SK Hynix is exploring plans to expand base die production—a critical component of next-generation high-bandwidth memory (HBM)—from Taiwan’s TSMC to Intel’s foundry. With the option of assigning volume to Intel starting from HBM4E, the seventh-generation HBM, gaining traction, the South Korean chipmaker’s efforts to shift from a single-foundry supply chain to a multi-vendor arrangement are becoming more concrete.

According to industry sources on the 31st, SK Hynix is pursuing a plan to collaborate with Intel as well as TSMC on base dies for HBM4E at the earliest. The base die is a controller chip positioned at the bottom of the HBM stack that supports and controls the entire structure; it is also referred to as a logic die. It serves as the interface that connects external processors such as GPUs or CPUs to HBM at high speeds.

SK Hynix had produced base dies in-house through HBM3E. However, as the importance of logic functionality and performance in base dies grew with each HBM generation, the company began outsourcing production to TSMC starting with HBM4. The HBM4 currently in mass production uses base dies manufactured on TSMC’s 12nm-class process.

The key issue is cost burden. According to industry sources, TSMC-supplied base dies for HBM4 are three to four times more expensive than the core dies SK Hynix produces on its 10nm-class fifth-generation (1b) process. Core dies are the DRAM chips stacked on top of the base die that directly store data. The base die cost burden is expected to grow further with the transition to HBM4E. Since HBM products carry a high proportion of long-term supply agreements (LTAs), it is difficult to immediately pass foundry cost increases through to product prices.

For this reason, industry observers believe that if Intel actually joins the base die supply chain, SK Hynix would gain flexibility on both cost and supply stability. The strategic decision to diversify a supply chain concentrated on TSMC over the long term is also cited as a key driver. With HBM4 as the inflection point, custom HBM—where base die designs are tailored to each customer’s AI accelerator architecture and performance requirements—is becoming increasingly important. This means SK Hynix needs to break free from an ecosystem tied to specific customers and foundries, and secure room to collaborate with a broader range of AI chip companies.

The possibility of Intel collaboration extending beyond base die production into advanced packaging is also being raised. At Hot Chips, a high-performance semiconductor and computing architecture conference held in Silicon Valley, Lee Jae-sik, vice president of package engineering at SK Hynix America, explained that the company is comparing and analyzing major advanced packaging approaches including TSMC’s CoWoS-S and CoWoS-L, as well as Intel’s EMIB.

Background of Supply Chain Diversification

SK Hynix’s latest move is intertwined with structural changes across the HBM supply chain. From HBM4 onward, base die functionality is becoming more sophisticated, making customer-specific optimization essential. This requires close collaboration with customers from the product development stage, and a broader set of foundry options makes it easier to respond to diverse AI chip designs.

SK Hynix is investing more than $4 billion (approximately 5.5 trillion won) to build an advanced packaging production base for next-generation AI memory in West Lafayette, Indiana. The company plans to complete the cleanroom by October 2028 and begin mass production of next-generation HBM in the second half of 2029. The vision is to ship cutting-edge wafers produced in South Korea to the United States, where they will undergo packaging and testing before being supplied directly to big tech customers.

This U.S. footprint reflects the memory industry’s shift from a standardized, high-volume production model to one centered on customized, high-value products. Some observers note that if demand visibility is secured early through long-term supply agreements with customers, the extreme boom-and-bust cycles of oversupply and price collapses seen in the past could be moderated.

Foundry Industry’s Price Hike Trend

Base die supply chain diversification also coincides with a supplier-favorable phase in the global foundry market. Samsung Electronics’ foundry division has implemented price increases of up to 15%, primarily for new 4nm process orders, with increases also extending to 5nm and 8nm wafer prices. With TSMC taking a wait-and-see approach on pricing, Samsung’s decision to move first underscores the shift in the foundry market’s balance of power toward suppliers.

Behind the rise in Samsung’s 4nm line utilization rates are increased orders for Qualcomm’s and Groq’s next-generation AI chips, as well as demand for the 4nm logic base dies needed for Samsung’s own HBM4 mass production. Analysts note that as global big tech companies sought alternatives amid saturation of TSMC’s leading-edge lines, Samsung gained substantial pricing leverage.

If SK Hynix adds Intel to its base die supply chain, the competitive dynamics in the advanced foundry market—currently being reshaped around TSMC and Samsung Electronics—are expected to become even more complex. In particular, if Intel integrates its EMIB advanced packaging technology into the HBM ecosystem, it could have significant implications for the competitive landscape in HBM back-end processing.