Liang Wenfeng has no life, Yang Zhilin has no way out

By Jia Liu, BeatZ

This week, as Kimi K3 was released, the US tech community collectively expressed regret: why didn't such an outstanding young person like Yang Zhilin stay in the United States back then?

Kimi K3 surpasses all other models in frontend capabilities.

This topic garnered over five million views on X. David Sacks, Trump's former head of AI at the White House and a close confidant of Trump, stated that he had switched from Claude to Kimi for handling large amounts of work, saying, "It's just much more fun because it gets things done directly instead of lecturing you." Meanwhile, veteran Silicon Valley venture capitalist Vinod Khosla blamed immigration policies, claiming that the US is driving away outstanding talent with its own hands.

Speculations continued to grow. Yang Zhilin's PhD supervisor, Salakhutdinov, publicly showed off his favorite student on social media while clarifying the reasons for not staying in the US: "If you don't even have the courage to try starting a business, Yang Zhilin said he would regret it for the rest of his life." On this side of the ocean, there is another name from Guangdong that is also well-known in China's AI circle: Liang Wenfeng.

Liang Wenfeng is seven years older than Yang Zhilin. Born in Wuchuan, Zhanjiang, he is a programmer who has engaged in quantitative trading for fifteen years. He has rarely given interviews, has no social media accounts, and his colleagues' only summary of his personality is "he has no hobbies other than programming." After the release of his DeepSeek R1 in January 2025, Nvidia lost nearly $600 billion in market capitalization in a single day, an event Silicon Valley dubbed the "Sputnik moment." While the whole world was looking for him, he hid away in his hometown and played football for a few days.

Two people from Guangdong, one born in Wuchuan and the other in Shantou, separated by the Leizhou Peninsula. Their companies ran on the same track, following almost perfectly mirrored trajectories, yet every divergence actually stemmed from their true personalities.

A Radio and a Band

Liang Wenfeng was born in 1985 in Mili Ling Village, Qinba Town. Both of his parents were primary school teachers in the town. There weren't many toys at home; the most important object of his childhood was a Feiyue brand radio, which he took apart and put back together repeatedly countless times.

This quiet child showed differences early on. His junior high school homeroom teacher remembered that he wasn't a bookworm, nor did he necessarily study harder than others, but he had self-taught high school mathematics by junior high and started reading university textbooks. "It seemed as if he could learn every subject without spending much time."

In the 2002 college entrance examination, he scored 806, becoming the top scorer in Zhanjiang City. Photos of him receiving the award can still be found today: wearing a jujube-red short-sleeved shirt with a large red flower pinned to his chest, his expression stiff, clearly pushed onto the stage by his teacher. That autumn, he entered Zhejiang University to study Electronic Information Engineering. But for the next twenty years, no occasion worthy of photography appeared in his life. Seven years later in Shantou, another Guangdong boy grew up. Yang Zhilin, born in 1992, ranked first in his grade for four years at Tsinghua University's Department of Computer Science, publishing over twenty papers. This wasn't particularly rare at Tsinghua; what was rare was that outside of these academic achievements, he formed a rock band named Splay, derived from a data structure, playing the drums himself. He later explained: "At that time, I felt there was a lot I wanted to express, including pressure from reality and a sense of absurdity in the broader environment." They wrote a song about a daydream of achieving entrepreneurial success and getting rich overnight. "Half out of empathy, half to remind myself not to become too utilitarian." Many years later, this band would return to the story in a way no one could have predicted: the bandmate Zhou Xinyu became a co-founder of Moonshot AI.

Outside the office of Moonshot AI stood a white Yamaha electric piano, topped with Pink Floyd's 1973 album *The Dark Side of the Moon*. The company's name came from this album.

The Dark Side of the Moon Album

One dismantled a radio without needing anyone to see; the other played drums because there was something that had to be expressed. Almost every choice these two made over the next twenty years grew from here.

