures are current as of 23 August 2026 and are subject to change. Won-to-dollar conversions are approximate, at roughly ₩1,400/USD. Readers should conduct their own research and consider consulting a licensed financial adviser.

Comments

Popular posts from this blog

US Open 2026 Underdog Stories: 5 Shocks From Week 1

Why the KOSPI and KOSDAQ decoupled, what index concentration does to foreign investors benchmarking Korea, and which indicators actually describe market breadth.