AI fever: South Korea exports $60 billion in chips in a single month

On October 1, 2026, South Korea's Ministry of Trade, Industry and Resources reported that the country exported $120.9 billion in September, a monthly record. More than half came from a single product: semiconductors, at $60.3 billion, up about 263% from September 2025. It is the first time the country's monthly chip sales have topped $60 billion.

What it is / how it works

South Korea is home to Samsung and SK hynix, two of the world's largest memory chip makers. Memory is where a computer keeps the data it is working on. There are different types:

DRAM: fast "working" memory, used in phones, PCs and servers.
HBM (high-bandwidth memory): several layers of DRAM stacked and connected very close to the processor. This is the type used in AI accelerators, because giant models must move huge amounts of data every second.
NAND (flash): permanent storage, as in SSDs.

Training and running AI models requires data centres packed with accelerators, and each accelerator needs a lot of memory. With demand growing faster than factory capacity, prices rise — and export value soars.

Why it matters

The figure shows where one of AI's physical bottlenecks lies: processors are not enough, you need memory to feed them. It also shows how dependent the South Korean economy has become on this wave: chips accounted for about half of everything the country sold abroad in the month, and the trade surplus reached an unprecedented $49.85 billion. Exports to the US rose 137% and to China 123%, with semiconductors weighing heavily.

Evidence

Total exports: $120.9 billion (+83.5% year on year); imports: $71.09 billion.
Semiconductors: $60.3 billion (+263%), which the government attributed to "increased shipments and higher product prices."
From January to September, cumulative exports passed $800 billion; Industry Minister Kim Jung-kwan said the country is on track for $1 trillion this year.
On the supply side, US maker Micron said memory should remain "tight through 2028," with more than 75% of its 2027 output already committed, according to a SemiEngineering summary.

What could still go wrong

Memory is the most cyclical part of the chip industry. When prices rise, makers invest in new fabs; when that capacity comes online, there can be a glut and prices collapse. This has happened many times — the last big downturn was in 2022–2023.

A large part of the jump is price, not just volume. If data centre demand slows — for example, if AI companies cut spending or models become more memory-efficient — revenue could fall quickly. The comparison base also helps: September 2025 was a much weaker month, which inflates the percentage.

The minister himself cited global protectionism and Middle East tensions as risks. And there is concentration: when a country relies so much on one product, a downturn hits the whole economy.

What to watch