GuShiio.com Article source: Smart Things
Image source: Generated by AI
DeepSeek, which shocked the world overnight during the Spring Festival, made Hangzhou become the country’s top seller. When the public’s eyes are all focused on Hangzhou, few people know that DeepSeek’s earliest registered company is located in Beijing, and it was not until Hangzhou was established two months later.
Now, DeepSeek’s Beijing office is crowded with visitors every day. With many questions, Zhixi visited the office of DeepSeek’s Rongke Information Center in Haidian District, Beijing, and communicated with many people who knew about DeepSeek and had worked in DeepSeek. They discovered many unknown stories behind this world-renowned company.
Although DeepSeek’s main body is located in Hangzhou, its subsidiary in Beijing is domineering in Haidian.An entire floor of Rongke Information Center, covers an area of up to2100 square metersleft and right. It is understood that DeepSeek currently has 160 employees. Even if they all work in Beijing, if 70% of the office space is calculated, the per capita office space is calculated.At least 9 square meters。Such a spacious office environment is very rare for startups.
Not only are the hardware conditions “luxurious”, but DeepSeek’s working atmosphere is also extremely relaxed. In today’s competitive and tight paced workplace environment,Recruitment is mainly for fresh graduates, and employees have no KPIs or check-ins, and are mainly interest-oriented.。
Behind the loose office environment, Liang Wenfeng is quite generous in terms of salary. The magic square quantifies that the annual salary of a senior data scientist is twice the competition. DeepSeek’s annual salary in recruiting positions is alreadyover one million。
Zhidongxi sorted out DeepSeek employee graduate schools based on public information and found that DeepSeek has more students graduating from Peking University than Tsinghua University. Some netizens believe that this is because Peking University students would rather go to small companies than choose Tsinghua AI companies when looking for internships.
Previously known as the “Six Little Dragons of Hangzhou” exploded in the circle. Why is the main R & D team located in Beijing? What is the difference between the development of Beijing and Hangzhou companies? How does DeepSeek’s success relate to its selection and employment standards? How is this almost “ideal” working model achieved? We try to find answers to these questions by disclosing information and interviewing relevant people.
01.
DeepSeek opens the key location for a turning point of the times,
Beijing is the starting point
When sorting out the development context of DeepSeek, a mysterious clue hidden in the long river of time gradually surfaced. It can be summarized in eight words: Beijing blooms, Hangzhou results.
By sorting out the timeline of the establishment of this company, we found that DeepSeek’s main body was first established in 2023.Established in Beijing in May, called Beijing Deep Quest Artificial Intelligence Basic Technology Research Co., Ltd.(hereinafter referred to as Beijing Deep Quest), is two months earlier than the current main company of DeepSeek, Hangzhou Deep Quest Artificial Intelligence Basic Technology Research Co., Ltd.(hereinafter referred to as Hangzhou Deep Quest).Hangzhou Deep Search was established in July 2023。
Interestingly, in the early days of its establishment, Hangzhou Deep Search was a wholly-owned subsidiary of Beijing Deep Search.
But what is more interesting is that in August 2023, the roles of the two companies reversed.
Beijing Deep Seeking to transfer all equity to Ningbo Cheng ‘en Enterprise Management Consulting Partnership (Limited Partnership) and Liang Wenfeng, and Hangzhou Deep Seeking to increase its registered capital from 1 million yuan to 10 million yuan.
▲ Beijing deeply seeks equity change information (left), Hangzhou deeply seeks equity change information (right)
Since then, the current layout of DeepSeek has been formed.Hangzhou Deep Search is the main company of DeepSeek, and Beijing Deep Search is a subsidiary company。
Since the Spring Festival, the Rongke Information Center, where Beijing conducts in-depth research, has become a popular place for players from all walks of life to check in.
