Samsung Electronics told analysts on 30 July 2026 that it plans to place 60 to 70 per cent of its memory capacity under multi-year supply contracts. That claim is the centre of the argument that this memory cycle will end more gently than the last ones, so we went looking for the documents behind it. Only one of the three large makers has published contract numbers you can audit, and it is not a Korean one.
Last verified: 10 August 2026. Built from Micron’s fiscal third-quarter 2026 earnings materials and its FY2024 Form 10-K, SK hynix’s 2Q26 results release and its 2024 Corporate Value-up disclosure, and Samsung’s Q2 2026 earnings call, then from TrendForce’s public press releases. The companies have signed the contract terms, and those should hold. The coverage percentages describe contracts still under negotiation, so they will move.
Editor’s opinion
The brake is real. Korea has not shown you the paperwork
Our view is that long supply contracts and long fab lead times do make a repeat of the old memory collapses less likely, and that the evidence for it sits in Micron’s filings rather than in Korean disclosure. If you are holding Samsung or SK hynix for this reason, you are holding it on a read-across.
Micron’s agreements are take-or-pay, carry floor prices, and appear in its accounts as roughly 100 billion dollars of remaining performance obligation under ASC 606, against 22 billion dollars of customer deposits and related commitments. That is an audited construct, not a management aspiration, and the company says the floor price alone supports a gross margin above its best quarter in any previous cycle.
Against that: the industry has made this promise before and broken it. Micron’s own FY2024 Form 10-K describes the previous generation of long-term agreements as contracts where “pricing, quantity, and other terms will be periodically negotiated to reflect market conditions” — a contract that renegotiates in a downturn is not a brake. Samsung’s 60-to-70 per cent covers contracts partly unsigned. Samsung said it in a Q&A answer, it does not appear in the company’s own earnings deck, and SK hynix has published no coverage percentage at all. The view survives because take-or-pay with cash already deposited is a different instrument from a price-renegotiable order book, and because a customer who has paid a deposit has spent something to keep the contract.
Andrew Lee. This section argues a view rather than establishing a fact. It is not a forecast and not a recommendation, and the material it draws on is listed in the sources below.
What is public about the supply gap, and what is not
Korean coverage of this cycle circulates precise-looking sufficiency ratios: minus 1.4 per cent for DRAM in 2026, widening to minus 3.8 per cent in 2027. We could not find either number in anything TrendForce has published. What TrendForce says publicly is that the 2026 DRAM sufficiency ratio is “approximately -1% to -2%,” a range and not a point, and that the gap will “widen further in 2027” with no figure attached. The granular model sits inside an 18,000-dollar subscription datasheet, which is why a wafer-start table broken out by maker reaches you through an intermediary rather than from a published forecast.
Two things TrendForce does publish are more useful than the decimals. NAND runs a 4 to 5 per cent supply deficit in 2026 and turns positive in the second half of 2027 — a quantified annual figure, openly stated. And HBM will take an estimated 18, 22 and 30 per cent of the top three makers’ total DRAM wafer input at the end of 2025, 2026 and 2027. That last series is the real supply story. The makers add capacity, and HBM eats it in the same breath.
Lead time is the first brake, and the companies say so themselves
The best thing about the lead-time argument is that you need no analyst to make it. Micron said in June 2026 that it cannot see the end. Even as it expects industry supply to improve gradually in 2028, it “currently do[es] not have line of sight as to when memory supply will be able to catch up with increasing demand.” SK hynix pointed at the same constraint in July, citing the “lead times required for constructing new production facilities.” TrendForce agrees from the outside, putting substantial output no earlier than 2028.
Only one company has stated an interval end to end. We found it while counting what Samsung and SK hynix have actually committed to 2028. SK hynix signed for its Indiana site in April 2024 and targets mass production in the second half of 2028. Four and a half years, contract to output. Think of it as ordering a building rather than buying a machine — you cannot rush the concrete. That single number does more work than any capacity table, because it is the company’s own clock on its own project.
