A post-90s professional from GL Ventures has become Liang Wenfeng's CFO.
On September 14, a personnel-related piece of news spread in the venture capital circle. According to multiple independent sources, the chief financial officer of DeepSeek is about to take office, and the candidate closest to the final selection is Yan Wentao, partner of GL Ventures, who is currently going through the resignation procedure. Although this position has not been fully confirmed, many people in the industry have taken this matter for granted. On the same day, Reuters also released a report stating that DeepSeek plans to hire this GL Ventures partner as the company's first CFO, labeling him a "dealmaker", a person who is good at closing transactions.
Yan Wentao was born in 1991 and is 35 years old this year. He has been working on primary market investments for a long time. One week earlier, on September 9, the news that DeepSeek hired CITIC Securities to prepare for its Sci-Tech Innovation Board listing was just confirmed by insiders. The two incidents happened one after another, making it difficult for the public not to associate them with each other.
A Position Left Vacant for One and a Half Years
DeepSeek has been looking for a CFO for a very long time. In February 2025, the company posted positions including chief financial officer and vice president of finance in its recruitment information, with job descriptions mentioning "optimizing capital structure" and "promoting strategic financing". At that time, R1 had just become popular globally, and the market's first reaction was that DeepSeek was about to start financing.
That did not happen. For more than a year afterwards, DeepSeek still relied on Hyperscaler to provide financial support, and the CFO position remained vacant. Until June this year, in the finance team recruitment published on DeepSeek's official website, the CFO position was still listed, and this position has been posted for more than a year.
The position has been vacant for so long, but there are quite a lot of people wanting to take it. Sources revealed that heads of many leading investment institutions such as 5Y Capital, HSG, and Dragon Capital have all communicated with Liang Wenfeng about this position. These names are all closing guests at any investment summit.
The industry's interpretation of this is somewhat delicate. In the view of many people, most of these "job applications" are just excuses, and the real intention is to take the opportunity to sit down and talk with Liang Wenfeng, get a clear picture of DeepSeek's situation, and see if there are any investment opportunities by the way.
After all, for a long time, Liang Wenfeng has been one of the most difficult people to make an appointment with in China's AI circle. Being able to talk to him face to face once is a scarce resource in itself. An industry insider put it bluntly that the attractiveness of this position comes half from DeepSeek's position in the industry, and half from the capitalization process it is about to launch. The person who takes this position is standing in the middle of the biggest capital story in China's AI sector at the moment.
After going through a lot of twists and turns, it is Yan Wentao who finally gets the position. According to people close to DeepSeek, Liang Wenfeng hopes that this CFO is a post-90s. This condition sounds a bit random, but it has actually screened out a large number of candidates with deeper qualifications and older ages.
This requirement is not surprising for DeepSeek. The company's team has always been known for being young, and reports say that more than 70% of its core research team members are under 30 years old.
Yan Wentao graduated from Fudan University, joined GL Ventures in 2020, and later became one of the representatives of post-90s investors at GL Ventures, and was promoted to partner not long ago. The projects he has invested in include ByteDance, J&T, as well as Zhipu AI and MiniMax. The latter two are exactly DeepSeek's peers in the large model track, and both have been listed on the Hong Kong Stock Exchange. From the perspective of an investor, he has fully witnessed how large model companies complete financing and go public. Reuters also specifically mentioned that GL Ventures is not currently a shareholder of DeepSeek, so he does not need to bear the interests of his former employer when he joins the company.
At the same time, Yan Wentao is not the precedent of leaving an investment institution to work at a large model company. Earlier, Xue Zizhao, a GL Ventures investor focusing on MiniMax, joined MiniMax in 2023 and served as vice president of investment and financing and joint company secretary. In addition, Zhang Yutong, managing partner of GSR Ventures, joined Moonshot AI in April 2024 and served as president, responsible for overall strategy and commercialization including financing.
Not only in the large model circle, there are many such examples in other industries.
A long time ago, for investors, growing to a certain stage and then starting their own business to set up a new fund was the goal in life. But now, more and more investors seem to prefer to start a business or take up senior management positions in enterprises.
