Long-term supply agreements have become the standard way AI buyers secure memory, Korean daily Hankyung reported on 2 November. Five years is now the base contract term for memory, and deposits of 20% to 25% or take-or-pay terms, under which the buyer pays even if it does not take the volume, have become default options, the paper said, citing the industry.
Adding up the deals each company has reported, Samsung Electronics, SK hynix, Micron and Sandisk hold 58 long-term memory agreements between them, by our count. The count is of agreements, not unique buyers: one buyer can hold deals with several makers.
The key insight: Samsung, SK hynix, Micron and Sandisk hold 58 long-term memory agreements between them by our count, and Hankyung reported on 2 November that five-year terms with 20% to 25% deposits or take-or-pay clauses are now the default, citing the industry.
Samsung has signed multi-year memory supply contracts with 10 large customers and is negotiating with 10 more, some of them near completion, according to Datanews’ reading of Samsung’s third-quarter results and earnings call. Datanews said that in the second quarter the figures were five signed and five in talks, so both have doubled.
On its 29 October call, Samsung said its goal of raising long-term agreements to 60% to 70% of production capacity by 2030 is more likely to be reached, Hankyung reported. About 80% of Samsung’s memory output for next year already has a confirmed buyer, Hankyung said, citing the industry.
SK hynix signed long-term agreements with 14 major customers, including big tech companies, in the third quarter, covering more than 50% of its total production capacity for 2027 and 2028, according to Hankyung. A company official told the paper that DRAM and HBM volumes for the three years to 2029 are effectively sold out.
Micron’s strategic customer agreements rose from 16 in the second quarter to 26 in the third, Hankyung reported, and Datanews said Micron expects revenue from those agreements to exceed 35% of its total by 2030. Sandisk has long-term agreements with eight customers, including big tech companies, according to Hankyung.
What Was Posted on X
According to Korean media, Samsung and SK Hynix have locked up most of their future supply with long term contracts.
Samsung has signed deals with 10 large customers and is negotiating with 10 more, and its 2027 HBM supply is already sold out.
SK Hynix has deals with 14 customers that cover more than half of its 2027 and 2028 capacity, backed by upfront deposits.
The interesting detail is how they are building.
Both companies are putting up factory shells early but adding cleanrooms in phases, because they are watching how big tech in the United States funds its AI spending.
Memory used to be priced quarter to quarter, and multiyear contracts with deposits make this cycle look very different from the old boom and bust.
Investor account MelvinInvests on 2 November, 07:45 UTC, citing Korean media: Samsung has deals with 10 large customers and is negotiating with 10 more, and SK hynix has deals with 14 customers covering more than half of its 2027 and 2028 capacity, backed by upfront deposits. The post adds that both companies are putting up factory shells early and adding cleanrooms in phases; we did not find that detail in the Hankyung and Datanews reports this piece is built on.
Long-term memory supply agreements by maker, one dot per agreement: Micron 26, SK hynix 14, Samsung 10 signed plus 10 in talks, Sandisk 8, as reported by Hankyung and Datanews on 2 November 2026. The 58 total is our count; counts are agreements, not unique buyers.
The Contract Terms
Samsung’s contracts run five years as a base and can be extended through yearly negotiation, Datanews reported. They include sizeable deposits that customers must keep in place for the contract term, and a price floor on commodity memory to limit Samsung’s investment risk if market prices swing, the report said.
Suppliers sometimes use the deposits to fund large fab construction, Hankyung reported. Memory and parts used to be sold on quarterly or spot-price contracts, the paper said, which left chipmakers struggling to forecast prices and demand.
A short explainer on backlog, revenue that is signed but not yet earned. It teaches the general idea only and says nothing about any company in this story.
Beyond Memory: Capacitors, Power and Cooling
The same contracts are spreading across AI data-centre parts, according to Hankyung. Samsung Electro-Mechanics recently signed an 879.3 billion won MLCC long-term agreement with a global big tech company, its eighth this year, and is signing or negotiating such agreements with about 50 customers, the paper said. We covered that order in Samsung Electro-Mechanics’ $658 million order.
