During the week Kimi K3 was released, the U.S. tech community collectively lamented why such an outstanding young person as Yang Zhilin didn't stay in the United States back then.
This topic has garnered over five million views on X. David Sacks, former White House AI lead and a close confidant of Trump, said he has switched from Claude to Kimi to handle large volumes of work, stating, “It’s just way more fun because it actually gets things done instead of lecturing you.” Meanwhile, veteran Silicon Valley venture capitalist Vinod Khosla attributes the issue to immigration policy, saying the U.S. is driving away top talent on its own.
The speculation grew larger, and Yang Zhilin’s PhD advisor, Salakhutdinov, publicly praised his student on social media while clarifying the reason Yang did not stay in the U.S.: “If he didn’t have the courage to try entrepreneurship, Yang Zhilin said he would regret it for the rest of his life.”
On this side of the ocean, another Guangdong native, Liang Wenheng, is also widely known in China’s AI community.
Liang Wenfeng is seven years older than Yang Zhilin, born in Wuchuan, Zhanjiang, a programmer with fifteen years of experience in quantitative trading, who has rarely given interviews and has no social media accounts; his colleagues summarize his personality as “having no hobbies other than programming.” After the release of his DeepSeek R1 in January 2025, NVIDIA lost nearly $600 billion in market value in a single day—a moment Silicon Valley called its “Sputnik moment.” While the world searched for him, he returned to his hometown to play soccer for a few days.
Two men from Guangdong—one born in Wuchuan, the other in Shantou—separated by the Leizhou Peninsula. Their companies competed in the same industry, following nearly mirror-image paths, with every divergence rooted in their inherent personalities.
A radio and a band
Liang Wenfeng was born in 1985 in Mililing Village, Tanba Town; both of his parents were teachers at the local elementary school. With few toys at home, the most important object of his childhood was a Feiyue-brand radio, which he repeatedly took apart and reassembled countless times.
This quiet child showed signs of being extraordinary from an early age. His junior high homeroom teacher remembered he wasn’t a bookworm, nor did he seem to work harder than others, yet he taught himself high school math by junior high and began reading university textbooks—“as if he could master every subject without spending much time.”
In the 2002 college entrance exam, he scored 806 points, becoming the top scorer in Zhanjiang. Photos of him receiving the award can still be found today: wearing a dark red short-sleeve shirt, a large red flower pinned to his chest, his expression stiff—as if pushed onto the stage by his teachers. That autumn, he enrolled in Zhejiang University’s Electronic Information Engineering program. But over the next twenty years, there was never another moment worth photographing in his life.
Seven years later in Shantou, another child from Guangdong had grown up. Yang Zhilin, born in 1992, ranked first in his class for all four years at Tsinghua University’s Department of Computer Science and authored over twenty papers. While such achievements aren’t unheard of at Tsinghua, what sets him apart is that, alongside these accomplishments, he formed a rock band called Splay—named after a data structure—with himself as the drummer.
He later explained: “At the time, I felt there was so much I wanted to express—both the pressures from real life and the absurdity of the broader environment.” They wrote a song about a fantasy of overnight success and wealth through entrepreneurship, “half out of empathy, and half as a reminder to ourselves not to become overly materialistic.”
Many years later, the band would return to the story in a way no one could have anticipated: Zhou Xinyu, a bandmate, became a co-founder of Moonlight Dark Side.
In front of the office door at Lunar Eclipse stands a white Yamaha digital piano, upon which lies Pink Floyd’s 1973 album, The Dark Side of the Moon. The company’s name is inspired by this album.
Taking apart a radio doesn’t require anyone to see it; drumming happens because something must be expressed. Every choice these two people made over the next twenty years grew almost entirely from this moment.
Rental apartments in Chengdu and the hallways of CMU
After graduating from Zhejiang University, Liang Wenfeng didn’t join a tech giant to earn a technical credential. Instead, he went to Chengdu, hid in a cheap apartment, and experimented with various algorithms, trying to equip traditional industries with AI—all of which failed. While the founding teams of China’s top quantitative hedge funds typically boast prestigious backgrounds from overseas hedge funds, Liang Wenfeng forged his path entirely on his own in that rented room.
In 2015, he co-founded QuantConnect with classmates from Zhejiang University. What followed was a series of actions that almost no one understood at the time: in 2019, he invested nearly 200 million yuan to build a custom cluster with 1,100 GPUs; in 2021, he added another billion yuan to stockpile approximately 10,000 A100s.