A Rental Room in Chengdu and a Corridor at CMU

After graduating from Zhejiang University, Liang Wenfeng didn't go to a big tech company to receive a technical business card. Instead, he went to Chengdu, hiding in a cheap rental apartment to try various algorithms, wanting to equip traditional industries with AI, but failed completely. The founding teams of China's top quantitative funds mostly had gilded resumes from overseas hedge funds, whereas Liang Wenfeng figured it all out himself in a rental room. In 2015, he co-founded High-Flyer Quant with classmates from Zhejiang University. This was followed by a series of moves that almost no one understood at the time: in 2019, he invested nearly 200 million yuan to build his own cluster with 1,100 GPUs; in 2021, he added another 1 billion yuan, hoarding approximately 10,000 A100s. Quantitative trading doesn't require that many cards, Liang Wenfeng admitted himself. For trading alone, a small number of cards would suffice. Someone who had interacted with him early on recalled seeing him hoard cards to train models, thinking only that this was a tech geek with a bad hairstyle burning money. But what he was doing was exactly opposite to everyone's understanding. He wasn't using AI to reduce costs and increase efficiency in finance; rather, he was using finance to fund AI research. The order for most Chinese AI companies was first to raise financing, then find products, then find cash flow. But he reversed the entire sequence: first create a cash-generating machine, then use it to buy freedom for research. Yang Zhilin took another path, a road paved with flowers and landmines. After graduating from Tsinghua, he went to CMU (Carnegie Mellon University) for his PhD, conducting research at both Google and Meta's AI labs. In 2017, he bet all his energy on language models, later calling it "the only important question." During his doctoral studies, he published two papers: one taught AI to remember longer contexts, and the other defeated Google's strongest model in 20 tests, accumulating nearly 20,000 citations combined. Years later, Kimi gained popularity for its ability to input long texts. Many thought this was a differentiated selling point found temporarily in 2023, but it was actually a shape-shifted version of the direction he had identified during his PhD. In 2016, while still a PhD student, he participated in founding Circulate Intelligence, which focused on sales call analysis, with Sequoia Capital and Vertex Ventures as shareholders. This entrepreneurial experience allowed him to witness the roughness of technology implementation early on, but it also buried a landmine under his feet, which wouldn't be detonated until eight years later. In 2019, upon graduating with his PhD, his supervisor connected him with an Apple executive who reported directly to Tim Cook, asking if Yang Zhilin was willing to join Apple, possibly even in Apple's China region. However, Yang Zhilin rejected Apple's email and offers from Silicon Valley, deciding to return to China. At that time, Liang Wenfeng was hoarding cards in Hangzhou, and Yang Zhilin was waiting for the wind in Beijing. Neither knew of the other's existence, nor did they know that these two names would come to represent China's AI circle today.

A One-Month Window and a Catfish

On November 30, 2022, ChatGPT launched, causing the Silicon Valley tech community collective insomnia. Yang Zhilin recalled that many friends around him were anxious, experiencing FOMO, and couldn't sleep; many turned to entrepreneurship. "Starting in February 2023, we concentrated on the first round of financing. If delayed to April, there would basically be no opportunity. But doing it in December 2022 or January also offered no chance due to the pandemic; everyone hadn't reacted yet." Yang Zhilin seized this one-month window, resting not a single day. In March 2023, Moonshot AI was established, with Tsinghua alumni Zhou Xinyu and Wu Yuxin as co-founders. It reportedly raised $60 million in seed funding and gathered about 40 AI researchers within three months. Then came the steepest curve in the history of Chinese large model financing: Sequoia and ZhenFund entered, Alibaba led with a $1 billion investment, followed by Tencent, Meituan, and Xiaohongshu, pushing the total to $3.3 billion. Kimi became the first large model product used frequently by many Chinese people, thanks to its 200,000-word long text capability. During that period, Yang Zhilin lived in a dual state. Outwardly, he spoke grand narratives, estimating the probability of scaling laws failing as close to zero, comparing entrepreneurship to driving towards endless snow-capped mountains, and calling the first year building a rocket prototype and touching a bit of fuel formula. Inwardly, he had to monitor the most trivial algorithms: when computing power was tight, a machine cost 260 one day, 340 the next, and dropped back a few days later. Buy or rent? Which channel? Following and modifying daily. Half scientist's certainty, half small boss's shrewdness, were embodied in this 31-year-old young man. But everything capital gave had already been priced. To sustain the growth curve, in October 2024, Kimi spent 220 million in a single month, and another 200 million in November, burning through more than the entire third quarter in two months. The drummer who wrote songs satirizing overnight wealth became the founder buying traffic most aggressively in the industry. It wasn't that he changed; it was the $3.3 billion valuation making decisions for him. Liang Wenfeng entered the field in Hangzhou in a manner almost deliberately contrary. In 2023, DeepSeek spun off from High-Flyer, accepting no external investment. The team had fewer than 140 people, almost no returnees, all fresh graduates from domestic universities and young people who had graduated only a few years ago. No KPIs, no hierarchy, ideas could directly adjust cards and personnel. A former employee recalled to *The Washington Post* that Liang Wenfeng would dive into the details of training strategies, reading papers and writing code alongside researchers, "completely unlike a boss, more like a geek." He explained why he hired fresh graduates instead of poaching industry experts: "Experienced people will tell you unhesitatingly that you should do it this way, but inexperienced people will explore repeatedly." In May 2024, DeepSeek-V2 compressed API prices to one yuan per million tokens, forcing ByteDance, Alibaba, Baidu, and Tencent to follow. The entire industry thought this was a premeditated business war, but his response was: "We didn't intend to become a catfish; we just accidentally became one." Pricing was merely slight profit above cost, "not subsidizing money, nor earning exorbitant profits." Internet people talked about market share, entry points, and network effects in price wars; he talked about cost accounting. Yet precisely this emotionless price drop was the most deadly, dragging large model APIs directly from high-margin narratives into infrastructure pricing logic. Yang Zhilin and Liang Wenfeng, two completely different business models, began to be compared by outsiders from some unknown point.

