Samsung and SK hynix Just Promised the Next Three Years
Two Korean boards met 48 hours apart last week and approved the largest shareholder payouts in the country's corporate history. SK hynix signed off on a 40 trillion won buyback-and-cancellation on August 19. Samsung Electronics followed on August 21 with a 2026 return program it sizes at 90 to 110 trillion won.
The easy read is that memory makers have more cash than they know what to do with. That read misses what the filings actually commit to. Both numbers are derived from a three-year free-cash-flow formula, which means both boards are making a statement about cash they have not earned yet.
KEY TAKEAWAYS
1. SK hynix will buy back and cancel 24.07 million shares, about 3.3% of shares outstanding, valued at roughly 40 trillion won. Samsung guided 2026 returns to 90-110 trillion won, about 5x its previous annual record.
2. Both rest on the same commitment: half of cumulative free cash flow over a three-year window. Samsung's window is 2024-2026, SK hynix's is 2025-2027. Neither number is final.
3. The collateral behind that confidence is long-term supply agreements. Every major supplier now says more than half of its shipment volume is contracted. What none of them disclose is the price.
What the two filings actually say
English-language coverage tends to collapse both events into one headline number. The Korean disclosures are more specific, and the specifics matter.
| Date | Company | What was approved |
|---|---|---|
| Aug 19 | SK hynix | 24.07 million common shares acquired on-market between August 20 and November 19, then cancelled in full. The payout floor was also raised, from "within 50%" of cumulative free cash flow to "50% or more." |
| Aug 21 | Samsung | 90-110 trillion won of 2026 returns, of which about 30 trillion won is a third-quarter cash dividend. A separate 15 trillion won buyback for employee compensation was approved the same day. |
Three qualifications belong on these numbers before anyone models them.
110 trillion won is not a commitment. Samsung's arithmetic is 50% of 2024-2026 cumulative free cash flow, minus the 29.3 trillion won already paid out in 2024 and 2025 (20.9 trillion in dividends, 8.4 trillion in buybacks). Second-half earnings and capex both move that residual. The company describes it as a current estimate.
40 trillion won is not a commitment either. It is a planned acquisition amount priced off a single closing print, so the realized figure moves with the share price over the three-month window.
The two figures are not comparable. Samsung's is a full-year return envelope. SK hynix's is a single buyback, with additional returns to be disclosed at third-quarter results. Adding them together, or ranking the two companies by them, produces a number that means nothing.
A payout policy is a forward statement
Committing half of free cash flow across a three-year window is not a distribution of past earnings. It is a declaration that future cash flow is predictable enough to pre-commit against.
Memory is the industry where that has historically been least true. Samsung's semiconductor division posted an operating loss of 14.88 trillion won in 2023. Three years later, both Korean suppliers are underwriting multi-year payout floors. The form of the promise is the signal, more than its size. SK hynix's net cash stood at roughly 69 trillion won at the end of the second quarter.
The collateral is long-term agreements
Start with what is public. Kioxia has guided to filling roughly half of its 2028 shipment volume with long-term agreements. SK hynix has disclosed contracts with about ten customers on roughly five-year terms, backed by deposits. Samsung has said it has completed agreements with its five largest customers.
The common thread in supplier commentary is that terms have tilted toward the seller: sizable prepayments, and binding obligations that now run in both directions rather than one.
The consequence is structural rather than cyclical. Once price-setting moves from a quarterly negotiation to a multi-year contract, near-term ASP upside gets capped and a floor appears underneath. That compresses the amplitude of the cycle in both directions.
This is also where the argument is weakest. Suppliers disclose volume coverage. None disclose contract pricing. "More than half the book is locked" is verifiable. "Locked at what price" is not, and it is the variable that determines whether these agreements read as a floor or as forgone upside.
The price argument has already moved on
Conventional DRAM contract prices rose 93-98% quarter over quarter in Q1 2026 per TrendForce, roughly 60% in Q2, and are guided to 13-18% in Q3. NAND contract prices are guided to 10-15% for the same quarter. Prices are still climbing; the second derivative turned some time ago.