The second brake is a contract, and one company shows the paperwork
Micron is the only maker that has put the new contracts into auditable form, and the contrast with its own past filings is the strongest evidence that something changed. Its FY2024 Form 10-K says customers were “generally reluctant to enter into long-term, fixed-price purchase contracts” and that terms were periodically renegotiated. In March 2026 the chief executive said the new Strategic Customer Agreements “are different from prior LTAs.”
Different in specific, checkable ways. Micron structures them as take-or-pay, with binding volume commitments. Its chief business officer said flatly that they “cannot be canceled” and contain no provision letting a customer walk away. Sixteen signed agreements cover roughly 20 per cent of Micron’s DRAM volume and a third of its NAND volume, typically over calendar 2026 to 2030. Fourteen of them carry about 100 billion dollars of contracted revenue at minimum prices.
Why the price floor is the part that matters
The largest agreements set a ceiling at the calendar second-quarter market price and a floor that holds for the term. Micron expects roughly 40 per cent of revenue to sit under fixed or ceiling pricing once it signs every planned agreement, and about half or more of company revenue under the agreements in total. On the floor it is unusually direct: it “enables a very robust gross margin for Micron, well above our peak quarterly margins in any past cycle.” Note what the ceiling costs. Micron capped its own upside on those volumes to win the floor. That is the trade you would expect, and the reason to believe the floor is real.
Korea has not shown the same paperwork
Samsung’s 60-to-70 per cent is a genuine company statement, not a brokerage estimate, and it deserves credit for that. But read the tense. Samsung’s memory sales head said it plans to allocate that share to long-term agreements, and that volumes under multi-year contracts “will easily account for 60%-70% of our planned capacity” once it finalises the pending ones. Samsung has closed agreements with the top five global data centre customers and is in final talks with five more. So it is a plan, delivered in Q&A, and absent from Samsung’s own earnings deck.
SK hynix discloses less. It has closed agreements with “around 10 customers” and says it has built deposits into them, calling the structures “designed to address price volatility.” It gives no percentage of capacity, bits or revenue at all. If you have seen 60 to 70 per cent attached to SK hynix, that is a read-across from Samsung, not an SK hynix figure — the same precision limit we ran into when measuring how far behind CXMT actually is.
The capital expenditure figure everyone is quoting is wrong
One correction worth making, because the wrong version is everywhere. SK hynix’s Value-up plan sets capital expenditure at a “mid-30%” share of revenue on a three-year moving average, not within 30 per cent. Management restated the mid-30 per cent target on the January 2026 call. The 30 per cent figure in circulation appears to come from the actual trailing figure of 27.8 per cent at the third quarter of 2025 — which is an outcome of a revenue surge, not a ceiling the company promised. The distinction matters here: a mid-30s target with a 27.8 per cent outturn means SK hynix has room to spend more, not less.
Contract coverage, by what the company itself has published
Three makers, three very different standards of disclosure, as at 10 August 2026
| Maker | Coverage figure | What kind of statement it is |
|---|---|---|
| Micron | ~20% of DRAM volume and a third of NAND volume signed; ~half or more of revenue when complete; ~$100bn RPO | Prepared earnings remarks plus an ASC 606 accounting disclosure |
| Samsung Electronics | 60–70% of planned capacity | Spoken answer in earnings Q&A; a plan contingent on unsigned contracts; not in the IR deck |
| SK hynix | None published — only “around 10 customers” | Press release and earnings call; no percentage of any denominator |
The three rows are not comparable, and that is the finding. Only the first has a defined denominator and reaches the accounts. Treating all three as equivalent coverage is how a read-across becomes a fact.
Sources
We take facts from these and write our own sentences. Where a figure originates in a company filing we say so, and where it originates in a research house estimate we name the house. We link to a document only where we hold the exact address for it; a link to a publisher’s front page would not get you to the source, so where we do not have the direct address we give you enough to find it.
Filings and official documents
- Micron Technology, Fiscal Q3 2026 earnings call prepared remarks, 24 June 2026. Source for take-or-pay structure, price floors and ceilings, the 16 signed agreements, ~20 per cent of DRAM and a third of NAND volume, ~$100bn of contracted revenue and remaining performance obligation, $22bn of deposits and related commitments, ~40 per cent of revenue under fixed or ceiling pricing, the margin statement, and the greenfield lead-time passage.
