Editor’s note: South Korea has basked in the glow of Nvidia’s epic rise and now stands exposed to the fallout from the sudden departure of Jensen Huang, a central architect and driver of the AI boom. His exit comes as scrutiny of soaring AI spending and resistance to data center construction test the durability of a boom that has powered Korea’s semiconductor earnings and export surge. In this four-part series, AJP examines the ripple effects on chipmakers, the wider AI supply chain and the broader economy.
SEOUL, November 13 (AJP) - President Lee Jae Myung and Samsung Electronics Chairman Lee Jae-yong turned the first shovels of earth on Thursday at a groundbreaking ceremony to scale up the company’s decades-old chip assembly and testing complex in Onyang into a cutting-edge high-bandwidth memory (HBM) packaging hub while Nvidia — a driving force behind South Korea’s ambitious chip republic project — was weighing the ouster of its founder and AI crusader Jensen Huang.
"This site, once focused on assembling and testing conventional semiconductors, will be transformed into a cutting-edge HBM hub at the heart of AI chips," the president declared in Asan, South Chungcheong Province.
The city is positioned to bridge a growing chip belt connecting the industry’s established bases in Gyeonggi Province with planned hubs in Honam, the southwestern region encompassing Gwangju.
Samsung plans to invest 46 trillion won ($34 billion) in the Onyang campus over time, including 6 trillion won for facilities and 40 trillion won for advanced equipment. Its new production line is scheduled to begin operations in June 2028, about a year earlier than originally planned.
The investment is part of a broader industrial buildout stretching from Samsung’s semiconductor complexes in Pyeongtaek and Yongin to its planned factories in Gwangju, alongside SK hynix’s expanding manufacturing network in Yongin and Cheongju.
Together, the projects represent a multiyear bet that demand for AI computing power will remain strong enough to justify new fabrication plants and advanced packaging facilities well into the next decade.
Huang’s abrupt departure adds uncertainty to that bet. Nvidia named Chief Financial Officer Colette Kress as interim CEO, and the transition has not been accompanied by any announced changes to Nvidia’s HBM procurement or capital spending plans. Nevertheless, it adds a new variable to an industry already questioning how long the extraordinary pace of AI infrastructure investment can last.
For Samsung, the expansion involves coordinating the factories that manufacture chips with the facilities that turn them into finished products.
HBM4 illustrates the division of labor. Samsung’s design combines sixth-generation 10-nanometer-class DRAM, known as 1c DRAM, which stores data, with a 4-nanometer logic base die that controls signal transmission.
The components are made through separate memory and foundry processes. Packaging brings them together into a single product, stacking, connecting and testing the chips before they can be supplied to AI processor makers.
More DRAM production in Pyeongtaek will therefore translate into more HBM shipments only if packaging facilities can handle the additional output. Expanding chip fabrication in the Seoul metropolitan region and advanced packaging in the Chungcheong region are complementary investments.
Giheung develops and validates next-generation processes before mass production. Samsung plans to invest 20 trillion won in its integrated research and development complex there by 2030.
Hwaseong houses DRAM production lines and foundry operations using extreme ultraviolet (EUV) lithography, a technology for printing fine circuit patterns. Pyeongtaek handles mass production of memory and advanced foundry chips, with its fourth production line expanding capacity for the 1c DRAM used in HBM4.
The Pyeongtaek and Onyang expansion schedules are closely aligned.
Samsung has identified 2028 as the year its fifth line in Pyeongtaek’s second complex will begin operating as a key HBM production base. Onyang’s fifth line, which broke ground Thursday, is targeting June of the same year.
The company stressed that the 46 trillion won commitment is a medium- to long-term investment in five HBM production lines at the Onyang campus, rather than the construction cost of a single building.
The campus, which began operations in 1991, is shifting from conventional semiconductor assembly and testing toward the advanced stacking required for high-performance memory. Nearby Cheonan is also expanding and modernizing facilities to handle HBM.
The importance of the Chungcheong region lies increasingly in the products it can complete, beyond its geographical position between existing and planned manufacturing hubs.
Samsung’s next major fabrication expansion in the metropolitan region is in Yongin’s Cheoin district. Separate from its existing Giheung campus, the company plans to build six fabs in stages at the advanced system semiconductor national industrial complex, with the first production line targeting October 2029.
The planned facilities will manufacture circuits on silicon wafers, extending Samsung’s fabrication footprint from southern Gyeonggi into southwestern Korea. Advanced packaging operations in Chungcheong would help complete the expanding production network.
SK hynix is also accelerating its expansion.
In August, the company approved 54.3 trillion won in additional investment for two new fabrication facilities: 35.2 trillion won for its second Yongin plant, known as Y2, and 19.1 trillion won for M17 in Cheongju.
The first cleanroom at Y2 is scheduled for June 2029, while M17 is targeting December 2028. The company is also developing its first Yongin facility, Y1, and expanding production capacity at its Cheongju manufacturing complex.
The scale of those commitments reflects the increasingly high stakes of the global AI race.
SK Group Chairman Chey Tae-won acknowledged the extraordinary sums required during a lecture at Seoul National University on Nov. 5.
"The scale and very concept of money have changed completely," Chey said, recalling the announcement of a 2,100 trillion won investment initiative associated with the government’s regional megaprojects.
"The whole world is in this game. We cannot keep up unless we make bets of this magnitude."
His remarks capture the dilemma facing Korean chipmakers. Investing too cautiously risks falling behind in the next phase of AI development. Expanding too aggressively could leave them exposed if demand slows before new facilities reach full production.
Chey also warned of heightened market volatility after a sharp selloff in semiconductor stocks following the U.S. midterm elections.
"This roller-coaster market is bound to continue," he said.
The warning comes as investors debate whether the memory industry’s extraordinary earnings growth is approaching a cyclical peak.
Samsung and SK hynix are still enjoying strong demand for HBM and conventional memory. But their longer-term outlook remains sensitive to AI infrastructure spending, memory prices and the timing of new capacity.
A slowdown or plateau in growth from 2027 would not necessarily end the boom. It could, however, change the conditions under which factories planned during the current heyday begin operating.
Major facilities scheduled to come online from 2028 onward will add supply to a market whose growth trajectory remains uncertain. Decisions made while demand is racing ahead will be tested against the orders and prices available several years later.
For both companies, maintaining profitability will increasingly depend on more than producing additional chips.
Energy efficiency is emerging as a critical competitive factor as data center operators confront rising electricity costs, power-grid constraints and pressure to improve computing performance per watt.
Advanced memory packaging, lower-power chips and improved thermal performance could help Korean suppliers strengthen their positions even if the pace of new data center construction moderates.
Customer diversification could provide another avenue for growth.
Nvidia remains a major force behind global HBM demand, but competing AI accelerator developers and cloud computing companies designing their own chips offer additional opportunities for memory suppliers.
Korea’s chip republic is therefore pressing ahead with a construction timetable that extends far beyond Huang’s departure. The strategic challenge is whether its chipmakers can sustain demand and profitability when the factories now breaking ground begin delivering output.
The investments underway in Onyang, Pyeongtaek, Yongin and Cheongju will be tested long after today’s semiconductor earnings records have been set.
Copyright ⓒ Aju Press All rights reserved.