HBM traces a different curve. Fubon Research estimates the HBM4 content cost in Nvidia GPUs at 31-32 dollars per gigabyte, against 17-18 dollars for HBM3E. Cantor, writing after FMS 2026, put SK hynix's HBM4 pricing at 32 dollars per gigabyte to Nvidia, 36 to Broadcom and 40 to AMD. All of these are broker estimates. No supplier or customer has confirmed them.
One point is being read backwards. The shift toward 8-high HBM stacks in place of 12-high is described in some coverage as a spec downgrade. Supplier commentary says the opposite: it reflects 12-high availability, not a relaxed performance requirement. Cantor's read was that Nvidia is prioritizing 8-high to protect unit shipment volume.
The new capacity is a 2029 story
SK hynix's August 7 disclosure puts the Cheongju M17 NAND fab at a February 2027 construction start and a December 2028 first cleanroom. Yongin Y2, for DRAM and HBM, breaks ground in July 2027 with a first cleanroom in June 2029. Samsung targets 2028 for Pyeongtaek P5 Fab 1 and 2029 for Fab 2.
A cleanroom opening is not bit output. Tool move-in, setup and yield stabilization typically add about a year on top. So bit growth over the next twelve to eighteen months has to come from process migration, not from new footprint.
A second effect sits on top. As HBM takes a larger share of DRAM capacity, usable die per wafer falls: HBM dies are larger, and stacking consumes several of them per finished part. Adding wafer capacity does not convert into conventional DRAM supply anywhere near one-for-one.
What I actually watch
| When | What | Why it matters |
|---|---|---|
| Late Oct 2026 | Samsung board, Q3 dividend detail | First look at the dividend-versus-buyback split inside the 30 trillion won |
| Q3 2026 results | SK hynix incremental return | Sizes the remaining capacity after the 40 trillion won program |
| Q4 2026 - Q1 2027 | LTA negotiations closing out | Blended ASP effect and whether price-adjustment clauses bite |
| Jan 2027 | Samsung board, residual allocation | Turns the 90-110 trillion won estimate into a fixed number |
Value chain read-through
| Segment | Implication |
|---|---|
| DRAM suppliers | Contracted volume plus HBM mix supports revenue visibility. Capped near-term ASP upside is the trade. |
| NAND suppliers | Enterprise SSD demand is doing the work. Consumer recovery is the swing variable for 2027. |
| Equipment | Order timing is set by 2027-2028 construction starts, not by current spot prices. |
| AI server buyers | Memory is a rising share of bill of materials. The pass-through holds only while accelerator margins hold. |
Risks to this view
- The estimate is not the outcome. A heavier second-half capex program shrinks free cash flow and shrinks the payout pool with it.
- LTAs are a black box. If market pricing runs well above contracted pricing, locked volume becomes opportunity cost rather than protection.
- Pass-through has a ceiling. Rising memory content cost lands on AI server bills of materials. The arrangement works while customer margins absorb it.
- NAND's consumer gap. If smartphone and PC demand stays soft, enterprise SSD strength may not fully offset it. Supplier and buy-side views on 2027 NAND balance are unusually far apart.
- China. Chinese capacity lands on consumer and domestic demand before it reaches server-grade product, but it can still pull down the consumer price floor.
The bottom line
Last week's two numbers are not a distribution of what has been earned. They are a signature on what these companies expect to earn, and the collateral behind that signature is a stack of long-term agreements whose pricing terms nobody outside the negotiation has seen.
For the next several quarters, the more informative variable is probably not the shape of the price curve. It is the thickness of the contract book. The next hard data point is Samsung's late-October board meeting.
Sources: SK hynix board resolution and disclosure (Aug 19, 2026); Samsung Electronics 2026 shareholder return resolution and disclosure (Aug 21, 2026); SK hynix Cheongju M17 and Yongin Y2 investment disclosure (Aug 7, 2026); TrendForce memory contract price surveys (Q1-Q3 2026); Fubon Research and Cantor estimates as reported; company comments at investor meetings in Seoul, Aug 20-21, 2026. Everything here is from public sources.
Disclaimer: This post is for informational and educational purposes only. It does not constitute investment advice or a recommendation to buy or sell any security. All investment decisions are your own responsibility.
Comments
Post a Comment