Samsung and SK Hynix Delay 3D IC Commercialization; CXMT and Winbond May Enter the Market First
Taylor Wilson
AI chip companies are flooding Samsung's foundry division with 3D IC inquiries, but a fundamental business-model clash is keeping Samsung and SK Hynix on the sidelines — opening a window for custom-memory makers CXMT and Winbond to move first.
3D IC demand is surging — who's asking and why?
Per ZDNet Korea, global IC design houses and AI chip firms have stepped up their evaluation of 3D IC adoption, with Samsung's foundry unit seeing a notable jump in related inquiries.
A Samsung foundry insider said: "Every client visiting lately is asking about 3D IC."
This means → the AI chip differentiation race has moved beyond packaging and interconnects into the memory architecture itself — whoever stacks compute and memory closest wins on speed.
Why aren't Samsung and SK Hynix rushing in?
Traditional DRAM runs on standardized, high-volume production — one design sold to everyone, with margins improving as scale grows.
3D DRAM is the opposite: highly customized, designed to each client's spec, effectively re-engineered for every customer.
In plain terms = Samsung and SK Hynix make money with a "one design fits all" model; 3D DRAM demands "one design per client," which fundamentally clashes with how they earn profits.
Industry insiders expect both companies to continue 3D IC R&D, but neither is likely to elevate it into a formal business unit in the near term.
Why could CXMT and Winbond get there first?
CXMT's (长鑫存储) strategy was never to challenge Samsung head-on in commodity DRAM — it is built around custom memory from the start.
U.S. semiconductor export restrictions have blocked CXMT's path into HBM — high-bandwidth memory, a technology that stacks multiple DRAM layers for AI chips. CXMT is reportedly investing heavily in 3D IC capabilities as its next-generation AI-memory play.
This reflects a deeper dynamic: the blocked high-end path has forced open a first-mover window on an alternative technology route.
What to watch next?
China has already produced several 3D DRAM sample chips; CXMT and Winbond are positioned to be the first to push samples toward commercialization.
Two key markers: ① whether the 3D IC track can form a scaled market on schedule; ② when Samsung and SK Hynix adjust strategy to follow.
This means → if custom-memory makers prove 3D IC can turn a profit, the two giants' "no rush" stance will turn into "no choice but to rush."
Content is for reference only, not financial advice.