The Floating Yang Zhilin

The landmine Yang Zhilin buried eight years ago exploded in November 2024. For Yang Zhilin's previous startup, Circulate Intelligence, five old shareholders filed for arbitration in Hong Kong, accusing him of initiating financing for the new company before obtaining full shareholder waivers. On December 5th, Zhu Xiaohu attacked on WeChat Moments, focusing fire on Zhang Yutong: the former Vertex Ventures partner received an initial 14% stake, or 9 million shares, for free at Moonshot AI, exceeding the 9.5% allocated to Circulate Intelligence as the "parent body." Zhu Xiaohu's solution was nearly humiliating: apologize, return shares, or cut ties between the company and Zhang Yutong. At 9:40 PM on December 6th, Yang Zhilin issued a 1,300-word long article. He didn't cut ties; instead, he closed the door: Zhang Yutong is a co-founder, and her shares are consideration for future years of work. The procedures for leaving Circulate Intelligence had signatures from every director. People close to the company relayed internal attitudes: she and Moonshot AI are now one entity; they cannot be separated. Zhu Xiaohu publicly stated he completely didn't understand. In a purely commercial coordinate system, indeed, there is no solution; cutting ties is the only rational option. But in Yang Zhilin's coordinate system for decision-making, there were other considerations. Salakhutdinov's later clarification provided a footnote: this is someone who "would regret it for a lifetime if he didn't try," someone who doesn't look back once committed, even if the cost is laid bare on the table. The real heavy hammer fell forty-some days later. On January 20, 2025, R1 was released. Free, open-source, with reasoning capabilities approaching OpenAI's o1. A team of over a hundred people in Hangzhou turned the global capital market upside down in a week. Carnegie researcher Matt Sheehan said something interesting: DeepSeek was not the company China had pre-selected; its viral success even surprised China. Meanwhile, Liang Wenfeng was celebrating New Year in his hometown of Wuchuan. On the afternoon of January 27th, he played football with junior high school classmates in the village. Tourists checking in crowded the village entrance, while the main character was on the pitch. For Yang Zhilin, this was a double kill. The arbitration was unresolved, and R1 directly sentenced his past year's route to death: users bought through advertising were insignificant in front of a free and stronger competitor. Public opinion shifted its gun barrel; an article's title read "Yang Zhilin, a 90s Idealist's Suspension." Suspension means feet not touching the ground. When he wrote songs, he worried about becoming utilitarian; now the whole world says he is both utilitarian and a failure. In early 2025, Moonshot AI still had money in the bank, but very little 话语权 (discourse power).