You don’t need so many GPUs for quantitative trading—Liang Wenheng himself admitted that only a few GPUs are sufficient for trading alone. Those who had interacted with him in the past recall seeing him hoard GPUs to train models and simply thought he was a tech nerd with a bad haircut burning through money.
But what he did was the exact opposite of what everyone understood: instead of using AI to reduce costs and increase efficiency in finance, he used finance to fund AI research. Most Chinese AI companies follow the sequence of securing funding first, then finding a product, and finally generating cash flow—but he reversed the order entirely: he first built a cash flow machine, then used it to buy the freedom to conduct research.
Yang Zhilin took a different path—one paved with flowers and mines.
After graduating from Tsinghua University, he went to CMU (Carnegie Mellon University) to pursue his Ph.D., during which he conducted research at AI labs at Google and Meta. In 2017, he devoted all his efforts to language models, later calling it “the only important problem.” During his Ph.D., he published two papers: one that taught AI to remember longer contexts, and another that outperformed Google’s own strongest model across 20 benchmarks, with a combined citation count of nearly 20,000.
Years later, Kimi gained widespread attention for its ability to process long text inputs—many assumed this was a temporary differentiator found in 2023, but it was actually a direction he had firmly committed to during his PhD, merely taking on a new form.
In 2016, while still pursuing his Ph.D., he co-founded Recurrent Intelligence, which focused on sales call analysis, with Sequoia and GSR Ventures as shareholders. This venture gave him an early glimpse of the roughness of technology implementation, but also planted a seed beneath his feet that would not ignite until eight years later.
He graduated with a Ph.D. in 2019; his advisor contacted an Apple executive who reported directly to Cook to inquire whether Yang Zhilin would be interested in joining Apple, even potentially in Apple China. However, Yang declined all Apple emails and offers from Silicon Valley, deciding to return to China.
At that time, Liang Wenheng was hoarding cards in Hangzhou, while Yang Zhilin was waiting for the wind in Beijing. Neither knew of the other’s existence, nor could they have imagined that these two names would come to represent today’s Chinese AI community.
A one-month window and a catfish
On November 30, 2022, ChatGPT launched, and the tech community in Silicon Valley collectively lost sleep. Yang Zhilin recalled that many of his friends felt anxious, experienced FOMO, and couldn’t sleep—many of them turned to starting their own businesses.
We began focusing on our Series A funding in February 2023, but by April, the window had essentially closed. Even if we had started in December 2022 or January 2023, there would have been no opportunity—back then, the pandemic was still ongoing, and everyone was overwhelmed. Yang Zhilin seized that one-month window, taking not a single day off.
In March 2023, Moonshot AI was founded by Tsinghua alumni Zhou Xinyu and Wu Yuxin. It reportedly raised $600 million in initial funding and assembled around 40 AI researchers within three months. This was followed by the steepest fundraising curve in China’s large model history: Sequoia and ZhenFund entered the scene, Alibaba led a $1 billion investment, with Tencent, Meituan, and Xiaohongshu following suit, pushing the valuation to $33 billion. Kimi became the first large model product widely used by many Chinese users, thanks to its ability to handle 200,000-character texts.
During that time, Yang Zhilin lived in a dual state. Externally, he spoke of the grandest narratives, estimating the probability of scaling laws failing as nearly zero, comparing entrepreneurship to driving toward an endless snow-capped mountain range, and describing the first year as building a rocket prototype and getting a glimpse of the fuel formula. Internally, he had to monitor the most mundane algorithms: when computing power was tight, a machine’s performance fluctuated from 260 one day to 340 the next, then dropped again a few days later—he had to decide daily whether to buy or rent, which channel to use, and constantly adjust. Both the certainty of a scientist and the shrewdness of a small business owner were embodied in this 31-year-old.
Everything capital provides comes with a price already marked. To sustain its growth curve, Kimi spent 220 million in October 2024 and another 200 million in November—burning through more in two months than the entire third quarter. The drummer who wrote a song satirizing overnight wealth became the industry’s most aggressive buyer of user acquisition. It’s not that he changed; it’s that a $3.3 billion valuation is making the decisions for him.
Liang Wenfeng entered the market in Hangzhou in a manner that seemed almost deliberately oppositional.
In 2023, DeepSeek spun off from Hangfang without accepting any external investment. The team consists of fewer than 140 people, with almost no returnees—mostly recent graduates and young professionals from domestic universities. There are no KPIs or hierarchical structures; if you have an idea, you can directly request resources and personnel.
