SK Hynix’s Reported Intel Plan Could Reshape HBM4E’s Supply Chain

The reported plan is a potential dual-sourcing arrangement: SK Hynix could have Intel Foundry manufacture part of the logic base dies used in its next-generation HBM4E stacks, while continuing to work with TSMC. The proposal has not been publicly confirmed by SK Hynix, Intel or TSMC, so it should be treated as an evaluation rather than a finalized production agreement. 234

The strategic logic is straightforward. HBM4E is arriving as AI-chip makers try to secure more memory, but its logic base die is becoming a more expensive and strategically important component. A qualified second supplier could help SK Hynix manage cost, capacity and negotiating leverage without abandoning its existing TSMC relationship.

Why SK Hynix is considering a second base-die supplier

HBM is built by stacking DRAM layers above a logic layer known as the base die. Rather than storing data like the DRAM layers, the base die helps manage the stack’s interface, routes signals and connects the memory to the surrounding accelerator package. SK hynix itself has highlighted the growing importance of the logic die as HBM advances toward higher connectivity and lower power consumption. 7

For HBM4, SK Hynix has relied on TSMC’s 12nm-class logic process for the base die. Industry reports cited in the coverage of the Intel proposal say that this base-die manufacturing can cost roughly three to four times as much as producing a core DRAM die. That cost gap creates pressure to find an additional manufacturing route, particularly for less customized products where the most advanced process may not always be essential. 2415

Dual-sourcing would also reduce dependence on a single foundry. If Intel could match the required yield, power, reliability and packaging compatibility, SK Hynix could divide production between Intel and TSMC. That would provide more flexibility when negotiating capacity and pricing, although the exact allocation and production timeline remain unknown.

HBM4E demand raises the stakes

SK Hynix has already shipped 12-layer HBM4E samples to major customers. The company says the samples reach speeds of up to 16 Gbps per pin and deliver more than 20% better power efficiency than the preceding generation. 15

Those specifications matter because HBM sits close to the center of modern AI-system performance. Faster memory can feed accelerators more quickly, while better power efficiency helps limit the energy and cooling burden of large AI deployments. But higher performance also increases the importance of dependable, scalable production across the entire stack—not just the DRAM layers.

The supply environment is tightening at the same time. NVIDIA’s reported supply and capacity commitments rose from $119 billion to $279 billion, with memory procurement described as a major driver. The figure covers broader supply-chain and infrastructure commitments rather than a single HBM purchase from SK Hynix, but it illustrates why memory makers are working to secure every critical step in production.

Base-die manufacturing and packaging are different decisions

The Intel discussions are easy to overstate because HBM production involves multiple linked technologies. Manufacturing the logic base die on a wafer is one step; connecting HBM stacks with an accelerator in the final package is another.

TSMC’s CoWoS is a 2.5D packaging family used to integrate accelerators and HBM through a high-density interconnect structure. Intel’s EMIB, or Embedded Multi-die Interconnect Bridge, instead uses small silicon bridges embedded in the package substrate to create dense die-to-die connections. Reports say SK Hynix has been testing HBM integration with Intel’s EMIB-based packaging and assessing related materials and components.

That creates the possibility of Intel participating in two parts of the supply chain:

• Foundry: fabricating some HBM4E logic base dies.
• Advanced packaging: providing or supporting EMIB-based integration between memory and logic.

These possibilities should remain separate analytically. Packaging research does not prove that Intel will manufacture the base dies, and a base-die qualification would not by itself establish Intel as the supplier of the finished AI package.

What the arrangement could mean for Intel and TSMC

For Intel, even a limited SK Hynix order would be an important external customer win for Intel Foundry. It would offer a practical demonstration that Intel can use its manufacturing and advanced-packaging capabilities for a demanding, high-growth application. The commercial impact would depend on whether Intel progresses from evaluation to qualified, repeatable volume production.

For TSMC, the immediate financial effect of losing only part of a potential base-die program would likely be limited. The larger issue would be strategic. A credible alternative could reduce TSMC’s exclusivity in HBM-related logic manufacturing and give SK Hynix more leverage over price and capacity. That remains a possibility, not an established outcome, because no volume allocation has been announced. 2

Other memory companies may also be watching. Reports indicate that Samsung and Micron have assessed the compatibility of Intel’s packaging approach, but that does not amount to an announced adoption plan. Any broader shift would depend on Intel proving that EMIB can meet customers’ requirements for performance, power, cost and production reliability.

How this fits Intel and SK Hynix’s broader relationship

The potential HBM collaboration would build on an existing commercial relationship, although the two situations are not the same. In 2020, Intel agreed to sell its NAND memory and storage business—including its SSD business, NAND assets and Dalian facility—to SK Hynix for $9 billion. The first closing, valued at $7 billion, was completed in December 2021.

There have also been reports that SK Hynix was considered as a possible partner for Intel’s Ohio fab. SK Hynix denied acquisition plans, and that report is separate from the HBM4E base-die proposal. It nevertheless points to wider interest in cooperation between a major memory producer and Intel’s foundry and packaging businesses.

The key question is qualification, not the headline

The reported Intel plan is best understood as a supply-chain hedge. SK Hynix has reasons to seek another source: base-die costs are reportedly high, AI customers are locking in capacity, and HBM4E places greater demands on logic, power and packaging.

But a second source is useful only if it can deliver production-scale results. Intel would need to demonstrate acceptable yield, power characteristics, reliability, intellectual-property integration and compatibility with SK Hynix’s HBM manufacturing flow. Until the companies confirm a production agreement or volume allocation, the most accurate conclusion is that SK Hynix is exploring Intel as a potential complement to TSMC—not replacing TSMC outright.