The AI a Rock Drummer Built — China's Six AI Founders, and the Ladder Korea Lacks

The man who built Kimi was a rock band drummer in college. The first time I read that line, I filed it away as a fun bit of trivia. So a founder had a band in his past. What of it?

But the more I dug into Yang Zhilin (杨植麟), the more my mind changed. He was born in 1992. His favorite band is Pink Floyd, his desert-island album The Dark Side of the Moon. The company he founded, Moonshot, took its name from that album, and he timed the launch to the record's 50th anniversary. The chatbot's name, Kimi, is his own English nickname. He put it this way: "A truly great company needs cultural depth. Technology alone has no soul."

One man's taste, maybe. Except the résumés sitting next to his were too similar to be coincidence. DeepSeek's Liang Wenfeng, MiniMax's Yan Junjie, StepFun's Jiang Daxin, Zhipu's Zhang Peng, Baichuan's Wang Xiaochuan. The six people steering China's AI right now share a set of traits. And following those traits led me, in the end, back to Korea.

Small-town boys, children of reform

The first thing that stood out was where they came from. Not the pedigreed families of Beijing or Shanghai. Yang Zhilin is from Shantou in Guangdong, Liang Wenfeng from Zhanjiang in the same province, Yan Junjie from a small county in Henan, Wang Xiaochuan from Chengdu in Sichuan. Mostly inland, or provincial cities in the south. In Korean terms, most of them grew up in mid-sized towns that don't even rank as metropolitan cities.

Their birth years cluster too. The oldest, Liang Wenfeng, was born in 1985, Yan Junjie in 1989, Yang Zhilin in 1992. They are the first generation to spend childhood and adolescence in the years when China threw its doors open under reform and opening. Poor, certainly, but watching the country grow visibly richer year after year. The way Koreans who lived through the high-growth 1980s watched buildings rise and paychecks climb overnight, these kids felt it as children. The reform generation knew in their bones that study, that starting a company, could change a life.

Provincial boys who go to Beijing, to Tsinghua, then abroad to study, then return to succeed. That path was itself a kind of grand Chinese hero's journey.

Kids for whom math was the exit

The second thread is more interesting: math. Each of them was obsessed with math or programming as a child, and most walked into that world on their own.

Yan Junjie found the school curriculum too slow, so he taught himself calculus. Wang Xiaochuan was writing software at seven and won gold at the International Olympiad in Informatics in 1996. Yang Zhilin, having never programmed before, placed first in Guangdong's informatics olympiad after a single year of middle-school study. He then passed Tsinghua's independent admission and took the top science score in all of Shantou on the gaokao. You could dismiss it as a story of geniuses, but Korea has had its share of perfect-score prodigies too, and Korean students routinely rank at the very top of the international math olympiad. Which only makes their story more worth asking about.

Of the stories I read, a line from MiniMax's Yan Junjie stuck with me: he "realized early that I could never be a top mathematician, and found my interest in AI instead." At that point their love of rock and their love of math suddenly made sense to me. A depth of devotion to what you love. And an aim that reaches past the technology itself. Does Korea, I wondered, ever produce people who genuinely love math or science and end up in AI or a startup?

The American experience

The third thread is America. Most of them experienced the world's most powerful nation, the Silicon Valley frontier, firsthand.

Yang Zhilin did his PhD at Carnegie Mellon after Tsinghua. His advisors are now senior figures at Apple and Google DeepMind, and during his doctorate he worked at Google Brain and Meta. He co-authored papers with Turing Award winners Yoshua Bengio and Yann LeCun. He is the first author of Transformer-XL and XLNet, core papers in the transformer lineage — in plain terms, methods that let a model keep notes on earlier sentences and read the next one in context.