A former employee recalled to The Washington Post that Liang Wenfeng would dive into the details of training strategies, reviewing papers and writing code alongside researchers—“he didn’t act like a boss at all, more like a geek.” He explained why he hired recent graduates instead of poaching seasoned industry experts: “Experienced people will instinctively tell you to do it this way, but inexperienced people will keep exploring.”
In May 2024, DeepSeek-V2 slashed its API pricing to one yuan per million tokens, forcing ByteDance, Alibaba, Baidu, and Tencent to follow suit. The entire industry assumed this was a carefully planned business strategy; his response was: “We didn’t intentionally set out to be a catfish—we just accidentally became one.”
Pricing is only slightly above cost—“no subsidies, no excessive profits.” While internet companies in price wars talk about market share, entry points, and network effects, he talks about cost accounting. Yet it is precisely this emotionless price reduction that is most devastating, as it directly pulls large model APIs out of high-margin narratives and into the pricing logic of infrastructure.
Yang Zhilin and Liang Wenhong—two entirely different business models—have somehow begun to be compared by the outside world.
Hovering Yang Zhilin
The bomb Yang Zhilin planted eight years ago detonated in November 2024.
Yang Zhilin's previous startup, CycleAI, had five original shareholders file an arbitration case in Hong Kong, accusing him of initiating fundraising for his new company before obtaining full shareholder waivers.
On December 5, Zhu Shuhu launched a scathing attack in his WeChat Moments, targeting Zhang Yutong: the former GSR Ventures partner had received an initial allocation of 9 million shares, representing 14% of the company, for free—exceeding the 9.5% allocated to Loop Intelligence as the “parent” entity. Zhu Shuhu’s proposed solution was nearly humiliating: apologize, return the shares, or sever all ties between the company and Zhang Yutong.
On December 6 at 9:40 PM, Yang Zhilin published a 1,300-word article. Instead of distancing himself, he made his position unequivocal: Zhang Yutong is a co-founder, and her equity represents compensation for years of future work; the formalities for her departure were approved with the signatures of every board member. Those close to the company relayed the internal stance: she and Moonshot AI are inseparable—she cannot be cut off.
Zhu Xiaohu publicly stated he completely doesn’t understand. In a purely commercial framework, there is indeed no solution—cutting ties is the only rational option. But in Yang Zhilin’s decision-making framework, there are other considerations. Salakhutdinov’s later clarification provided a fitting footnote: this is the kind of person who would regret not trying for the rest of their life; once they’ve made up their mind, they won’t turn back, even when the costs are already clear.
The real hammer falls over forty days later.
On January 20, 2025, R1 was released—free and open-source, with reasoning capabilities rivaling OpenAI’s o1. A team of over a hundred in Hangzhou turned global financial markets upside down in just one week. Carnegie researcher Matt Sheehan made an interesting remark: “DeepSeek was not the company China had preselected; its explosive rise even surprised China.”
Liang Wenfeng spent the Lunar New Year in his hometown in Wuchuan. On the afternoon of January 27, he played a football match with his junior high school classmates in the village. The entrance to the village was crowded with tourists taking photos, while the main event was unfolding on the pitch.
For Yang Zhilin, this was a double blow. With arbitration still ongoing, R1 directly declared the death of his strategy over the past year: users acquired through paid acquisition meant nothing against a free and superior competitor. Public opinion turned against him; one article titled “Yang Zhilin: The Floating Idealism of a 90s Generation” captured the sentiment. “Floating” meant being disconnected from reality. When he wrote songs, he feared becoming opportunistic—now the world said he was both opportunistic and a failure.
In early 2025, Moonlight had funds on its books but little influence.
A comeback against gravity
Next year will be a crucial one for Yang Zhilin.
Yang Zhilin almost entirely rejected his own actions over the past year. He halted advertising spend, cut redundant businesses, and refocused on foundational models, shifting to open source. In a conversation with Geek Park, he said that an organization’s inertia tends to drive it to do more and more—“We must resist this gravity.”
It sounds easy to say, but doing it means admitting the strategy was wrong, firing the people he hired, and bowing to the very opponent who nearly destroyed him. Most 33-year-old founders can’t cross this threshold. But Yang Zhilin crossed it with remarkable decisiveness—perhaps because open source and long-termism were his original settings; closed-source, paid-growth tactics were merely the clothing imposed by capital, and now he’s simply taken them off.
In July 2025, the trillion-parameter K2 was open-sourced. In November, K2 Thinking outperformed GPT-5 on several of the most challenging agent benchmarks, prompting Hugging Face co-founder Thomas Wolf to ask on Twitter: Is this another DeepSeek moment?