LG Innotek is negotiating long-term agreements with big tech for FC-BGA package substrates, LS Cable has five-year busduct agreements with Meta and Google, and HD Hyundai Electric is pursuing agreements for data-centre transformers, Hankyung reported. One industry official told the paper that big tech companies are now accepting unfavourable terms to secure volume.
What One 100MW Data Centre Needs
A 100-megawatt hyperscale AI data centre typically holds about 40,000 Nvidia GPUs, which need about 330,000 HBM stacks to run, Hankyung reported in a companion piece. It also uses about 170,000 FC-BGA substrates, about 330 million MLCCs and about 80 coolant distribution units, the paper said.
By our arithmetic from those figures, that is about eight HBM stacks and about 8,250 MLCCs per GPU. If any one of those parts is short, the data centre cannot run, Hankyung said.
The Risk the Contracts Leave
Long-term agreements do not remove all risk, Hankyung reported: if the chip market turns sharply, customers could cancel and absorb the penalty, and suppliers that built capacity on the strength of the contracts would be hit hard. Choi Woo-young, a professor of electrical and computer engineering at Seoul National University, told the paper that suppliers would have little choice but to ease terms if a memory downturn led customers to ask for renegotiation.
As we read the reports, the change is in who carries the risk: memory used to be priced quarter by quarter, so the chipmaker absorbed the cycle; deposits, take-or-pay clauses and Samsung’s price floor on commodity memory, as described by Hankyung and Datanews, move part of it to the buyer.
As we read Hankyung’s account, the same terms reaching capacitors, substrates, busducts and transformers means the AI build-out is being booked years ahead across the parts list, not only for HBM. Its own risk passage is the counterweight: a buyer can still walk away and pay the penalty.
Choi Woo-young, Seoul National University professor, to Hankyung (Korean original; in English: if a memory downturn hits and customers ask to renegotiate, suppliers will have no choice but to ease the existing terms)
FOR MEMORY BUYERS Hankyung reported that five-year terms with 20% to 25% deposits or take-or-pay clauses have become default options, and that about 80% of Samsung’s memory output for next year already has a confirmed buyer.
FOR MEMORY MAKERS SK hynix’s 14 agreements cover more than 50% of its 2027 and 2028 capacity, and Samsung is targeting 60% to 70% of capacity under long-term agreements by 2030, according to Hankyung.
FOR OTHER AI PARTS SUPPLIERS Hankyung reported long-term agreements at Samsung Electro-Mechanics for MLCCs, LS Cable for busducts with Meta and Google, and talks at LG Innotek for FC-BGA substrates and HD Hyundai Electric for transformers.
The framework opens: “While attention remains fixed on GPU shortages and frontier model capabilities, the binding constraint on AI scaling has quietly shifted to High Bandwidth Memory (HBM).”
As we read it, five-year contracts with deposits are what a binding constraint looks like in a contract: buyers pay ahead to hold a place in line. The framework is the owner’s argument and does not discuss these agreements.
The 20% to 25% deposit range and the five-year standard term are Hankyung’s description citing the industry; the companies have not disclosed deposit amounts for individual contracts. The 100MW parts list is also an industry estimate reported by Hankyung, not a company disclosure.
Hankyung dates Samsung’s 10 signed agreements to the second quarter, while Datanews says Samsung had five signed in the second quarter and 10 by the third. The reports do not name the buyers or say how many hold contracts with more than one maker.
This Week In The AI Supercycle
Anthropic’s Reality Check, The Capital Squeeze, & The New World Order
This week’s issue: for two years the AI build-out was priced as a growthstory. This week it started being priced as a credit story.
Samsung, SK hynix, Micron and Sandisk hold 58 long-term memory agreements between them by our count, and Hankyung reports five-year terms with 20% to 25% deposits or take-or-pay clauses as the new default. The same contracts are reaching capacitors, substrates, power gear and cooling, the paper said, while a Seoul National University professor warned that terms could be eased in a downturn.
A note on sourcing. This piece is based on two Korean-language Hankyung reports and one Datanews report of 2 November 2026, translated by us; the deposit range, five-year term and 100MW parts list are those outlets’ reporting citing the industry. The 58 total and per-GPU figures are our arithmetic.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
