Late on July 16, 2026, when Yang ZhiLin's Kimi K3 model went live, no one anticipated what would happen next.
On day one, it topped the Arena AI frontend coding competition with 1679 points, surpassing Claude and GPT. Overseas, some are calling it the “DeepSeek 2.0 moment.”
The next day, Musk wrote "Impressive" on social media and announced that his new 2-trillion-parameter model "may surpass Kimi," to which Moonshot AI responded: "Welcome to the '2 Trillion+ Club,' Musk."
On the third day, Kimi user requests exceeded estimates and approached the cluster's capacity limit, prompting the team to issue a late-night announcement suspending new user subscriptions.
On day four, U.S. White House officials publicly accused Moonshot AI of "stealing technology" and "circumventing chip export controls".
On day five, Nasdaq opened down 1.8% due to factors including K3's release, and the Philadelphia Semiconductor Index fell 5.2%.
On day six, Bloomberg reported: K3 is believed to potentially narrow the AI gap between China and the U.S., with some scholars suggesting the gap could shrink to two to three months.
On the seventh day, Hu Xijin posted a message reminding Yang Zhilin, "Don't go to the U.S. for now."
In seven days, a 90s-born man from Shantou sent Silicon Valley, the White House, and Wall Street into chaos.
Code is the tool, rock is the bone
Born in 1992 into an ordinary family in Shantou, Guangdong, Yang Zhilin’s journey was different from the start.
In his youth, he had two passions: rock music and coding. He attended Shantou Jinshan High School, and despite having no prior programming experience, he was selected for the Informatics Olympiad training program. Most of his classmates had been coding since middle school, so he started alarmingly late. Yet, just one year later, he won first prize in the Guangdong regional round of the National Youth Informatics League and earned a direct admission offer from Tsinghua University.
But he was determined to prove he wasn’t just a “recommended student.” He qualified again through independent admission, then chose to take the national college entrance exam as a regular candidate—scoring 667 points and becoming the top science student in Shantou. He was admitted to Tsinghua University three times, becoming a local legend.
In 2011, Yang Zhilin, the top science student in Shantou's college entrance exam (center), was offered free airline tickets by Air China. (Source: Hualong Chaoshan Network)
After entering Tsinghua University, he was reassigned to thermal energy engineering—commonly known as “burning boilers.” In his sophomore year, he made a decision that puzzled those around him: switching his major to computer science. His reason was refreshingly “rock and roll”: he liked it. And this fondness came from a short story by Haruki Murakami—a character who stayed up late coding to bring technology to life filled him with awe.
Switching majors meant he had to make up all the freshman programming courses. Yet he graduated at the top of his class, with 90% of his major courses scoring above 95%. Meanwhile, he founded the rock band Splay at Tsinghua University, serving as drummer and songwriter. The band’s name is a clever pun derived from the data structure “Splay Tree”—and in 2014, they won the “Best Original Song Award” at Tsinghua’s campus song competition.
Top GPA, numerous papers
Years later, Yang Zhilin recalled his biggest regret during undergraduate years: “My band didn’t win the championship in the original music competition; we only received an award for original song.”
Entrepreneurship is written into the genes of Chaoshan people.
In 2015, Yang Zhilin graduated first in his class from Tsinghua University’s Department of Computer Science and went on to pursue his Ph.D. at Carnegie Mellon University under Ruslan Salakhutdinov, Apple’s first AI director. He graduated four years later—two years faster than the typical six-year timeline.
During his PhD, he conducted two highly cited works: Transformer-XL and XLNet, with a combined citation count exceeding 20,000. Transformer serves as the foundational architecture for all large models today, and his contributions amounted to key improvements on this core framework. As a result, he became the most cited researcher in China under the age of 35 in the field of NLP.
After the release of K3, Yang Zhilin became famous, and many people are asking: Why didn't he stay in the U.S.? Some speculate it's due to visa issues, others say he didn't get selected in the H-1B lottery.
