Large corporations will never understand Liang Wenfeng.
Internet giants will never understand Liang Wenfeng, just as old players find it hard to keep up with new card games. They think it's a game where money gets you a seat, but in fact, they may not even be at the same table.
Late on the night of May 8th, a piece of news that was both surprising and reasonable came - DeepSeek is seeking to raise up to 50 billion yuan (about $7.35 billion) in its first - round financing, which will set a new record for single - round financing of Chinese AI companies.
Just before the news was officially announced, a more meaningful detail emerged. According to Baijing Laboratory, Alibaba, which had previously been high - profile in the competition, was in a deadlock in the negotiation of this round of financing.
In the past few weeks, around the rarest ticket in the Chinese AI field - DeepSeek, Tencent, Alibaba and a group of state - owned institutions have engaged in a silent game. According to Chao News, Tencent plans to invest 6 billion yuan, accounting for about 2% of the equity; while founder Liang Wenfeng plans to personally invest up to 20 billion yuan, accounting for 40% of the total fundraising. However, Alibaba's participation remains uncertain.
In addition, the post - investment valuation of DeepSeek will reach 350 billion yuan. In the secondary market, Zhipu currently has a market value of HK$411.5 billion (about 357.4 billion yuan), and MiniMax has a market value of HK$232.8 billion (about 202.2 billion yuan). In the primary market, the latest post - investment valuation of Kimi is $20 billion (about 136 billion yuan). In late February, Caijing reported that Jieyue Xingchen plans to submit its prospectus on the Hong Kong Stock Exchange before June 30th this year, with an expected cornerstone pricing of about $10 billion (68 billion yuan).
So far, the total valuation/market value of the top five new Chinese AI companies has reached 1.1 trillion yuan.
Behind these five companies stand different Internet giants. According to an incomplete statistics by Phoenix Tech, in the multiple rounds of financing of the above - mentioned AI companies, many Internet giants and mid - sized companies such as Tencent, Alibaba, Meituan, Ant Group, Xiaohongshu, Xiaomi, and Mihoyo have appeared. Among them, Tencent and Alibaba have made the most investments. Tencent has invested in almost all the top five AI companies, Alibaba has participated in the investment of Kimi, Zhipu, and MiniMax, and Meituan has recently heavily invested in Kimi. Xiaohongshu has participated in the investment of Kimi and MiniMax.
However, when it comes to the most important ticket - DeepSeek, the setbacks of Internet giants are perhaps the most obvious. The misalignment is not about price, but about the underlying logic. Among all the AI stars, DeepSeek is the closest to original innovation and most advocates anti - KPI and anti - OKR, while the latter is precisely the organizational culture that Internet giants are most proud of and have thrived on in the past few years.
While Internet giants are used to understanding everything from their own perspective, Liang Wenfeng has never intended to fit into that framework.
01
Alibaba and DeepSeek: Several Encounters and Denials
The story starts at a rather dramatic point: DeepSeek and Alibaba are both located in the same city - Hangzhou.
In February 2025, there were rumors in the market that Alibaba would invest $1 billion to subscribe for 10% of DeepSeek's equity at a valuation of $10 billion. At that time, this local alliance was regarded as a perfect match: on one hand, there was an e - commerce giant eager to establish its position in the AI large - model track; on the other hand, there was a star startup with a solid technical foundation but yet to start commercialization, lacking money, computing power, and application scenarios. Logically, this deal could meet the needs of both sides.
Moreover, Alibaba has shown an extremely determined attitude in its AI strategy. From the infrastructure investment plan of 380 billion yuan to the "golden triangle" setup of T-head Semiconductor, Alibaba Cloud, and Tongyi large - model, Alibaba has almost become one of the only two global AI full - stack players.
However, on the night when the news spread, a relevant person in charge of Alibaba Group posted a WeChat Moments message, politely clarifying: "As a company in Hangzhou, China, we applaud DeepSeek, but the information that Alibaba is investing in DeepSeek is false."
