Growing up, nobody in my school knew what Korea was. Now it's the world's No. 3 AI power.

Growing up in the States in the '80s and '90s, there were exactly two kinds of Asian as far as my school was concerned. You were Chinese or Japanese. That was it. I'd tell my classmates that I was Korean and get a blank stare, and then usually a follow-up question about which of the other two it was closer to. As I got older and people asked, "What nationality are you?" I would say Korean, and they would ask me, "North or South?" That used to drive me crazy, and Korea has come a long way.

Then things started slowly changing. Samsung and LG went from the budget shelf to the entire TV wall at Best Buy. Hyundai and KIA stopped being a punchline. In 2012, you heard a Korean-language song, "Gangnam Style," on the radio. Then Parasite won Best Picture. K-pop exploded. BTS was selling out stadiums, Squid Game became the biggest show Netflix ever put out, and Korean fried chicken started showing up in strip malls that used to have a Chinese takeout spot and nothing else. Every single one of those felt like a fad at the time. None of them stopped.

Third place, with an asterisk

Here's the part I didn't see coming, and I watch this stuff daily. Back in April, Stanford's AI Index 2026 put South Korea third in the world in notable AI models. Third! Behind the United States and China, ahead of the UK, France, and Canada. The same report had Korea first in the world in AI patents per capita, at 14.31 per 100,000 people, and the government isn't being subtle about where it wants this to go. Korea's science minister Bae Kyung-hoon put it this way: "We will continue to strengthen policy support so that Korea can become one of the world's top three AI powers." Seoul calls the goal "AI G3," and it means that literally. The US, China, and Korea.

Now, here's the honest part, because third place comes with an asterisk. Third is a distant third. The US produced 50 notable models last year and China produced 30, and Korea built 5. The money gap is worse. Private AI investment last year came to $285.9 billion in the US and $12.4 billion in China, against $1.8 billion in Korea, which is less a gap than a different sport. Nobody in Seoul thinks a Korean chatbot is taking down ChatGPT anytime soon, and the same Stanford report flags a net outflow of AI talent on top of everything else.

And this didn't turn out to be one report that came and went. On August 13, the Seoul Economic Daily went back through the updated numbers and Korea was still sitting third. Their critique was a different one, though. Korea can build the models. What it doesn't have is a proving ground, a home market big enough to commercialize them and improve them at scale. One professor the paper quoted, Kang Jun-young of Hankuk University of Foreign Studies, put the smarter play in one line: "What matters is a strategy to seize the critical bottlenecks that cannot be left out."

The government's answer to that problem is honestly kind of wild. Seoul is making AI free for every citizen. The project is called "AI for Everyone," a free, unlimited AI chatbot and public-service agent for all 52 million Koreans, built mostly on Korean models, with the government covering costs and even handing the operators 512 Nvidia B200 GPUs this year. Today, August 28, the science ministry picked three consortiums led by SK Telecom, KT, and Kakao to run it. No other G20 country has tried anything like this. If your home market is too small to be a proving ground, you turn the whole country into one.

The bottleneck runs through Korea

Which brings me to the part Wall Street should actually care about, because Korea already owns one of those bottlenecks. Every AI accelerator Nvidia (NVDA) ships needs high bandwidth memory stacked right next to the GPU. That market runs through Korea. SK Hynix (SKHY) alone held 58% of it in the first quarter of 2026, according to Counterpoint Research, and Samsung held another 21%, so nearly 80% of the market is Korean. Micron (MU) has the rest. We built our HBM memory sector hub around this exact supply chain, and we covered how SK Hynix's HBM4 may cost Nvidia nearly twice as much as HBM3E. That's what owning a bottleneck gets you. The buyer complains about the price, and then the buyer pays it. SK Hynix is also breaking ground on a $4 billion HBM plant in Indiana, so this isn't even staying overseas anymore.

America designs the chips and Taiwan manufactures them, but the memory that makes them usable comes overwhelmingly from two Korean companies. Hold that up against the country I grew up explaining to people. The place my classmates couldn't find on a map is sitting on one of the few genuine choke points in the biggest trade on Wall Street, and it didn't get there with a moonshot or some Sputnik moment. It got there the way it got everywhere else, quietly, product by product, while everyone assumed the fad would pass.

I don't think the kid version of me would believe any of this.

I am not a financial advisor, and nothing here is investment advice.

Sources

• The Korea Times (April 14, 2026): Korea ranks 3rd globally in key AI models, securing top spot for patents
• Stanford HAI: The 2026 AI Index Report
• Seoul Economic Daily (August 13, 2026): Korea ranks 3rd in AI model development but lacks a proving ground
• The Korea Times (August 28, 2026): SKT, KT, Kakao consortiums selected for free AI service for public
• Counterpoint Research: Global DRAM and HBM market share, quarterly