GuShiio.com Article source: Tech Planet
Image source: Generated by AI
DeepSeek, the AI model, is popular around the world with its three major characteristics of ease of use, open source, and free. It has become the fastest APP in history to exceed 30 million daily active users. Nowadays, all walks of life are being transformed by DeepSeek.
Before it exploded, the outside world knew very little about such a company. It rarely appeared in media reports, headhunting, and peer research, and it never stirred up a splash in the market like the “AI Six Little Dragons”.
Its founder, Liang Wenfeng, is also a low-key person. There are only a few words of information about him on the Internet, and the number of public speeches that can be found is very small.
Even practitioners in the big model industry don’t pay special attention when talking about DeepSeek. “I read it in headhunting recommendations in 2023, but I didn’t pay any attention at all at that time,” said a big model practitioner. “In the past few years, the focus has been on the large model ‘Six Little Dragons’, and research on DeepSeek has been limited to past products.” said one investment manager.
But now it is completely different. Colleagues and investment institutions are all looking for Liang Wenfeng.
“To be precise, we began to pay special attention to this product after DeepSeek’s open source reasoning model DeepSeek-R1,” said the practitioner of the above-mentioned big model. After DeepSeek-R1 was open source, many industry insiders re-examined DeepSeek’s technical strength.
At the same time, investment institutions have taken action and tried every means to reach more people related to DeepSeek. “Of course we want to meet Liang Wenfeng,” said an investment manager at a private equity firm.
But just like before the explosion, DeepSeek and Liang Wenfeng are still hidden in this wave of craze, and no one can easily find their traces.
Nearly a hundred people visit DeepSeek’s office a day
“Strict security” is the most intuitive feeling of an employee working in a company in Huijin International Building on the first day of returning to work after the Spring Festival. In early February, when she stepped into the building, she was keenly aware that the security situation was completely different and was much stricter than before. “Later, I heard from security personnel that DeepSeek and we worked in the same building.”
In the following week, as DeepSeek’s popularity soared, Huijin International Building became a hot “check-in point”. A building staff member said that at most, nearly a hundred people came to visit in one day.
However, most of these “visitors” only left behind their backs in disgrace. “Although DeepSeek’s specific office address can be found online, security personnel will not disclose the specific floor it is located on, let alone expect to see their employees,” the employee said.
Among the visitors looking for DeepSeek and Liang Wenfeng, investment institutions who have long smelled market opportunities are also making efforts.
An investment manager told Tech Planet that it is not difficult to find the contact information of DeepSeek staff and even founder Liang Wenfeng, but the difficulty lies in communicating with them.”Through the connections I have accumulated in the AI industry in the past, I have added several DeepSeek employees, but I didn’t get much feedback.”
Another investment manager also said that he was trying to contact DeepSeek during the Chinese New Year, but there was really no result.
For a time, those who had interacted with DeepSeek on different occasions and in different ways, no matter how insignificant the connection, became the “rich mine” that everyone competed to explore.
A headhunter who had sent recruitment requests for DeepSeek said,”In the hottest days of DeepSeek, I was asked questions about DeepSeek by many people, some asking about recruitment needs, and some asking about the company’s situation.”
And a practitioner who was once invited by DeepSeek to go for an interview was also added to WeChat by many “strangers” after sharing his experience with friends around him.
DeepSeek has been surrounded by almost fanatical attention, but it has never signaled that it is looking forward to communication.
Shortly after resuming work after the Spring Festival, DeepSeek declined all interview requests through external feedback from Huijin International Building Property, saying that it would focus on its own research and development. Since the explosion, Liang Wenfeng has only briefly appeared in the public eye.
But despite this, discussions around DeepSeek and Liang Wenfeng’s past experiences are still hot.
“One month after the holiday, our work still focused on DeepSeek research.” The investment manager said that he had read almost all the public information of Liang Wenfeng and DeepSeek in the past.
On social platforms, discussions about DeepSeek and Liang Wenfeng also continue to emerge. From technology nerds discussing the architectural optimization of the DeepSeek model to business people analyzing its potential market valuation, everyone’s enthusiasm remains undiminished.
As the outside world explores DeepSeek and Liang Wenfeng deeper and deeper, this previously unknown company seems to have become more mysterious, and everyone wants to solve the many mysteries surrounding it.
Investment institutions and DeepSeek: A misplaced game amid the financing boom
Compared with the curious gaze of the outside world, investment institutions in the primary market smell money. They ultimately want to squeeze into the ranks of DeepSeek investors.
Even Zhu Xiaohu, managing partner of Jinshajiang Venture Capital, has reversed his attitude towards AI investment because of DeepSeek. Zhu Xiaohu once said he would no longer invest in big model companies, but DeepSeek changed his skepticism about General Artificial Intelligence (AGI).
In an interview with Tencent News, he said that if DeepSeek opens up financing, he will definitely participate in the investment because it is of great significance to “witness the birth of human AGI.” He even said that price is no longer important, the key is to participate in it.