This is not the first time Alibaba has shown goodwill, nor the first time it has been rejected. According to a previous report by The Wall Street Journal, since March 2025, Liang Wenfeng has continuously rejected investment proposals from companies including Tencent and Alibaba, and even said bluntly that he is "not in a hurry to raise funds." Liang Wenfeng has set an iron rule: not to accept external financing, not to dilute equity, and not to be tied to anyone's commercialization schedule.
In the eyes of Internet giants, this is almost a form of arrogance; but in Liang Wenfeng's logic, this is precisely the last line of defense for a researcher to maintain the purity of scientific research.
In the past, Alibaba had a clear technical cooperation plan, using the relevant computing power requirements of Alibaba Cloud to offset the investment amount and taking a stake at a discounted price with strategic resources, ultimately achieving a double - binding effect.
However, for a team that has established its status through technological open - source and architectural innovation, the word "being bound" may be an unacceptable bottom line.
In the eyes of Internet giants, it seems like the best deal in the world: they provide money, resources, and channels, and the invested company just focuses on technology. But in the view of DeepSeek, which advocates original innovation, this is a path - dependence and a wrong path that may stifle innovation.
That's why Internet giants lacking an innovation gene may never understand why someone doesn't want to sit at the table they've carefully set up, because they firmly believe that the end of all startup stories should be a business empire.
02
It's Never About the Money
The essence of the negotiation deadlock has never been about money.
According to reports, Tencent proposed to subscribe for up to 20% of DeepSeek's shares in this round of financing, but was flatly rejected by DeepSeek. Alibaba's conditions were even more demanding: it not only required a considerable equity ratio but also tried to embed strategic decision - making power in the board of directors. For Liang Wenfeng, these are all touches on the red line.
A crucial time point is worth noting: on April 27, 2026, just before the financing news started to spread intensively, DeepSeek quietly completed an adjustment of its equity structure. Industrial and commercial information shows that the company's registered capital increased from 10 million yuan to 15 million yuan. Founder Liang Wenfeng increased his direct stake from 1% to 34% through direct capital increase, and the shareholding ratio of the original major shareholder, Ningbo Chengen Enterprise Management Consulting Partnership, was diluted to 66%. After the adjustment, Liang Wenfeng controls about 84.29% of the company's equity through direct and indirect shareholding. More directly, Liang Wenfeng plans to personally invest up to 20 billion yuan in this round of financing, accounting for 40% of the total fundraising.
First, hold tight the control with his own money, and then open the door to welcome strategic investors. This sequence of operations sends a clear signal: Liang Wenfeng welcomes capital, but will never tolerate capital taking over, as this would be the greatest harm to DeepSeek as an innovative organization.
Another AI company is a typical example. Phoenix Tech learned that in early 2025, after a popular AI company received investment from an Internet giant, it was required by the latter to provide DAU data. This is the most observable indicator in the Internet era, but in the AI era, excessive pursuit of short - term DAU may not only have no positive effect but also erode the innovation motivation of a technology - research team. "The founder starts to deal with shareholders and DAU indicators, which will definitely distort R & D," an investor in the primary market told Phoenix Tech.
However, Liang Wenfeng's insistence on independence makes Internet giants, who are used to "I invest in you, and you have to listen to me," feel at a loss. In the cognitive framework of Internet giants, investment is essentially an equivalent exchange of resources: they provide money and scenarios, and in return, they get equity and decision - making power. This is the market rule.
Liang Wenfeng's confidence comes from Magic Square Quantitative, a quantitative giant with a management scale of over 70 billion yuan. In 2025, it achieved an annualized return of 56.55%, and the performance fees alone may bring in more than $700 million in cash flow in a year. More importantly, Liang Wenfeng holds a majority stake in Magic Square, and the profits of Magic Square can be fully converted into DeepSeek's R & D budget without being diluted by any external shareholders. This is why, since its establishment, DeepSeek has always had the strongest confidence in financing negotiations among all Chinese AI startups.