The Comeback Against Gravity

The next year was a critical one for Yang Zhilin. Yang Zhilin almost completely negated himself from the past year. Stopped advertising, cut redundant businesses, contracted to basic models, and switched to open source. In an interview with Geek Park, he said organizational inertia wants to do more and more things, "we must fight against this gravity." This sounds light, but doing it means admitting the route was wrong, firing people he hired, and bowing to a rival that nearly killed him. Most 33-year-old founders cannot pass this hurdle. Yang Zhilin passed it exceptionally decisively, perhaps because open source and long-termism were originally his factory settings; closed source and ad-buying were clothes put on him by capital, now simply taken off. In July 2025, the trillion-parameter K2 was open-sourced. In November, K2 Thinking pressed GPT-5 beneath itself on several of the hardest Agent benchmarks. Hugging Face co-founder Thomas Wolf asked on Twitter: Is this another DeepSeek moment? In the early hours after the release, Yang Zhilin, along with Zhou Xinyu and Wu Yuxin, held an AMA on Reddit, answering 21 questions consecutively. He clarified that the $4.6 million training cost was not an official figure, admitting that the number of GPUs was inferior to American peers, "but we squeezed the performance of every card to the extreme." Someone asked how he viewed OpenAI's burning of money; Zhou Xinyu answered relaxedly: "We don't know either; only Sam knows. 'We have our own rhythm.'" Our own rhythm. In 2024, Moonshot AI couldn't say these five words; at that time, its rhythm was the investors' rhythm, the advertising ROI rhythm. What returned these five words to Yang Zhilin was precisely Liang Wenfeng. R1 proved that open source plus algorithmic efficiency works in China, equivalent to giving Yang Zhilin a roadshow to his own board. The person nearly killed by DeepSeek had to thank it for his life. Market returns were also direct. On the last day of 2025, Moonshot AI announced a $500 million Series C round, with Alibaba, Tencent, and Wang Huiwen all adding investments, valuing the company at $4.3 billion, with over ten billion yuan in cash on hand. Less than 20 days after the release of K2.5, revenue exceeded that of all of 2025, personal subscription orders increased eighty-fold month-on-month, breaking into the top ten of Stripe's global list. Then came this week's K3. The PhD student who said "you'll regret it for a lifetime if you don't try" seven years ago now makes the US tech community doubt life, becoming an example used to rattle the White House.

The Hermit's Bill

But reality never only slaps one side of the face. After 2025, it was Liang Wenfeng's turn to pay the bill. He had no life, but his employees did. Luo Fuli, Wang Bingxuan, Wei Haoran, Ruan Chong—these names were prominent core backbone members inside DeepSeek. Starting in 2025, they left successively, with many taking direct business leadership roles elsewhere. The popular saying is heart-wrenching: When colleagues at similar levels jump out and get so much, why can't I? A research utopia with no hierarchy, no KPIs, and no talk of money relies on members' loyalty to the problem itself. But R1 raised everyone's market value tenfold; loyalty faced a priced opponent for the first time. Money was also becoming a problem. High-Flyer's assets under management shrank from its peak to over 20 billion yuan; the battery powering love was leaking. Thus appeared a scene previously unimaginable: the person who said "no financing plans in the short term" began meeting investors. The opening bid was 5 billion yuan per transaction, later lowered to 1.5 billion. But throughout the negotiation, what he repeatedly emphasized was not valuation or equity ratio, but the same condition: do not poach DeepSeek people, do not incite them to start businesses elsewhere. A financing deal turned into a non-poaching agreement. He could give up shares, but he could not give up the atmosphere of that laboratory. As the two lines reached 2026, a situation emerged that no one expected. Yang Zhilin was cutting ads, squeezing the performance of every H800 card, talking about his own rhythm, becoming more and more like Liang Wenfeng. Liang Wenfeng was meeting VCs, worrying about retaining staff, and for the first time distracted by secular issues like organization, forced to step out of the lab. One accumulated water, the other rode the tide; now the accumulator finds the reservoir leaks, and the tide-rider finally waits for his own tide.