Another investor also said that if they could really participate in DeepSeek’s investment, who would bother with price and share?
However, unlike the eagerness shown by investors in the primary market, DeepSeek has never had a record of external financing in the past, and now it has not signaled that financing is needed.
According to an enterprise survey, Liang Wenfeng directly and indirectly holds 84.29% of the shares of Hangzhou Deep Search Company behind DeepSeek. He has absolute voting rights on DeepSeek.
After the explosion, many investors approached DeepSeek’s team in charge of IR (Investor Relations). One of the PE (private equity) investors said that they have been in contact with DeepSeek in the hope of gaining share of the investment, but DeepSeek made it clear that it is not open to financing at present.
In Liang Wenfeng’s personal resume, money does not seem to be the core factor hindering him from starting a business. The quantified profits of Liang Wenfeng’s quantitative hedge fund Magic Square support his investment in DeepSeek.
Magic Square Quantification was established in 2015. The original main body was Zhejiang Jiuzhang Assets. In 2016, Ningbo Magic Square Quantification was established. Data from the China Foundation Association shows that as of July 31, 2024, the asset management scale of Zhejiang Jiuzhang is more than 10 billion yuan; as of April 26, 2024, the quantified management scale of Ningbo Magic Square has also reached more than 10 billion yuan.
His strong financial strength gave Liang Wenfeng a relatively loose financial environment when he joined the development of DeepSeek.
However, DeepSeek’s development model relying on its own funds and its growing popularity in the market have also discouraged many investors.
“On the one hand, the other party has made it clear many times that it will not raise funds. On the other hand, based on the current popularity of DeepSeek, its valuation must exceed the pricing we can afford and cannot afford it,”said a PE investor.
According to Bloomberg analysts ‘forecasts for DeepSeek’s median valuation, DeepSeek’s valuation may reach US$30 billion. And this price is close to the total valuation of the current “AI Six Little Dragons”.
However, Liang Wenfeng did not consider financing. In an exclusive interview with Liang Wenfeng by “Dark Waves” in 2023, he revealed that he had taken the initiative to communicate with multiple investors and tried to introduce funds to help DeepSeek develop. However, the negotiations highlighted the contradiction-venture capital (VC) institutions are restricted by the exit mechanism and are eager to commercialize the project in order to quickly withdraw funds and obtain benefits.
The core of the development blueprint planned by Liang Wenfeng for DeepSeek is “research first”. This conceptual difference has become a huge obstacle between DeepSeek and VC. Even today, this contradiction cannot be easily resolved.
Waiting for the next DeepSeek
On March 1, the DeepSeek team’s Zhihu account released the first news, announcing key information such as optimization technical details and cost and profit margins of the DeepSeek V3/R1 reasoning system for the first time.
Based on theoretical calculations, DeepSeek claims that its AI model’s reasoning services have a “cost-profit margin” of 545%. This means that for every $1 spent in costs, DeepSeek earns $5.45 in profit.
However, 545% is only a theoretical data. In fact,”night discounts”,”V3 low-price strategy”, and “free web/App entry” are all factors that affect this data.
The theoretical figure of 545% is like the current best situation for DeepSeek. No one knows whether it will last and for how long.
A large-scale model industry practitioner told Tech Planet that the challenges facing DeepSeek have already arrived.”DeepSeek shocked the market this time, which actually represents a victory for open source.”
In his view, on the one hand, it is very difficult for open source models to continue to launch products that can compete with closed-source models. On the other hand, open source model companies cannot avoid the topic of commercialization and monetization. If they want to continue, they must find monetization. way.
Not long ago, Wu Bingjian, a partner of Heart Capital, posted a sigh on social media: He said that last year he went to ask DeepSeek people,”Why are your models so good?” The other party replied that because our boss reads papers, writes code, and does recruitment, wherever time is spent, efficiency is easy to be achieved.
Facing the next round of challenges, the DeepSeek team cannot slack off at all. “Their team must work hard to explore various technical issues in a big model, otherwise they may fall behind,” the above-mentioned practitioner said.
For investors who have missed DeepSeek, now is the time for them to re-examine the key factors that have influenced investment decisions in the past.
An investor reflected on, for example, looking at the team. Liang Wenfeng’s team is a team with a quantitative fund background and is doing an AI open source model, which is not in line with past experience. For another example, looking at business models, venture capital tends to have a clear business route, but even today, it is difficult to see DeepSeek’s future business direction.
“The explosion of DeepSeek has given us an opportunity to rethink.” In his view, DeepSeek has broken the conventional understanding of past investment and cultivated stronger forward-looking and inclusiveness, so that we can not miss the next DeepSeek needs. ability.
But no matter what, DeepSeek has broken the giant’s narrative in the AI space, and now investors are eager to seize it before the next DeepSeek appears.