In the investment history of Internet giants, this is almost an unfamiliar experience. They are used to suppressing with scale, enticing with traffic, and coercing with the ecosystem, but when facing a self - sufficient opponent, all these methods fail. That's why Internet giants will never understand Liang Wenfeng. In their eyes, control is a bargaining chip on the negotiation table, while in Liang Wenfeng's eyes, control is an untradeable belief.
03
How Can Old Players Fit into the New Card Game?
Just as Internet giants were deeply involved in the game with DeepSeek, a real "super variable" entered the scene.
On May 6th, it was reported that the National Integrated Circuit Industry Investment Fund (the "National Big Fund") is in talks with DeepSeek to lead its first - round financing, and the post - investment valuation is expected to be around $45 billion. In just two weeks, the outside world's valuation of DeepSeek has doubled from $20 billion.
The entry of the National Big Fund brings much more than just a doubled valuation. The first and second phases of the Big Fund focused on the chip hardware as the only entry point. The third phase extended its reach systematically to all industrial chain links such as large - models, embodied intelligence, and AI security by setting up a special AI industry investment fund. This contact with DeepSeek is called the first direct investment case in an AI large - model company, with a clear ecological goal behind it.
Many investors told Shanghai Securities News: "This project is different from other startup projects. Although the current commercialization efforts are limited, investors still have high hopes for its development potential. Now, it's definitely the investors chasing after Liang Wenfeng to see which one he will finally choose." Some analysts further pointed out that in the final investor lineup, "state - owned capital will definitely play a very important role."
This also means that when national - level capital enters the market collectively, the bidding ability of Internet giants is completely crushed. It's not about who has more money, but about who can provide the strategic resources that DeepSeek really needs: industrial chain integration ability and support for computing power infrastructure.
These structural forces far exceed the value that an Internet giant can provide. For Liang Wenfeng, choosing state - owned capital instead of Internet giants is not simply "bowing to capital," but choosing a "strategic partner" that can maximize the preservation of governance independence. In the verification by Shanghai Securities News, the industry generally believes that the end of the financing is likely to be a three - layer structure of "led by the Big Fund, with multiple Internet giants participating with small amounts, and Liang Wenfeng continuing to make a large personal investment," which can ensure the supply of funds and prevent any party from gaining overwhelming decision - making power.
However, in the new - era card game, old players are still struggling to find their positions.
Needless to say about the first - tier Internet giants led by Alibaba, Tencent, and ByteDance. Even mid - sized companies are now actively making arrangements.
Just before the May Day holiday this year, Xiaohongshu announced a large - scale organizational upgrade in an internal letter. The core actions are twofold: first, establish a first - level department Dots centered around AI to build a complete technical system from model R & D, infrastructure, engineering to products; second, establish an "Enterprise Intelligence Department" at the organizational level to re - establish the foundation for the "AI - era organization" from four dimensions: intelligence, talent, data, and resources. Xiaohongshu's AI construction is ultimately entrusted to Conan, who has a technical background.
Meituan started to reflect earlier. On March 13, 2026, Meituan held a management communication meeting with a scale of 2,000 people. CEO Wang Xing made a strong statement, saying that "mobile Internet and the Internet are like roses and peonies, while AI and the Internet are like monkeys and flowers - the magnitude and influence are much greater." Then, at the end of April, Meituan quietly launched an invitation - only test of a large - scale model with trillions of parameters. In May, it publicly announced a strategic investment in Kimi, but this battle has just begun.
Judging from the differences in the investment in DeepSeek, this is the first collision between old players and the new card game.
Internet giants will never understand Liang Wenfeng, just as old players find it hard to keep up with new card games. They think it's a game where money gets you a seat, but in fact, they may not even be at the same table.
This article is from the WeChat official account "Phoenix Tech", author: Dong Yuqing. Republished by 36Kr with permission.