SK hynix told analysts on its third-quarter earnings call on 28 October that some customers are moving toward 8-high instead of 12-high stacks of high-bandwidth memory (HBM), and that the move reflects a shortage of supply rather than weaker demand, according to Korean outlets iNews24, ZDNet Korea and Digital Daily (our translation from the Korean).
The company also said that the volumes customers are requesting exceed more than twice what it can supply. By our arithmetic, an 8-high stack uses a third fewer DRAM dies than a 12-high one, so the same dies yield 1.5 times as many stacks, which is the logic SK hynix described.
Short explainer · SK Hynix: HBM Requests Top 2x Supply as Buyers Cut to 8 Layers
A vertical news Short narrated from this piece’s headline and opening line.
Business Pill · OPERATING LEVERAGE
A short explainer on operating leverage, why profits swing more than revenue when costs are largely fixed, as in memory chipmaking. It teaches the general idea only and says nothing about this earnings call.
The key insight: SK hynix calls the move to 8-high HBM a stopgap to get more cubes from limited supply; by our arithmetic the same dies fill 1.5 times as many stacks, and requested volumes still exceed twice its supply.
iNews24 reported that SK hynix called the recent downgrade by some customers a representative example of demand overwhelming supply. ZDNet Korea quoted the company as saying that, with HBM in short supply, customers have to build more AI accelerators and systems with limited volumes, and that lowering stacks from 12 to 8 layers is a stopgap to increase the number of cubes they can secure for the same capacity.
ZDNet Korea reported the company saying that even allowing for that possible downgrade, requested volumes remain more than double what it can supply. iNews24 carried the same more-than-double figure and quoted the company as saying the medium-to-long-term HBM demand outlook is unchanged despite the downgrade. Digitimes headlined its report “SK Hynix says customers want more than twice the HBM it can supply”; only the headline was readable to us.
Digital Daily described the customers as considering fewer layers, a response to the supply shortage. It quoted Park Jun-deok, head of DRAM marketing, as saying that raising the capacity per DRAM die can largely offset the capacity lost by going from 12 to 8 layers, and that the change was already reflected in production and supply plans, so existing delivery schedules would not slip.
Illustration of the 12-high vs 8-high trade-off SK hynix described on its 28 October 2026 earnings call: the same 24 DRAM dies fill two 12-high stacks or three 8-high stacks, 1.5 times as many (our arithmetic, one die per layer). Source: SK hynix call as reported by ZDNet Korea, iNews24 and Digital Daily (our translation).
What Was Posted on X
According to Korean media, SK Hynix says some customers are cutting the number of HBM stacks per chip because there simply is not enough supply.
Customers are asking for more than DOUBLE what SK Hynix can deliver.
The company also said it will stay on its current DRAM design with EUV for now, and move to next generation 4Fยฒ or 3D DRAM only when the economics make sense.
When buyers start redesigning products around a shortage, you know pricing power is staying with the memory makers.
This post summarises Korean media reports: SK hynix says some customers are cutting the number of HBM stacks per chip because there is not enough supply, customers are asking for more than double what it can deliver, and it will stay on its current DRAM design with EUV for now.
Sold Out for Three Years, Half of 2027-28 Under Contract
Digital Daily reported that SK hynix has signed long-term supply agreements with 14 major customers, including a global cloud provider, and that the contracted volume covers more than 50% of its total production capacity for 2027 and 2028. It quoted Park saying that, combining those agreements with customers’ stated demand, HBM volume for the next three years is effectively sold out. ZDNet Korea reported the company describing HBM and DRAM volumes as sold out through 2029.
The company reported quarterly revenue of KRW 104.6 trillion and operating profit of KRW 83.3 trillion, both records, an operating margin of 80%, according to Digital Daily. We covered the results and the stock’s reaction on 28 October.
The Roadmap: Stretch Today’s DRAM Before the Next Design
DigitalToday reported that SK hynix plans to use its current 6Fยฒ DRAM cell structure as far as possible before moving to 4Fยฒ or 3D DRAM, and that it says EUV-based process improvements give it room to extend 6Fยฒ further. It quoted Park saying that, for now, making the most of a platform with proven mass-production stability and cost competitiveness is the most efficient strategy, and that each new technology will be applied when its maturity and economics are in place.