Rumors had spread too far, so his advisor, Russ Salakhutdinov, had to step in to clarify: the truth is that, given Yang ZhiLin’s level, he had countless opportunities in the U.S. as he neared graduation—Apple wanted to hire him, Google and Meta had offered him positions, and both Stanford and MIT asked if he would consider a postdoctoral role. An Apple executive even offered him a position at their Beijing office. Salakhutdinov said, “I remember him telling me that he would regret it for the rest of his life if he never even tried to start a business. I respect his decision, and he’s right.”
In February 2023, Yang ZhiLin began focusing on the first round of funding.
He later recalled that it was an extremely narrow window: “If we delayed until April, there would be virtually no chance. But if we tried in December 2022 or January 2023, there was also no chance—there was still the pandemic, and people hadn’t reacted yet—so the real window was just one month.” While still in the U.S., he did a precise calculation one evening and concluded he needed at least $100 million within a few months. At the time, very few in the market were raising funds, and many doubted he could secure that much. It turned out he could—and even more.
On April 17, 2023, Moonshot AI was founded in Beijing. The company’s name is inspired by Pink Floyd’s album “The Dark Side of the Moon,” which the founders had listened to for years. Among the co-founders is Zhou Xinyu, the lead guitarist of Splay, a band from Tsinghua University. From music to AI models, the two transitioned from playing in a band to building models.
Six months later, the Kimi AI assistant was officially launched, pioneering a direction no one else dared to pursue at the time—ultra-long context. With lossless input support for 200,000 Chinese characters, it offered the longest context length among all large model products globally. For comparison, Claude supported around 80,000 characters, and GPT-4 only about 25,000. Yang Zhilin believed that context length was the key to unlocking new applications—from analyzing entire codebases to building long-term assistants that never forget critical information—both requiring this foundational capability. While others competed on general benchmarks, he bet that long-form text would be the next gateway.
You called it right.
In November 2023, Kimi officially opened its services to the public. In February 2024, Moonshot AI completed an A+ round of financing exceeding $1 billion, led by Alibaba, with participation from Sequoia Capital China, Xiaohongshu, Meituan, and others, resulting in a post-money valuation of approximately $2.5 billion—the largest single-round financing ever for a Chinese AI large model company at the time. Within less than a year of its founding, Moonshot AI became one of the most highly valued unicorns in China’s large model sector. In March of the same year, Kimi increased its context length from 200,000 to 2 million characters. Monthly user traffic rose by 321.58%环比, causing the platform to temporarily crash due to overwhelming demand.
From its establishment in April 2023 to becoming an AI unicorn, Yang Zhilin achieved this in less than a year.
Yang and DeepSeek’s founder Liang Wenhong are often compared. Two fellow Guangdong natives, they have built the two most prominent AI companies in China.
But their backgrounds differ. Liang Wenhong took a local path—entering Zhejiang University at 17, starting with quantitative investing, and using his own earnings to invest in AI. Yang Zhilin followed an academic route.
In terms of business model, Kimi is a star company driven by funding and closely tied to its product; DeepSeek is more like a research institution, operating long-term on its own capital, with model capabilities always being the most discussed topic in the industry.
The narrative differences are more pronounced; someone summarized it directly: Liang Wenhong claimed the赛道 with three keywords—open source, low cost, and domestication; Yang Zhilin, with globally leading model capabilities, is fighting on a battlefield defined by others.
However, by 2026, the two paths began to converge. Kimi increasingly focused on foundational capabilities, while DeepSeek started paying attention to reasoning efficiency and productization, blurring the boundaries.
China's AI talent is drawing the attention of Silicon Valley
Yang Zhilin and Liang Wenhong are not the only ones in the spotlight.
In the same month, Zhipu's market capitalization briefly surpassed HK$810 billion, exceeding Xiaomi. The company, founded by Tsinghua University professor Tang Jie, had only gone public on the Hong Kong Stock Exchange six months prior. Tang Jie, now 49, is the eldest of the four founders. In July, he sent an internal memo announcing that the company would focus exclusively on foundational AGI research over the next two years, without pursuing short-term monetization. Interestingly, Yang Zhilin was once Tang Jie’s student during his time at Tsinghua—now, teacher and student find themselves on the same battlefield, becoming each other’s most familiar rival.
