President Lee Jae-myung released the "Korea great leap three mega projects (hereafter mega projects)" through a national briefing on June 29, a high-stakes plan for Korea's economy that will inject 4,755 trillion won into semiconductors, artificial intelligence (AI) data centers, and physical AI (AI that affects the physical world). The plan to build four semiconductor fabs (factories) by investing a total of 800 trillion won by Samsung Electronics and SK hynix in Honam areas including Gwangju immediately grabbed the market's attention.
But after that, the market was crushed by fears of oversupply. From June 26, the last trading day before the national briefing, to July 13, Samsung Electronics and SK hynix shares fell 25% and 31%, respectively. The KOSPI also fell 19% during this period, with the index plunging to 6,806.93. Samsung Electronics released its second-quarter results with the largest quarterly operating profit (8.94 trillion won) in the history of Korean corporations, and SK hynix successfully listed ADRs (depository receipts) on the U.S. Nasdaq market, but these positives failed to reverse the worsening investor sentiment.
A high-stakes move to end the investment ice age
Behind the mega projects that push an AI great transition centered on a Honam semiconductor cluster, Chungcheong and Gangwon AI data centers, and a Yeongnam physical AI industrial complex lies an investment ice age that has gripped Korea's economy over the past three years. The growth rate (year over year) of gross fixed capital formation, which includes facility and construction investment, was negative in nine quarters from the third quarter of 2023 to the fourth quarter of 2025, except for the first quarter of 2024. A contraction in construction investment due to bad real estate project financing (PF) loans and a slowdown in new fab investment by semiconductor corporations during the 2023–2025 memory semiconductor (hereafter memory) downcycle created an investment gap across the economy. Although the AI investment boom in the U.S. is expected to lift this year's economic growth rate to around 3%, the warmth of growth is concentrated in semiconductors. Exports in the first half of this year were $496.7 billion, up 48% from last year, but most of the increase ($162 billion) came from semiconductors ($119.5 billion). The mega projects are based on the judgment that the government must step in to link a super-gap in memory to a national growth engine.
Seeing that leaving it solely to the market would make large-scale investment for the AI great transition difficult, the government proactively set the investment direction.
Memory is still a cyclical industry
However, Michael Burry, famous for the movie "The Big Short," posted on his investment site on July 1 (hereafter local time) that he views Korea's large capital expenditure plans as a signal that the AI investment cycle is nearing its peak. In a July 8 column, Bloomberg noted that Korea's mega projects could disrupt the existing competitive order in the memory industry and, together with the aggressive capacity expansion by China's ChangXin Memory Technologies (CXMT), amplify the impact of the coming downcycle. Global investment leaders reacted this way because, even if the AI industry follows a structural growth trajectory, memory as a component cannot escape cycles of shortage and glut. Allianz Research and Bloomberg Intelligence forecast that from 2028, when Samsung Electronics and SK hynix finish expanding their Pyeongtaek and Yongin fabs, excess demand will ease and memory prices will decline. Some say the Korean government underestimates that memory remains a cyclical industry even if the AI ecosystem grows structurally. This view, which suggests the possibility that hyperscalers could modulate investment based on profitability, is also evident in Morgan Stanley's July 6 report recommending an underweight in Samsung Electronics and SK hynix. The Bank for International Settlements (BIS), which released its annual economic report on June 28, warned, "Disappointment with returns on AI investment could turn a capital expenditure boom into an investment bust." Summarizing this perspective and global memory corporations' expansion plans, Kim Kwang-doo, head of the Korea Future Institute, said in a report, "From the second half of 2027 to the second half of 2029 corresponds to a barley hump in semiconductor supply and demand."
Building the AI ecosystem will decide success or failure
AI value added is created not only in memory but across the value chain spanning system semiconductors, packaging, servers, networks, software, and platforms. But in Korea, the semiconductor skew within the electronic industry and the broader AI ecosystem is severe.
According to the Ministry of Data and Statistics (MODS) Korean Statistical Information Service (KOSIS), using 2020 as the base of 100, the production capacity index for semiconductors rose to 195 (as of May 2026), expanding the production base to nearly double in six years. By contrast, electronic components were 66, communications equipment 97, special-purpose machinery such as robots 89, and computers 28, all below the base. In particular, after the AI revolution in 2023, semiconductor production capacity expanded 40%, while the rest of the electronics sectors contracted 10%, weakening the fundamentals.
In particular, servers—core to AI data centers—have a tiny domestic market, as shown by KTNF, the only domestic corporation with server design and production capacity, posting just 25 billion won in sales last year. Regarding the AI ecosystem, the government said it would invest 81 trillion won in the Chungcheong region to foster a cutting-edge semiconductor packaging base and develop Yeongnam into a materials, parts, and equipment (MPE) supply chain hub, but these do not seem to be priorities. There is no visible strategy to foster servers and networks at all. At this rate, the AI data centers built under the mega projects will depend on foreign-made servers and systems. This is why Vice Chair Kim Sung-sik of the National Economic Advisory Council said, "Expanding semiconductor fabs is only part of the AI ecosystem strategy; we must foster inference chip fabless firms, network technologies, software, and MPE together." The success or failure of the three mega projects hinges on how to expand memory competitiveness into AI ecosystem competitiveness. Building memory fabs alone does not establish the AI value chain. The winning edge in AX lies more in the direction of building the AI ecosystem than in the size of investment.
Plus Point
Taiwan's AI server exports surpass TSMC
While Korea "went all in" on maintaining a super-gap in memory, Taiwan has been bringing its entire manufacturing sector into the AI value chain centered on AI servers. According to Taiwan's Ministry of Economic Affairs, in January–May 2026, Taiwan's computer and server exports were $120.56 billion, accounting for 35.3% of total exports (about $342 billion), surpassing semiconductors ($109.06 billion, 31.9%). Computer and server exports, which were $84.6 billion in 2024—about half of semiconductors ($165 billion)—grew to nearly match semiconductors ($209.2 billion) in 2025 at $183.5 billion, and this year became Taiwan's top export item. Taiwan's economy, once dependent on a semiconductor industry centered on TSMC, added AI infrastructure as a new growth axis. The Taiwan government's forecast of 9.64% GDP growth this year, with even the possibility of double-digit growth mentioned, is thanks to its AX (AI transformation) strategy for manufacturing. Taiwan's "AI Taiwan Action Plan," which has supported AI infrastructure since 2017, is bearing fruit. Taiwan achieved manufacturing AX through servers, power, and networks, turning it into a new growth engine.