South Korea's semiconductor exports rose 331.2 percent year-on-year in the first 10 days of November, reaching $16.6 billion. Semiconductor products accounted for 49.7 percent of all South Korean exports during the period, with overall exports climbing to $33.5 billion from $15.7 billion a year earlier.
The surge reflects a fundamental tightness in the high-bandwidth memory market. Samsung Electronics and SK Hynix control the majority of global HBM production—the specialized memory architecture that Nvidia GPUs require for AI training and inference. SK Hynix has already sold out its HBM capacity through the end of 2025, a rare position that allows the company to operate at full utilization with no pricing pressure to compete.
Hyperscale cloud providers and AI startups are in a capital allocation race to build data center footprints. Each GPU cluster demands multiple HBM packages per accelerator, and the installed base of Nvidia H100 and H200 chips deployed across major cloud regions has exhausted available supply. Samsung and SK Hynix have announced capacity expansions, but lead times for advanced memory fabs stretch 18 to 24 months—meaning orders placed today won't reach production until 2026.
The concentration risk for South Korea is real but asymmetric. Memory commodity cycles have historically lasted three to five years before oversupply flattens prices. The current cycle, driven by AI infrastructure spending rather than consumer device upgrades, appears more durable: cloud providers are locking in multi-year contracts and building redundant data centers across geographies. That durability supports higher capacity utilization and pricing power for manufacturers.
Micron Technology, the third major global memory producer, has signaled it will accelerate HBM output. A shift toward competitive supply could tighten margins by 2026, but Samsung and SK Hynix have a 12-month window to convert current pricing into manufacturing investment and capacity lock-in.
