SK Hynix Weighs U.S. Memory Chip Production as AI Customers Seek Secure Supply

SK Hynix is examining whether to expand its planned U.S. manufacturing footprint beyond advanced chip packaging and into memory wafer production, as artificial intelligence customers and government policy change the economics of building semiconductor capacity in America.

The Korean chipmaker has not made a final decision. It said it is exploring options to strengthen its global competitiveness after reports that it could lease part of Intel’s Ohio complex or form a venture involving Intel and major cloud companies. Solidigm, its U.S.-based NAND subsidiary, is also reportedly considering a separate American NAND plant.

SK Hynix already has a more than $4 billion project underway in Indiana. The facility, which broke ground last month, is scheduled to begin operations in 2029. It will receive advanced DRAM wafers made in South Korea and package them into high-bandwidth memory, the specialized chips used in AI accelerators. Under the current plan, the wafers themselves would continue to be fabricated in Korea.

Moving front-end production to the United States would represent a larger and more expensive commitment. South Korea retains important cost advantages because chip factories, skilled engineers and suppliers are concentrated around established manufacturing hubs. Construction and labor are also more expensive in the United States.

Those disadvantages may no longer settle the decision on their own. Federal incentives, possible tariff benefits and long-term purchase commitments from large cloud companies could reduce the commercial risk of a multibillion-dollar plant. For major AI customers, guaranteed access to memory supply has become increasingly important as demand for high-bandwidth memory grows. Producing closer to those customers could also strengthen coordination with research partners and equipment suppliers.

Intel’s unfinished Ohio facilities offer another potential advantage. Intel began construction there in 2022, but the project has fallen behind schedule, with initial operations now expected around 2030 or 2031. A deal could give SK Hynix access to a site where infrastructure work is already advanced while helping Intel improve future factory utilization. However, such an arrangement would not eliminate the higher cost of operating in the United States.

Solidigm faces a different geographic calculation. Its NAND production is concentrated in Dalian, China, so a U.S. factory could diversify manufacturing and reduce exposure to tariffs or tighter restrictions on semiconductor equipment shipped to China.

Any American expansion would supplement rather than replace SK Hynix’s Korean production base. The company approved a combined 54.3 trillion won in August for a second DRAM plant in Yongin and a NAND plant in Cheongju. Those investments underscore that Korea remains central to its capacity strategy even as pressure grows to place more production near U.S. AI customers.

The review reflects a broader shift in the semiconductor industry. Manufacturing cost remains critical, but supply security, trade policy and guaranteed demand are carrying more weight in decisions about where the next generation of memory capacity will be built.