That night after launch, Yang Zhilin, along with Zhou Xinyu and Wu Yuxin, hosted an AMA on Reddit, answering 21 questions. They clarified that the $4.6 million training cost was not an official figure and acknowledged that their GPU count lagged behind that of their U.S. counterparts, but added, “We’ve pushed every single card to its absolute limit.” When asked about OpenAI’s spending, Zhou Xinyu replied calmly: “We don’t know—only Sam knows. We have our own pace.”
On one's own rhythm. In 2024, Moonshot couldn't say these five words—its rhythm was dictated by investors and ROI-driven advertising. It was Liang Wenheng who returned these five words to Yang Zhilin. R1 proved that open-source combined with algorithmic efficiency works in China, effectively serving as a pitch to Yang’s own board. The very company that nearly killed him is now the one he must thank for saving his life.
The market's returns have been equally straightforward. On the last day of 2025, Moonshot officially announced a $500 million Series C round, with Alibaba, Tencent, and Wang Huiwen all increasing their investments, pushing the valuation to $4.3 billion and cash reserves above $10 billion. Less than 20 days after the launch of K2.5, revenue surpassed the entire year of 2025, with monthly personal subscription orders rising more than 80-fold month-over-month, propelling it into the top ten globally on Stripe’s leaderboard. Then came K3 this week.
Seven years ago, that PhD student who said, “If you don’t try, you’ll regret it for life,” has now made the U.S. tech industry question everything and become an example used to challenge the White House.
The Hermit's Bill
But reality never only hits one side of the face. After 2025, it will be Liang Wenheng’s turn to pay the bill.
He had no life, but his employees did. Luo Fuli, Wang Bingxuan, Wei Haoran, and Ruan Chong—these names were prominent core team members within DeepSeek, and starting in 2025, many of them departed, with several taking on leadership roles in business operations at other companies.
A harsh reality circulating among peers: if colleagues at a similar level can earn so much by leaving, why shouldn’t I? A utopia of research without hierarchies, no KPIs, and no talk of money relies on members’ loyalty to the problems themselves. But R1 has raised everyone’s market value tenfold—loyalty now faces its first priced rival.
Money is also having problems. Huanfang’s assets under management have shrunk from their peak to just over 20 billion yuan, like a battery powering love that’s now leaking. An unimaginable scenario has emerged: the person who once said, “There are no financing plans in the short term,” is now meeting with investors.
The initial minimum investment offer was 5 billion, later reduced to 1.5 billion. Yet throughout the negotiations, what he repeatedly emphasized was not valuation or equity stake, but the same condition: no poaching DeepSeek’s talent, and no encouraging them to leave and start their own ventures. What began as a funding discussion ended up becoming a non-poaching agreement. He was willing to give up equity, but not the atmosphere of that laboratory.
By 2026, the two lines had reached a situation no one anticipated.
Yang Zhilin is cutting marketing spend, squeezing every bit of performance out of each H800, and staying true to his own rhythm—becoming increasingly like Liang Wenfeng. Liang Wenfeng, on the other hand, is meeting with VCs, worrying about retaining talent, and for the first time, diverting his attention from the lab to deal with these mundane organizational issues.
Back then, one was hoarding water, the other was chasing the tide; now, those hoarding water have discovered that reservoirs can leak, and those chasing the tide have finally seen their own come.
Two Guangdong natives are reshaping China’s AI landscape.
This is the most vibrant time for AI entrepreneurship in China, with two names brought to the forefront: Liang Wenheng and Yang Zhilin.
They don’t resemble the founders familiar to Chinese internet users over the past decade—there are no iconic quote posters, no legendary dinner stories, and no strong desire to cast themselves as legendary entrepreneurs. Yet their position places them closer than most previous entrepreneurs to the convergence of money, power, and the spirit of the times.
What sets Liang Wenfeng apart is that he appears to have almost no “life.” Public reports portray him as someone consumed by work: his story revolves almost entirely around quantitative investment, building his own computing clusters, and DeepSeek—with virtually no mention of family, consumption, hobbies, or social life; instead, it is filled with models, architectures, computing power, organization, and originality.
What sets Yang Zhilin apart is that he appears to have no "escape route." Since the founding of Moonshot AI, Kimi has rapidly become one of China’s most watched AI applications, with pressure mounting simultaneously from fundraising, valuation, user growth, model iteration, commercialization, and shareholder arbitration. The faster it moves, the less time it has to stop and explain.