From No Financing to Two Rounds of Hundred-Billion-Level Financing
Looking back a year ago, DeepSeek was almost never associated with the word "financing". Liang Wenfeng publicly stated that what DeepSeek lacked was high-end chips, not capital. The turning point came in the spring of this year. In April, The Information reported that DeepSeek was seeking its first external financing, with a scale of at least 3 billion US dollars and a valuation of no less than 100 billion US dollars. This figure was quickly far exceeded. In mid-June, DeepSeek's first round of financing was completed, with an overall scale of about 510 billion yuan, and a post-investment valuation of nearly 4 trillion yuan, making it the largest first round of financing in the history of China's large model sector.
In the list of investors, Liang Wenfeng himself contributed about 200 billion yuan, making him the largest single investor. Tencent contributed about 100 billion yuan, the CATL system contributed about 50 billion yuan, NetEase, JD.com, IDG Capital and other institutions also participated, and the National Artificial Intelligence Industry Investment Fund contributed nearly 10 billion yuan. Internet giants, industrial capital and established US dollar funds all squeezed into this round.
What is more thought-provoking than the numbers are the terms. According to insiders, except for the National Artificial Intelligence Industry Investment Fund which directly injected capital, the funds from other external investors had to enter a limited partnership managed by Liang Wenfeng first, and then hold shares of DeepSeek indirectly. These investors have no voting rights, cannot get board seats, and have to accept a five-year lock-up period. What they get in return is mainly the priority to obtain financial information and the preemptive subscription right for subsequent rounds.
Liang Wenfeng's team even requested to verify who the LPs behind each contributing fund are. Before the financing, Liang Wenfeng first increased his direct personal shareholding from 1% to 34% through capital increase. Together with his indirect shareholding, his final beneficial shareholding reached 84.29%, and his voting rights reached 100%.
Some comments described this round of financing as buying a back-row seat ticket: investors can contribute money, but the steering wheel is always in Liang Wenfeng's hands. For other companies, such conditions would probably be very difficult to negotiate. Even so, people who want to get on the bus are still queuing up, and even giants like Tencent have to sit in the back row.
The second round came faster, but with more twists and turns. In mid-July, DeepSeek launched the second round of financing, but soon pressed the pause button suddenly, and some investors on the waiting list were notified that the signing plan was postponed. According to media reports, one of the reasons was that Liang Wenfeng was dissatisfied with a "transcript of the investor exchange meeting" circulated on the Internet. The exchange meeting lasted for nearly 4 hours, and its content was later widely forwarded in the industry. In early August, the financing restarted, with a planned fundraising of 500 billion yuan and a pre-investment valuation of about 5 trillion yuan. The signing was originally scheduled for late August, and both parties hoped to keep it as low-key as possible this time. Up to now, there is no public information about the final completion of this round of financing.
This pause has brought the differences between the two sides to the table. In the future, it will probably be this new CFO who passes messages between the two sides and grasps the proper sense of propriety. This task is far more difficult than simply negotiating a financing deal.
Behind the capital issue, there are also personnel issues. In the past, DeepSeek delayed financing for a long time, and the options held by employees had no external valuation as a reference, so it was hard to tell how much they were worth.
At the same time, poaching from large factories has never stopped: Luo Fuli joined Xiaomi, Wang Bingxuan joined Tencent, and Guo Daya, a core researcher of R1, was confirmed to join ByteDance's Seed team in April this year. Of course, there are also different voices. *Caijing* sorted out the author lists of 27 papers published by DeepSeek in more than two years, and believed that the core members of this team are actually quite stable, and the team is still expanding. In any case, once the equity has a clear price, retaining talents will always become easier.
The Sci-Tech Innovation Board Is Waiting for a Large Model Enterprise
On June 17, Wu Qing, Chairman of the China Securities Regulatory Commission, announced at the Lujiazui Forum that the scope of application of the fifth set of listing standards for the Sci-Tech Innovation Board has been extended to the field of artificial intelligence large models. In the afternoon of the same day, the Shanghai Stock Exchange released the supporting audit guidelines, with only a few hours apart. This day is regarded by many people as the starting point for large model companies to land on the A-share market.
