SK hynix Sets $713 Billion Korea Memory Plan As HBM4E Customers Stay Unnamed

SK hynix Names A 1,100 Trillion Won Manufacturing Plan

SK hynix is linking its next phase of AI-memory supply to a 1,100 trillion South Korean won domestic investment plan, equal to about $713 billion, as it expands manufacturing capacity in South Korea and prepares a U.S. capital-markets listing.

The memory chipmaker plans to fund the expansion partly through American depositary receipts on the Nasdaq Global Select Exchange.

The listing is planned for July 10 and is intended to raise 45.45 trillion won, or about $29 billion.

The disclosed spending is centred on fabs, packaging, NAND, HBM and memory manufacturing capacity.

SK hynix is already a central supplier in the AI accelerator supply chain, but the new disclosure is about manufacturing scale and capital needs rather than a finished customer deployment.

It also adds public-market financing to a capacity plan whose returns depend on future AI hardware orders.

Yongin And Cheongju Carry The Largest Fab Commitments

SK hynix said it will invest KRW 600 trillion, or about $390 billion, in the Yongin semiconductor cluster.

The company plans to complete construction of a fourth fab there by 2033, which EE Times described as 12 years earlier than the original plan.

The company also said it will allocate KRW 100 trillion, or about $65 billion, in Cheongju.

The Cheongju plan includes KRW 80 trillion for the M17 NAND production fab and KRW 20 trillion for the P&T7 advanced packaging facility.

The Southwestern-region allocation is KRW 400 trillion, or about $260 billion, with spending planned in phases.

SK hynix president and chief executive Kwak Noh-Jung said the Yongin site alone would not satisfy future market needs as AI services move beyond training and enter wider deployment.

SK Group's related infrastructure plan adds another scale marker.

The group plans to support the manufacturing locations by building 15 gigawatts of AI data centre infrastructure throughout South Korea, leaving the memory plan affected by power, land, cloud demand and wafer capacity constraints described by EE Times.

Data centre operators or cloud customers for that capacity remain outside the public record.

HBM4E Samples Add Product Evidence To The Spending Plan

The manufacturing plan is not only a long-range factory announcement.

SK hynix said it recently shipped 12-layer HBM4E memory samples to major customers.

The reported HBM4E modules reach up to 16 gigabits per second per pin, and the company says they use 20% less power than earlier versions.

The company's Advanced Mass Reflow Molded Underfill technology is described as raising memory-chip heat resistance by 17%.

Ahn Hyun, SK hynix's president and chief development officer, said HBM4E strengthened the company's AI-memory position and that SK hynix was working with partners to prepare mass production.

Those claims remain company-owned until customers, qualification status and production volumes are disclosed.

Nasdaq Proceeds Depend On AI Spending Staying High

The planned U.S. listing is presented as a funding tool for fabs, machinery and equipment rather than as proof that AI-memory demand is guaranteed.

SK hynix plans to deploy the 1,100 trillion won investment in phases, depending on market demand and cash flow.

According to Nomura Securities, global AI data centre investment will rise from $466 billion in 2025 to $3.379 trillion by 2030, with annual growth of about 48%.

The Bank for International Settlements warned that technology spending could fall sharply if AI tools do not produce enough financial returns to cover high upfront costs.

Those external warnings make the financing path part of SK hynix’s capacity plan for sustained hyperscaler demand.

SK hynix said it expects demand for advanced computing hardware to remain strong enough to justify additional manufacturing capacity.

The major HBM4E customers, the customer qualification timetable, the volume commitments behind the 12-layer samples, or the SK Group tenants and delivery schedule for the 15 gigawatts of AI data centre infrastructure remain outside the public record.