Jiang Daxin's path is also striking. He spent 16 years at Microsoft in the United States, building the Bing search engine and the Cortana voice assistant, and rising to chief scientist and vice president at Microsoft Research Asia. A man who went deep into the heart of American Big Tech and helped shape an era. He left because he couldn't do what he wanted to do. When ChatGPT dropped in late 2022, he handed Microsoft a ten-page proposal to build their own large model. But Microsoft chose to invest in and support OpenAI rather than build its own. "If I stay here," he said, "I'll have to watch AGI grow up with nothing to do with me," and he resigned. There's something admirable in that.

"We never intended to build a small company. We're going for AGI. Otherwise, why would these people have gathered?"

Big ambition. Not a repetition of small wins, but a vision to build the whole board, backed by the ability to pull it off. People who lived through the world's best arenas came back to China to lay out a new one.

The return, and China's cub-raising strategy

To boil it down: the core of the story is a provincial science prodigy who passes through America's best universities and companies, then comes back to start a company. (There are exceptions, of course.) But the fact that this story can repeat in China forces a hypothesis — that China has a structural system underneath it. Four layers of it.

First, a shortcut for prodigies. China's exam hell is as brutal as Korea's, if not worse. The gaokao is far more savage than the Korean CSAT. And yet China has a shortcut for its prodigies. Excel at a math or informatics olympiad and you skip the gaokao entirely, straight into an elite university. Yang Zhilin and Wang Xiaochuan both took that route. Tsinghua, the top science-and-engineering school, even has a special track called the "Yao Class" (姚班), the computer-science genius program built by Yao Chi-Chih, China's only Turing laureate, after he left Princeton for Beijing in 2004. Tens of thousands, hundreds of thousands walk this path every year. The statistics say China produces 5 million STEM graduates a year — ten times the roughly 500,000 in the US, and more than 30 times Korea's roughly 150,000. The prodigies, cherry-picked from those 5 million, pour into AI and math.

Second, geniuses raising geniuses. Zhipu's founder Zhang Peng and chief scientist Tang Jie both came out of Tsinghua's Knowledge Engineering Group (KEG). And Professor Tang personally trained a generation of young AI leaders, Yang Zhilin among them. A single lab birthed both Zhipu and Moonshot. America once had its "PayPal Mafia," but in China the master-and-apprentice relationship often means founding companies together and pulling each other up. Whether that's entirely healthy I'm not sure, but China's dense university networks map directly onto its industry.

Third, first-generation founders who raise their competitors. However good the idea and the talent, in AI you can't win without capital. This is a field that demands astronomical investment. What's astonishing about China here is that its first-generation founders keep investing and cultivating the next wave. Alibaba has its own model, Qwen. And still it invests heavily in Moonshot. Alibaba and Tencent hold stakes in a large share of the startups known as the "Six Little Tigers" (六小虎). Literally raising tiger cubs. Baichuan raised $700 million from Alibaba, Tencent, and Xiaomi within six months of founding. Where American Big Tech absorbs promising talent and startups to bulk itself up, China's first generation leans toward growing the board and buying an option on future value. That backdrop, of course, is inseparable from the role of the Chinese government.

Fourth, the state's patience. Beijing set up an $8.2 billion AI fund, and local governments hand out compute vouchers and model subsidies to buy computing power. It applies regulation strictly to the giant platforms while leaving room for the cubs, so innovation isn't stifled. Behind individual genius sits the structural support of the state. Something I felt in Shenzhen was exactly this: that another invisible hand — the government — was correcting the market's contradictions and imperfections.

None of this is to call the story ideal. Alibaba's patronage is no act of pure goodwill; it's a strategic bet. Everyone knows Tencent has taken stakes in countless companies and startups worth money, in Korea as well as China. For years Alibaba went around with cash in hand, pledging to invest in domestic unicorns. The more the cubs depend on Alibaba's cloud and capital, the more their future independence wobbles. The state's support, too, is better read as strategic positioning for national growth and the US-China rivalry than as investment in the future.

So, are they actually that good?

Honestly, I still don't know how effectively China's AI is used in the field once you set the benchmarks aside. I've tried DeepSeek briefly, and the impression was that it was more inconvenient than I'd expected. But after living through the recent Anthropic Fable 5 episode and feeling American control firsthand, it struck me that a real opening is arriving for the Chinese models too.

