SK Hynix Evaluates Intel Foundry for HBM Base Die Production
nashnova research
SK Hynix is evaluating a shift of part of its HBM base-die foundry work from TSMC to Intel, potentially starting with the seventh-generation HBM4E — this means the most critical outsourced link in the HBM supply chain could see genuine multi-vendor competition for the first time.
Why is SK Hynix looking beyond TSMC?
The core issue is cost: TSMC's 12 nm base die — the bottom chip in an HBM stack that handles connection and control — costs 3 to 4 times more than SK Hynix's own core die produced on its 1b-node process.
As the next-generation HBM4E grows more complex, that cost gap is expected to widen further.
This means → HBM is sold largely on long-term contracts, so rising foundry costs cannot be passed through to customers quickly — the margin squeeze hits directly.
What would bringing in Intel solve?
The most immediate gain is pricing leverage: with a second supplier, TSMC's quote is no longer take-it-or-leave-it.
In plain terms = TSMC is currently the sole source for base dies; SK Hynix has almost no room to negotiate. Once Intel enters the picture, the buyer gains a bargaining chip.
A second source also strengthens supply resilience — reducing the risk of a single-point disruption.
Could the partnership go beyond chip fabrication?
Reports suggest the potential collaboration extends to advanced packaging — the process of assembling multiple dies into one unit.
SK Hynix's Americas packaging VP Lee Jae-sik revealed at the "Hot Chips" conference that the company is benchmarking TSMC's CoWoS-S and CoWoS-L against Intel's EMIB packaging technology.
This reflects that Intel's packaging capabilities have formally entered SK Hynix's evaluation pipeline — no longer a theoretical alternative.
Why does "custom HBM" make multi-sourcing more urgent?
Starting with HBM4, AI chip makers — Nvidia, AMD, Google and others — increasingly demand customized HBM specs, with each customer requiring different configurations.
Relying on TSMC alone locks SK Hynix into one set of process options and capacity allocations, making it hard to serve multiple customers' divergent needs at once.
This means → a multi-vendor foundry model is not just about saving money — it is a prerequisite for SK Hynix to serve a broader AI client base.
Is this confirmed?
SK Hynix's official response: the company said details about its technology roadmap are "difficult to confirm" and that "some of the reported content does not match the facts" — neither confirming nor denying.
In plain terms = in the semiconductor industry, this kind of wording typically signals that discussions are underway but have not reached a stage for public acknowledgment.
Whether Intel ultimately secures a place in this supply chain still depends on substantive progress in technical evaluation and commercial negotiation.
市场有风险，内容仅供研究参考，不构成投资建议。