Yan Junjie of MiniMax focuses on mixture-of-experts models and multimodal technology. Around the same time, he also sent an all-hands email announcing that he would forgo his salary until AGI is achieved and pledged 4% of his personal shares to incentivize the team. At that time, MiniMax’s stock price had dropped from HK$1,330 to HK$297.
Four individuals—Liang Wenheng, 41; Yang Zhilin, 34; Tang Jie, 49; and Yan Junjie, 37—are known in the market as the "Four Great Kings of Domestic AI." Their AI companies collectively hold a valuation exceeding 1.1 trillion RMB. Coming from diverse backgrounds—quantitative trading, computer science, and artificial intelligence research—they are all top-tier academics. Each follows a distinct technological path: DeepSeek focuses on cost reduction, Kimi challenges leading U.S. models, Zhipu AI bets on foundational AGI research, and MiniMax expands into multimodal applications.
But what puts them on the same stage is the same fact: China’s large models have taken a leading position in the global AI race.
Over the past few years, the narrative around China’s AI development has largely been one of “catching up.” In early 2025, DeepSeek shocked the world with its low-cost models, marking the first “DeepSeek moment.” The arrival of Kimi K3 has been dubbed by industry insiders the “DeepSeek 2.0 moment”—as high-performance Chinese large models now emerge as competitors within the global narrative of artificial intelligence. OpenAI President Greg Brockman publicly acknowledged that estimates he has seen suggest China may be only four months behind the U.S. in model development.
What troubles Silicon Valley even more is the open-source model commonly adopted by Chinese companies. The six most popular models on OpenRouter all come from Chinese companies. Global developers can freely download, self-deploy, and flexibly modify them, significantly lowering the barrier to adopting AI technology. Le Monde of France noted that China’s mainstream large models generally follow an open-source approach, sharply contrasting with the closed-source, paid models of leading U.S. companies. Some analysts believe China’s AI industry has successfully turned its “chip shortage” into an “algorithmic efficiency advantage”—despite having far fewer GPUs than its American counterparts, it has developed exceptional engineering optimization capabilities.
Yet, behind all of this is an increasingly younger face.
At the 2026 World Artificial Intelligence Conference, many noticed a striking change: founders born in the 2000s and 1990s made up a very high proportion.
Zeng Guoyang, co-founder and CTO of FaceWall Intelligence, has been named “AI Personality of the Year” at CCTV’s “2026 China AI Gala.” Tianwu Technology’s protein design agent, “Xiao Wu,” has been selected as one of the event’s “treasures,” developed by 00s PhD student Tan Yang. In early 2026, Chen Boyuan, still ungraduated, founded Inverse Matrix and secured over $100 million in seed funding from Hillhouse Capital and Peking University-affiliated funds, with the company’s valuation now exceeding RMB 5 billion.
From Yang Zhilin to Chen Boyuan, a new, younger generation is rewriting the narrative of Chinese AI—following different paths and pursuing varied directions, but united by one common trait: they care little for existing rules or what others think.
(Source: People Magazine)
Employees of Moonshot AI say there are no departmental silos, and in some ways, there aren’t even departments. Yang Zhilin’s email signature consists of just four words: “Communicate directly.” If something requires someone else’s cooperation to complete, simply reach out to them directly—no approvals, no coordination, no meetings needed. More than ten employees have said they’re more comfortable interacting with AI because it’s more reliable and straightforward. Everyone is lively in group chats, yet unusually quiet in person. One employee used a single word to describe this atmosphere: “shy.”
Most brands want a story, but Kimi’s employees will remind visitors: don’t write about Pink Floyd, and don’t mention the piano by the office door.
Returning to Shantou Jinshan High School’s 2011 leaderboard, Yang Zhilin wrote these words for younger students: “Believe that things will fall into place. You can do anything you set your mind to.”
Looking back years later, this statement was not only his own epitaph, but also a prophecy for an entire generation. They didn’t take shortcuts or detours—they simply did what needed to be done, and the world naturally listened.
Author: seed.eth
Twitter: https://twitter.com/BitpushNewsCN
BitPush Telegram community: https://t.me/BitPushCommunity
BitPush Telegram subscription: https://t.me/bitpush