Where China is strong right now is open weights — public models anyone can download and run. There, China is the strongest. Zhipu's GLM-5.2 ranked fourth overall on an intelligence index, and Moonshot's Kimi K3, released in July, became the largest public model ever at 2.8 trillion parameters. Competition in coding is ferocious right now, and it took the world's number one spot on a front-end coding benchmark — genuinely not to be dismissed. DeepSeek's inference cost is a tenth of the American models'. Four of the top five public models are now Chinese.

Of course, ask which AI is the best and the answer is still an American company. The top spot for user preference is Anthropic's Claude, and Claude is what I find most comfortable too. But the gap has closed to a razor's edge. Talk of a "two- or three-month gap" is no longer a joke. What's left is originality. Liang Wenfeng said it in an interview: "The real gap between Chinese and American AI isn't one or two years — it's the difference between original creation and imitation. It's time to stop following and lead."

In our terms, the fast-follower strategy has reached its final stretch, and they've climbed to pace-setter. The day they can boast original capability, real originality, may not be far off.

In Korea, too

Having interviewed plenty of domestic founders myself, I felt a lot of envy watching the stories from China's AI scene. Korea is famous for its deep talent pool, and its students go through a ferocious admissions gauntlet. So why do we repeat the exam two or three times over to get into medical school, and then, after graduating, do it again and again to land at a conglomerate?

Covered in the KBS documentary War for Talent, China's top science prodigies go through the olympiads and the Yao Class into AI and math. Korea's top talent, by contrast, goes into medicine. The documentary title "China Mad for Engineering, Korea Mad for Medicine" didn't come from nowhere. For the 2025 cycle, applications to the science-and-technology institutes (KAIST and others) fell 28%, while medical-school applications rose 29% over the same period. At Seoul National University's engineering college, more than 100 of the 850 first-years drop out each year; in the department of chemical and biological engineering, one article reported a quarter of the incoming class quit within a year to prepare for medical school. What about semiconductors? At Yonsei and Korea University, 71% of those admitted to the contract semiconductor programs declined to enroll. And even the talent we do raise, once graduated, heads for American Big Tech rather than staying home.

The nature of the capital differs too. Korea's unicorns are mostly consumer platforms. Toss is the classic example. Only the grow-fast, exit-fast market has taken root. For capital-intensive deep tech, the investment and the patience are thin. Over the past four years the US minted 229 new unicorns, mostly AI-centered. Korea managed two. It wasn't until 2026 that Upstage became Korea's first generative-AI unicorn.

The posture of the big corporations differs as well. Where China's Alibaba and Tencent lean toward growing a competitor first, Korea's conglomerates lean toward absorbing the promising thing and internalizing it. Hyundai's 2022 acquisition of 42dot is a good example. The 420 billion won it poured in was billed as the core of its software-defined-vehicle strategy, but inside the conglomerate it hasn't shown the standout results many expected. A tendency to put application over pure science or technology, commercialization over origination, runs strong.

Thankfully, the current is shifting a little here too. A recent statistic showed that the share of gifted- and science-high-school graduates going into medical and dental programs fell 44% in two years. It may be a flash of popularity riding the semiconductor cycle, but it may also be a sign of talent flowing back to science and engineering. Still, I'd rather it not be a boom driven by rewards alone. Moonshot's Yang Zhilin named Steve Jobs as his hero, calling him "the man who scaled taste." He saw the code inside Apple's success for what it really is.

There's a line in 20th Century Boys, a manga I loved: "Rock and roll will save the world." The boy who played drums in a college band gave his company the name of an album he loved, and that company is now shaking the world. For him, taste was the dream and the passion. Maybe what we lack isn't talent, and isn't capital. It's the gaze that waits for someone to push what they love all the way through — the room to not treat that romance as a luxury. That's what I envied most.