- Micron Technology, Fiscal Q2 2026 earnings call prepared remarks, 18 March 2026. Source for the first five-year agreement and the company’s own distinction between the new agreements and prior long-term agreements.
- Micron Technology, Form 10-K for fiscal year 2024, filed October 2024. Source for the description of the previous contract regime, in which pricing and quantity were periodically renegotiated.
- SK hynix, 2Q26 business results, 29 July 2026. Source for agreements with around 10 customers, the capital expenditure discipline language, and the lead-time argument.
- SK hynix, shareholder return programme and Value-up plan announcement, 27 November 2024. Source for capital expenditure discipline at an average mid-30 per cent range of revenue. The three-year moving-average qualifier appears in the Korean disclosure and the 2025 implementation report rather than in this English release; the 27.8 per cent trailing figure at the third quarter of 2025 comes from that implementation report, filed with the Korean regulator on 27 November 2025.
- Samsung Electronics, Q2 2026 earnings conference call, 30 July 2026. Source for the 60 to 70 per cent capacity allocation plan, the top five data centre customers, and the framing of long-term agreements as hedging. Samsung publishes the webcast and its earnings deck at its investor relations site; the coverage figure appears in the spoken question-and-answer session and not in the deck, and we could not locate a Samsung-hosted transcript.
Research houses and reporting
- TrendForce, “Diverging Memory Market Outlook in 2027”, 30 July 2026. Estimate. Source for the 2026 DRAM sufficiency ratio of approximately -1 to -2 per cent, for the statement that the gap widens further in 2027 without a figure, and for output contributions not arriving in volume until 2028.
- TrendForce, NAND Flash supply and demand release, 21 July 2026. Estimate. Source for the 4 to 5 per cent NAND supply deficit in 2026 turning positive in the second half of 2027.
- TrendForce, HBM wafer input release, 2 June 2026. Estimate. Source for HBM taking approximately 18, 22 and 30 per cent of the top three suppliers’ DRAM wafer input at the end of 2025, 2026 and 2027.
- TrendForce, memory market release, 4 August 2026. Estimate. Source for NVIDIA halving next-generation SOCAMM capacity on constrained LPDDR5X supply through 2027. We checked one further claim against this and the July releases and are not carrying it: that mobile DRAM flips to oversupply while server DRAM tightens. TrendForce’s public position runs the other way — it describes low-power DRAM as tight and frames China’s CXMT as filling an LPDDR4X gap rather than creating a glut.
Material we could not open
- TrendForce, DRAM Monthly Datasheet product listing, last updated 22 July 2026. Cited for price and scope only. This is the subscription product, listed at 18,000 dollars, that contains the supply and demand sufficiency model. We did not purchase or read it, and we name it to explain why the granular sufficiency and wafer-start figures in circulation are not publicly attributable to TrendForce.
- Micron Technology, fiscal Q3 2026 post-earnings analyst call, 24 June 2026. Cited for the statement that the agreements cannot be cancelled. Micron hosts the audio at its investor relations site; we read the remarks in a third-party transcript rather than a company-published one, and flag that difference.
- Shinhan Securities, “Semiconductor Equipment — CapEx Speed Up!” sector report, 6 August 2026. Cited, original not read. This report is the origin of much of the Korean-language framing of this subject, including sufficiency ratios and per-maker wafer-start additions. It is distributed through brokerage channels with no public address, we could not open it, and we do not reproduce its tables or estimates.
Corrections
None yet. If you can point to a primary document that contradicts anything above, we will correct it and say what changed.
Disclosure. The author holds no position in Samsung Electronics, SK hynix, Micron Technology or NVIDIA as of the date of publication. This article is general information and analysis about memory semiconductor supply, demand and customer contracts. It is not investment advice, not a recommendation to buy or sell any security, and not a forecast of any price or index level. The section headed Editor’s opinion is comment: it argues a view, and we fence it off from the sourced material for that reason. We have not tailored anything here to your circumstances, and only a qualified adviser in your jurisdiction can do that. We are not registered as an investment adviser in any jurisdiction, and this site has no commercial relationship with any company, exchange or research provider named above. See the Editorial & Sourcing Policy and Disclaimer.