From a geographical point of view, the surrounding areas of Rongke Information CenterUniversities, well-known companiesnumerous. More than 40 well-known universities and research institutes such as Tsinghua University, Peking University, and China Academy of Sciences are gathered around the park. There are also technology companies with the world’s top market value such as Apple, Microsoft, AMD, and Nvidia, as well as a number of domestic technology startups.
▲ Around Rongke Information Center (Source: Baidu Map)
In May 2024, DeepSeek settled in the Rongke Information Center and occupied an entire floor with an area of about 2100 square meters. Liu Yonghao, chairman of New Hope Group, recently revealed that during an exchange with DeepSeek founder Liang Wenfeng, he learned that DeepSeek currently has 160 employees. In other words, even if all employees work in Beijing, if the office space is calculated at 70%, the per capita office space can reach at least 9 square meters.
Although DeepSeek’s main body is located in Hangzhou, our research found thatMost of DeepSeek’s R & D personnel and R & D progress are completed in Beijing。
Among the employees displayed by Sky Eye, DeepSeek’s overall size is not large, and in comparison, Hangzhou’s main company is smaller than Beijing.
Based on the Tianyan big data model, combined with the operating data of enterprises in different industries, such as the company’s total assets, operating profits, personnel size, etc.(among which listed/bond issuing companies use consolidated reporting standards), the L (large), M (medium), S (small) and XS (micro) enterprise scale classification system is formed by comprehensive calculation. Beijing is medium-sized and Hangzhou is micro-sized.
There is a significant difference in the size of the two companies. This shows that the company’s business and R & D focus may be in Beijing.
Interestingly, DeepSeek currently has multiple domain names, of which deepseek.cn was launched in Beijing on September 4, 2023.ICP filing (Internet information service filing), and deepseek.com completed its filing in Hangzhou on January 23, 2024.
For large model companies, algorithm filing and large model filing are crucial, which are two necessary conditions for companies to open their models to the public.
Currently,The DeepSeek algorithm filing has been completed in Beijing and Hangzhou in April 2024 and June 2024 respectively。
Large model filing will be approved in Beijing in May 2024。This may be related to the earliest batch of Beijing City’s layout in the national large-scale model filing. In August 2023, Beijing City has disclosed the first batch of large-scale models that have passed the filing.
Another interesting thing is thatDeepSeek’s two trademark agencies are both Beijing companies。
It can be seen that the early development of DeepSeek was inseparable from Beijing.
02.
The unique way to win: fresh graduates carry the lead
Liang Wenfeng himself has no other hobbies except programming
While top talents gather in the “encirclement”, why did DeepSeek take the lead in rushing out and changing the competitive landscape of the entire big model industry? There are indications that perhaps DeepSeek’s unique employee system and Liang Wenfeng’s unique personality charm contain some kind of winning code.
It is worth mentioning that DeepSeek, located in a gathering place of prestigious schools, has a high number of employees from well-known universities in Beijing such as Qingbei.
In May 2023, DeepSeek founder Liang Wenfeng mentioned in an interview with Dark Waves that DeepSeek’s core technical positions are basically people who have graduated one or two years ago. According to a student who had interned at DeepSeek, when DeepSeek recruited,They are mainly fresh graduates, with a master’s degree and a working background related to the big language model。
According to incomplete statistics from Zhidong, among the authors of the papers that DeepSeek has published areMore than half of the employees still study or graduate from well-known universities such as Peking University, Tsinghua University, Beijing Normal University, Beijing University of Posts and Telecommunications, and Beijing University of Aeronautics and Astronautics。In addition, a small number of its employees have graduated from well-known overseas universities such as the Massachusetts Institute of Technology and Carnegie Mellon University, and most of them have interdisciplinary backgrounds.
For example, the first author of a latest paper released by DeepSeek last month is Yuan Jingyang, a master’s degree student at Peking University. When he was an intern at DeepSeek, he was one of the main authors of the DeepSeek-V3 technical report and also participated in the DeepSeek-R1 project; the key innovative MLA architecture of DeepSeek-V2 was developed by graduates from Peking University and Beijing University of Posts and Telecommunications; Luo Fuli, a post-95 AI genius girl named DeepSeek, who is widely rumored to have earned tens of millions of yuan, also graduated from Peking University…
▲DeepSeek’s latest paper
The current public recruitment information also shows this trend.
