Korea: HBM Memory and What Comes Next

Korea makes most of the high-bandwidth memory that sits next to AI GPUs. This door follows who gets paid when HBM is allocated, which step is stuck, and the dates that matter — in plain English. Education only, not investment advice.

What Korea does in the AI chain

Korea is the AI memory door. Most of the high-bandwidth memory that sits next to an AI GPU is made by SK hynix and Samsung. Without that stack, a leading-edge die from Taiwan cannot become a finished accelerator.

Buyers are the same hyperscalers and neoclouds that book Taiwan packaging. Korea gets paid when HBM is allocated under long agreements, not when it is sold freely on a spot market. Pricing power shows up in memory makers' margins when supply stays tight.

Micron also makes HBM outside Korea; the Korea door still covers the largest share of AI-class output. For the layer map, read Semiconductors alongside HBM, explained.

As at September 2026, HBM remains allocated. Exports and supplier profits are high because the step is scarce, not because every buyer can get what they want.

Status table

StepWhoHow they get paidStatus (Sep 2026)
HBM / advanced DRAMSK hynix, Samsung (Micron smaller share)Allocated under long-term agreements; high ASP vs commodity DRAMStill stuck / allocated
Logic foundry (smaller)Samsung FoundryWafer pricing; competes with TSMC on some nodesSecondary to memory for this door
Packaging joinMostly Taiwan CoWoSSee the Taiwan doorStill a finish-line gate
Chip exportsKorea exporters (memory-led)Volume at elevated memory pricesRecord (Aug 2026 ~US$46.7bn)

Status is Second Order's read from SK hynix, Samsung, MOTIE and trade-press figures in the latest Korea Pulse.

Companies by role

NameWhat they doWhich step
SK hynixLargest AI-class HBM supplier by many 2026 readsHBM
Samsung Electronics (DS)HBM and DRAM; foundry on the sideHBM / memory
MicronThird HBM supplier (US-listed)HBM

Examples of exposure roles only — not recommendations.

Dates to watch

DateEventWhat to look for
Late Oct 2026SK hynix Q3 2026 resultsHBM mix, margin, allocation comments
Late Oct 2026Samsung Q3 2026 resultsDS memory profit; HBM capacity language
MonthlyKorea chip export printsWhether memory-led exports stay elevated
OpenHBM4 qualification / customer mixWhether next-gen supply stays under LTAs

What would change this view

  • Credible evidence that HBM supply meets allocated demand without rationing
  • Major buyer cuts that free memory without new capacity
  • A sustained drop in memory makers' operating margins while shipments stay high
  • Architecture shifts that cut HBM intensity per unit of useful compute
  • Packaging catching up so memory becomes the louder public bottleneck

Start here

FAQ

Why is Korea important for AI chips?

Korea makes most of the high-bandwidth memory that sits next to AI GPUs. Without that memory, leading-edge dies cannot ship as finished accelerators.

What does HBM allocation mean?

It means supply is booked under long agreements to a short customer list, not sold freely to whoever offers the highest spot price.

Korea supplies HBM; Taiwan does most leading-edge chipmaking and CoWoS packaging that joins the two. The stuck step can sit on either side.

Is this investment advice?

No. This door maps who gets paid and which step is stuck. It is education only.

Updated 19 Sep 2026 · Next update after SK hynix and Samsung Q3 2026 results (late Oct 2026)