The Business Lever: Buyers Redesign Around Scarce Memory
As we read the call reports, the customer response SK hynix described is a design change, fewer layers per stack, made to get more stacks out of a fixed supply of DRAM dies.
Digital Daily reports contracts with 14 major customers covering more than half of SK hynix’s 2027 and 2028 capacity, and quotes Park Jun-deok saying that, combined with customers’ stated demand, the next three years of HBM are effectively sold out.
As we read DigitalToday, SK hynix is choosing to stretch its current 6F² DRAM with EUV rather than rush to 4F² or 3D designs, timing each change to its economics.
Digitimes headline, 28 October 2026
“SK Hynix says customers want more than twice the HBM it can supply”
Three Implications
FOR AI CHIP BUYERS SK hynix says some customers are moving toward 8-high stacks to get more cubes from limited supply, according to ZDNet Korea and iNews24.
FOR SK HYNIX Digital Daily reports contracts with 14 major customers covering more than 50% of 2027 and 2028 capacity, and Park Jun-deok says the next three years of HBM are effectively sold out.
FOR THE DRAM ROADMAP DigitalToday reports SK hynix will extend 6F² with EUV and move to 4F² or 3D DRAM when the economics are in place.
The framework says “Memory is becoming the new compute. HBM growth is outpacing accelerator demand, making memory the rate-limiting factor in the AI scaling equation.”
As we read it, buyers trimming HBM layers to stretch supply is the kind of behaviour that thesis describes. The framework predates this call and does not discuss it.
What happened next (31 Oct): an HBM chief for the Americas
SK hynix said on 30 October that it has created a president-level Americas headquarters in the United States and appointed Ahn Hyun, the head of development who led its HBM business, as its first president, effective 1 November, Chosunbiz reported. The aim is to strengthen cooperation with US big tech customers such as Nvidia; HBM has to be developed to fit each customer’s GPU design and packaging. Cha Sun-yong succeeds Ahn as head of development and Kwak Noh-jung stays CEO. No investment figure was given.
What Is Not Established
The reports do not name which customers moved to 8-high stacks or how many, and SK hynix did not give the share of its HBM orders affected. We read the call through Korean press reports, not a transcript; translations are ours. The first-paragraph framing of customers asking for more than double supply is the company’s description of requested volumes, not shipped orders.
This Week In The AI Supercycle
Execution Fault Lines, Concentrated Demand, and Geopolitical Firewalls
This week’s issue: the AI build is not short of demand for its parts. It is short of cheap money, and its end revenue rests on a narrow base of large, often venture-funded buyers.
What happened next (31 Oct): SK hynix’s supply contracts usually run five years, and some buyers want longer
On its 28 October call SK hynix said its long-term agreements now cover 14 customers, including large global cloud companies, and at least half of its 2027–2028 capacity, iNews24 reported. The contracts usually run five years; some customers are asking for longer terms and the company is discussing extensions with them. Prices move with the market inside a range agreed with each customer, and deposits are used so that purchase plans turn into actual orders and shipments. That is up from about 10 customers described on its July call, NewsTomato reported. The company’s view that DRAM and HBM volumes for the three years to 2029 are effectively sold out combines the signed LTA volume with demand still under negotiation, not signed contracts alone.
The Bottom Line
SK hynix said on 28 October that some HBM customers are moving toward 8-high instead of 12-high stacks because there is not enough supply, and that requested volumes still exceed twice what it can deliver. An 8-high stack needs a third fewer dies, so the same output stretches to 1.5 times as many stacks, by our arithmetic.
A note on sourcing. SK hynix’s statements come from its 28 October earnings call as reported in Korean by ZDNet Korea, iNews24, Digital Daily and DigitalToday; the English is our translation. Digitimes’ report was readable only in its headline. We did not contact SK hynix. Figures labelled our arithmetic are our own.
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.