This is where the two people are most alike—and most different.
Liang Wenfeng appears as a technological idealist emerging from the depths of capital markets. He first demonstrated at Huanfang Quantitative that AI could directly alter the flow of money, then redirected this capability toward large models. DeepSeek’s story was never driven by fundraising, but rather by the capital, computing power, and engineering culture accumulated by Huanfang. The outside world later remembers R1, V3, low-cost training, and the open-source shockwave that reverberated through the U.S. market—but what’s more critical is that Liang Wenfeng defined the core issue early on as the gap between originality and imitation, not as “how to monetize Chinese large models.”
It made him seem anti-business, yet extremely business-oriented.
What’s anti-commercial is that, in public interviews, he repeatedly downplays short-term monetization and refuses to frame DeepSeek as an internet story about grabbing users or market share. What’s hyper-commercial is that every step he takes targets the most fundamental cost structures 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 significantly lower cost, it doesn’t just change a product ranking—it reshapes the entire industry’s imagination around capital expenditure.
This is why the rise of DeepSeek triggered such a strong reaction in global markets. It wasn’t just another Chinese chatbot—it reminded investors that the most expensive logic chain in America’s AI narrative over the past two years may not be as unshakable as once thought. Larger models, more GPUs, and higher capital expenditures do not necessarily equate to an insurmountable moat.
Yang Zhilin faced a different fate.
Moonshot has been in the spotlight since its inception. The founders boast an impressive background: undergraduate degrees from Tsinghua University, PhDs from Carnegie Mellon University, and participation in pivotal research such as Transformer-XL and XLNet. Founded in 2023, Moonshot differentiated itself with long-context capabilities, quickly becoming an AI product noticeable even to everyday users. Unlike DeepSeek, which first gained recognition within the tech community as a research organization, Moonshot entered the consumer market, capital markets, and industry narratives much earlier.
This gave Yang Zhilin a significant advantage, as well as a heavy burden.
The advantage is that Kimi has a product mindset. Many people first seriously used a domestic AI not because they read a technical report, but because Kimi can read long documents, organize materials, and handle workflows. For AI to move from the lab to the office, it needs an accessible entry point that ordinary users can remember—Kimi once secured that position.
The burden is that once a product’s mental model is established, it must be continuously nurtured. Users will wait for stronger models, investors for higher revenues, the team for greater option value, and competitors for you to make a mistake. The more Moonshot raises, the higher its valuation, and the less likely Yang Zhilin is to return to a quiet life as a researcher.
By 2026, this pressure becomes clearer. Public reports indicate that Moonshot AI completed approximately $2 billion in new funding by May 2026, pushing its post-money valuation beyond $20 billion; the company’s official website has also placed Kimi K2 and its code and agent capabilities at the core. The capital market has not offered applause, but rather a bill: you must prove you are not just “the AI company best at building products,” but also that you can maintain both technological and commercial leadership amid intense competition from DeepSeek, Alibaba, ByteDance, MiniMax, and Zhipu.
More complicatedly, Yang Zhilin is still shadowed by his previous startup venture. In 2024, media outlets such as the South China Morning Post reported that Yang Zhilin, founder of Moonshot AI, and his co-founder Zhang Yutao were subjected to arbitration in Hong Kong by certain former investors of Cyclica. Yang’s side stated that he had completed all necessary procedures to leave Cyclica and launch a new venture, while the investors offered a different account. At the heart of this dispute is not merely a personal conflict between founders, but a sharper question arising from China’s AI startup boom: Where exactly do the boundaries lie between a new company’s soaring value and its ties to the old company, former shareholders, previous teams, legacy intellectual property, and new funding?
Yang Zhilin cannot merely portray himself as a brilliant researcher. The scale of funding, commercial revenue, product iteration, IPO expectations, and legal disputes will require him to become a more complete and ruthless CEO. A researcher can prove themselves with papers; a CEO must prove themselves through their organization. Papers can be signed, but an organization cannot run on signatures alone.
The more humble a person is, the more the era amplifies them. The more a person tries to move forward, the more the past catches up to them.
In this round of China’s AI story, the winner may not be the best storyteller—but rather those who can endure three things simultaneously: technological uncertainty, the erosion of capital’s patience, and the personal cost of being both mythologized and judged.
It is not actually known whether Liang Wenheng has a life outside. Whether Yang Zhilin has a backup plan has not yet been decided.
But at least for now, neither of them has much room to return to ordinary life.
律动 BlockBeats