Two Guangdong People Rewrite China's AI Table

This is the liveliest time for China's AI entrepreneurship, with two names pushed to the forefront: Liang Wenfeng and Yang Zhilin. They don't quite resemble the founders familiar to the Chinese internet over the past decade. There are no golden quote posters, no banquet legends, nor a strong desire to mold themselves into legendary entrepreneurs. But the positions they stand in are closer to the intersection of money, power, and the zeitgeist than most entrepreneurs in the past. Liang Wenfeng's uniqueness lies in the fact that he appears to have almost no "life." In public reports, he looks more like a person swallowed by work: from quantitative investment, to building computing clusters, to DeepSeek, his narrative rarely includes family, consumption, hobbies, or social life; almost everything is filled with models, architectures, computing power, organization, and originality. Yang Zhilin's uniqueness lies in the fact that he appears to have no "way out." After Moonshot AI was founded, Kimi quickly became one of China's most watched AI applications. Financing, valuation, user growth, model iteration, commercialization, and old shareholder arbitrations all pressed down on this company simultaneously. The faster it runs, the less time it can stop to explain. This is where the two are most alike and most unlike. Liang Wenfeng seems like a technological idealist emerging from the depths of the capital market. He first proved in High-Flyer Quant that AI could directly change the flow of money, then turned this capability toward large models. DeepSeek's story initially wasn't driven by financing but by the funds, computing power, and engineering culture accumulated by High-Flyer. What the outside world later remembered was R1, V3, low-cost training, and open-source shocks to the US market, but more crucially: Liang Wenfeng defined the problem early on as the gap between "originality and imitation," not "how Chinese large model applications make money." This makes him appear anti-commercial yet extremely commercial. Anti-commercial because he repeatedly downplayed short-term monetization in public interviews, unwilling to tell DeepSeek's story as one of grabbing users and market share like an internet story. Extremely commercial because every step he took pointed directly to the underlying cost structure of the AI industry: training efficiency, inference costs, model architecture, talent density, and chip constraints. If a model can approach top-tier capabilities at a lower cost, what it changes is not just a product leaderboard, but the entire industry's imagination regarding capital expenditure. This is why the global market reaction was so intense when DeepSeek went viral. It wasn't just another Chinese chatbot; it reminded investors that the most expensive logical chain in the US AI narrative over the past two years might not be as solid as imagined. Larger models, more GPUs, higher capital expenditures do not necessarily equal an uncatchable moat. Yang Zhilin faced a different destiny. Moonshot AI stood in the spotlight from birth. The founder's resume was sufficiently impressive: Tsinghua undergraduate, Carnegie Mellon PhD, participation in important research like Transformer-XL and XLNet; the company was founded in 2023, and Kimi differentiated itself with long contexts, quickly becoming an AI product perceptible to ordinary users. Unlike DeepSeek, which was first discovered by the tech circle as a research organization, it entered mass products, capital markets, and industrial narratives earlier. This gave Yang Zhilin huge advantages but also huge burdens. The advantage is that Kimi has product mindshare. Many people's first serious use of domestic AI wasn't because they understood a technical report, but because Kimi could read long texts, organize materials, and handle workflows. For AI to move from the lab to the office, an entry point memorable to ordinary people was needed; Kimi once grabbed this position. The burden is that once product mindshare is established, it must be continuously fed. Users wait for stronger models, investors wait for higher income, teams wait for larger option values, and competitors wait for you to make mistakes. The more Moonshot AI raises, the higher the valuation, the less likely Yang Zhilin is to retreat to a quiet researcher position. By 2026, this pressure became clearer. Public reports showed that Moonshot AI completed about $2 billion in new financing in May 2026, post-money valuation breaking $20 billion; the company website also placed Kimi K2 alongside code and Agent capabilities in core positions. What the capital market gave it wasn't applause, but a bill: you must prove you are not just "the largest model company best at making products," but also prove you can maintain dual leadership in technology and commerce amidst attacks from DeepSeek, Alibaba, ByteDance, MiniMax, Zhipu, etc. More troublesome is the shadow left behind by Yang Zhilin's old startup projects. In 2024, media outlets like the *South China Morning Post* reported that Moonshot AI founder Yang Zhilin and co-founder Zhang Yutao were sued for arbitration by some former investors of Circulate Intelligence in Hong Kong; Yang Zhilin's side claimed necessary procedures for leaving Circulate Intelligence and starting the business were completed, while investors presented a different account. The core of this controversy is not just a personal dispute of a founder, but a sharper issue in China's AI entrepreneurship wave: when a new company's value skyrockets, where exactly are the boundaries between the old company, old shareholders, old teams, old intellectual property, and new financing? Yang Zhilin cannot just present himself as a genius researcher. Financing scale, commercial revenue, product iteration, IPO expectations, and legal disputes all demand he become a more complete, colder CEO. A researcher can prove himself with papers; a CEO must prove himself with organization. Papers can be signed; organizations cannot run solely on signatures. The more low-key a person is, the more amplified by the era they become. The more forward-looking a person wants to be, the more pursued by the past they are. China's AI story may not ultimately be won by the best storyteller. It is more likely to be won by those who can endure three things simultaneously: technological uncertainty, the patient consumption of capital, and the cost of the founder being mythologized and judged. Whether Liang Wenfeng has a life, the outside world actually doesn't know. Whether Yang Zhilin has a way out, the final answer hasn't arrived yet. But at least now, neither has much space to return to the state of ordinary people.