This guideline has a total of 15 articles, and its core meaning is to support large model enterprises that have not yet formed a certain scale of revenue but have obvious technical advantages to list on the Sci-Tech Innovation Board. The threshold is also clearly stated, one of which requires that when applying for listing, the enterprise has at least one large model product that has been launched and realized large-scale application. Checking this requirement against DeepSeek, it hardly needs additional preparation. After R1, its influence among global developers is obvious to all, and the V4 series also maintains the rhythm of open source and iteration. The AI companies on the Sci-Tech Innovation Board before are mainly chip enterprises such as Moore Threads and MetaX, and for companies that purely develop large models, this path has been clearly opened for the first time.
DeepSeek followed up very quickly. According to reports from Bloomberg, as early as July, DeepSeek cooperated with accounting firms and investment bank consultants to prepare financial materials required for listing. At that time, the public's attention was still focused on the second round of financing, and the preparation for listing had already been carried out in secret.
On September 9, Reuters cited people familiar with the matter as saying that DeepSeek has hired CITIC Securities to prepare for its Sci-Tech Innovation Board IPO, hoping to launch the listing process within this year. Relevant media later confirmed this news from insiders: CITIC Securities has entered the site to conduct due diligence, but the two sides have not yet signed a formal listing counseling agreement, and the listing time, fundraising scale and issuance valuation have not been determined. Another report mentioned that DeepSeek's revenue in the first 7 months of this year was about 475 million yuan, almost 10 times that of the whole year of 2025. On the same day, DeepSeek also lowered the API price of its Flash series of models, and some analysis believes that this is to exchange price reduction for calling volume.
According to Reuters, the CFO in a company is in charge of investor relations, financial internal control, capital allocation, and information disclosure required for listing. These are exactly the things that DeepSeek has touched the least in the past. For a company that is used to conducting research behind closed doors, this is almost a brand new set of rules.
This company is famous for being silent. After Liang Wenfeng was interviewed by *Waves* in July 2024, he did not speak publicly for nearly two years, and the researchers rarely appeared in public. When V4 was released, in the face of various public discussions, DeepSeek's only response was "Do not be tempted by praise, do not be afraid of slander, act according to the principle, and correct yourself properly". But once it becomes a listed company, regular financial reports, temporary announcements, and performance briefings will come one after another. How to explain the relationship between open source and commercialization is also an unavoidable chapter in the prospectus. The fact that the model is widely used does not directly equal revenue, and someone has to explain this clearly to secondary market investors.
Peers have taken earlier actions on this matter. Xue Zizhao, vice president of investment and financing and joint company secretary of MiniMax, also came from GL Ventures, and is responsible for capital market operation and financial affairs in the company. MiniMax has gone all the way from the primary market to being listed on the Hong Kong Stock Exchange, and there are always special persons in charge of this part of work.
On the side of Moonshot AI, Zhang Yutong, former managing partner of GSR Ventures, joined the company as a co-founder and is responsible for financing and other affairs, and publicly appeared as president around the end of last year. Yang Zhilin once commented on her that she not only understands business and strategy, but also has rich experience in investment and financing, and is willing to step out of the comfort zone to take risks. Now Moonshot AI has secretly submitted its application for listing in Hong Kong. According to reports from *South China Morning Post*, it is also considering landing on the Sci-Tech Innovation Board afterwards. Zhipu AI and MiniMax have been listed on the Hong Kong Stock Exchange one after another, taking the lead. The path of entering large model companies from investment institutions to manage capital has been taken by others long ago, and Yan Wentao is not the first one.
At this 4-hour investor exchange meeting this year, Liang Wenfeng said that the biggest core interest of DeepSeek is to keep the team stable. DeepSeek, which has exercised restraint for three years, is now going to push open the door of the capital market. What is behind the door will soon be seen by this post-90s CFO in person.
This article is from the WeChat Official Account "Rongzhong Finance" (ID: thecapital), written by Wang Tao, and published with authorization from 36Kr.