Korea’s $592B AI Memory Cluster Targets 2030 Output
2 min readSouth Korea has finally put dates on the largest industrial project in its history. Samsung Electronics and SK hynix are targeting June 2030 for first mass production at their joint 800 trillion won memory chip cluster in the southwestern Honam region, a build worth roughly 592 billion dollars and aimed squarely at the memory that AI accelerators run on.
What Is Being Built
The cluster will comprise at least four memory fabrication plants. Groundbreaking is slated for 2028, which means the two companies have a two year runway to finish site preparation before any concrete is poured. Gwangju, the regional hub, starts building out the power and water capacity next year. That sequencing is the real story: for a modern memory fab, grid connection and water supply are the long lead items, not the cleanroom.
Why Memory Is the Bottleneck
High bandwidth memory has become the choke point in AI hardware. SK hynix currently leads the HBM segment that feeds Nvidia accelerators, and Samsung has been fighting its way back into that supply chain while also chasing foundry work. Both companies are selling into demand that keeps outrunning supply, and both have concluded that the answer is capacity measured in decades rather than quarters.
Samsung alone has flagged annual capital investment above 73 billion dollars focused on AI semiconductors. The Honam cluster is that thinking taken to its logical end.
Why It Matters
A 2030 production date is a bet that AI memory demand is structural rather than cyclical, and it is being placed with public coordination in a single Korean region. It also concentrates risk. South Korea and Taiwan each passed Japan in total exports for the first time in the first half of 2026, driven by AI chip demand, and that export strength now rests on a small number of sites. Anything that disrupts Honam power, water or politics disrupts a meaningful slice of global AI supply.
The nearer term question is whether the schedule holds. Four fabs, two companies and one regional grid is a lot of coordination, and the market that justifies it has to still be there in 2030.