Samsung and SK Hynix Already Make Most of AI's Memory. Now Korea Wants to Make That Lead Permanent.

South Korea is not racing into AI chips. It is defending a position it already holds.

Samsung Electronics and SK Hynix together control about 80% of the high-bandwidth memory chips that sit beside every Nvidia AI accelerator, according to Yonhap's English service. High-bandwidth memory, or HBM, is the stacked memory that AI processors use to hold the model weights and intermediate calculations they need to run. Without it, the GPUs sit idle. In late June 2026, the country's new president, Lee Jae Myung, stood alongside the chairmen of both companies and announced more than $576 billion in planned chip and AI investment. The pledge dwarfs the U.S. CHIPS Act, the 2022 American semiconductor subsidy law, in announced scale. Two of the four new fabs will sit in the country's underdeveloped southwest, where the bottleneck is power, water, and skilled labor rather than capital.

The number is real as a pledge and aspirational as a schedule. Samsung committed roughly 400 trillion won (about $259 billion) for fabs near Gwangju. SK Hynix committed a comparable sum. Each will build two new fabs in the underdeveloped southwest. Local governments in Gwangju and South Jeolla province will layer in 5 to 20 trillion won (about $3 billion to $13 billion), according to the Korean financial press. Another 81 trillion won (about $52 billion) goes to a packaging cluster in the Chungcheong region, and 30 trillion won (about $19 billion) funds research into next-generation memory, on-device AI chips, and defense semiconductors, SBS reported. Lee called the strategy a "triple axis" of semiconductors, physical AI, and AI data centers.

The market has already priced the entrenchment. SK Hynix overtook Samsung as South Korea's most valuable listed company in June 2026 after its shares surged more than 300% year to date. That surge tracks one customer. SK Hynix supplies the majority of the high-bandwidth memory that goes into Nvidia's AI systems. Samsung and SK Hynix together produce about two-thirds of the world's memory chips overall. The wager is that whoever keeps that share keeps the toll booth.

The bottleneck is not capital. It is the southwest grid.

SK Group Chairman Chey Tae-won offered the cleanest falsifier in the room. Building out SK Hynix's major Gyeonggi Province manufacturing cluster took nine years, he noted. The new southwest plants, he said, will require "vast sites, along with sufficient power, water and skilled workers." That is the constraint the $576 billion has to clear, and it is the one the wire stories tend to bury under the headline number.

The southwest region has the land and the political backing. It does not have the high-voltage transmission, the industrial-grade water treatment, or the trained semiconductor workforce that the Seoul-area clusters spent a decade assembling. Chey's nine-year benchmark is a hint that the announced capex is the easy part of the bet. The hard part is the cadence. Fab construction, tool installation, and yield ramp have to overlap with the buildout of the grid and the water system underneath them.

The Korean financial press has made the same point: the $576 billion figure is a multi-year sum, not a near-term spend, and the southwest's infrastructure gap, not the dollar total, will decide whether the country doubles its HBM share or merely holds it. For the United States, China, and Japan, the read is similar. If Korea's bottleneck binds, AI's most concentrated supply chain gets longer and more fragile just as the buildout accelerates.

The next test is not another announcement. It is whether the first new southwest fab breaks ground on a grid and a workforce the country has not yet built.