According to the public information on the recruitment platform Boss Direct Recruitment, DeepSeek currently has a total of 34 recruitment positions, involving core technical personnel such as large model full-stack engineers and deep learning R & D engineers. Judging from the number,There are 24 recruitment positions in Beijing and 10 in Hangzhou。
In addition to the composition of employees, DeepSeek’s entire management model is also different.
Zheng Qing, assistant director of the Beijing General Artificial Intelligence Research Institute, told Zhidongxi that judging from DeepSeek’s employee structure, they break traditional standards when selecting people, mainly young people, and value people’s abilities and enthusiasm for work more than qualifications and experience.
Unlike many CEOs,Founder Liang Wenfeng will deeply participate in front-line work。She added that DeepSeek’s management is flat, DeepSeek founder Liang Wenfeng will be on the front line, and there is no hierarchical management model. An article by smart investors on the Weixin Official Accounts mentioned that colleagues ‘evaluation of Liang Wenfeng in 2018 was that he had no other hobbies except programming.
It has top talents and can effectively stimulate its innovative vitality, complementing DeepSeek’s unique employee structure.
03.
Liang Wenfeng creates a “R & D Utopia”:
Million annual salary, no check-in without KPI
What kind of unique working atmosphere do such relaxed office conditions create? This is also a key piece of the puzzle on DeepSeek’s rise.
Looking at public information, we can find that DeepSeek’s registration location in Beijing was originally located atNew Materials Entrepreneurship Building, No. 7 Fenghui Middle Road, Haidian District, Beijing City, changed to November of the same yearRongke Information Center at No. 2 Xueyuan South Road, Haidian District, Beijing City, and will officially settle in May 2024.
It is understood that the Rongke Information Center of Guangna International’s top company covers an area of 8 hectares and has a total construction area of 328,000 square meters. However, the floor area ratio is only 3.1, and it is the only park within the Fourth Ring Road in Beijing City. With a greening rate of 42%.
Because there were many people from all walks of life who wanted to visit DeepSeek at this time, its property management company also installed a separate gate and added security personnel on the floor where DeepSeek was located.
The hardware conditions are full, and the company’s office atmosphere is relatively relaxed.
According to students who have interned at DeepSeek, they have no fixed commuting hours and no KPIs. The work model is mainly small and interest-oriented, and the research progress is discussed regularly every week.
Talking about his impression of Liang Wenfeng, he added that Liang Wenfeng has no restrictions on employees and will participate in the company’s major project research and development. On the premise of ensuring work, he will provide more benefits and additional subsidies to interns.
▲ Liang Wenfeng, founder of DeepSeek
In May 2023, Liang Wenfeng once mentioned in an interview with Dark Wave that innovation requires as little intervention and management as possible, so that everyone has the space to play freely and the opportunity to trial and error. Innovation is often self-generated, not deliberately arranged, let alone taught.
An employee who left DeepSeek in September last year said: “Liang Wenfeng gave us control and treated us like experts. He kept asking questions and learning with us. DeepSeek put me in charge of core R & D, which is very exciting.”
Such a talent structure and company atmosphere are more suitable for companies like DeepSeek that are more R & D and are not complex at the product level.
Li Xinyu, deputy director of the Beijing Academy of Science and Intelligence, explained that AI has developed to this day,Higher requirements are placed on the learning ability and practical ability of talents. Against this background, students who are still in college or have just graduated have these two abilities at their peak.And they are less restricted by the outside world, so this group of people is more suitable for making breakthrough results in DeepSeek, a startup that focuses on algorithm innovation.
Therefore, DeepSeek’s recruitment requirements are more important-basic abilities, creativity, love, etc. Liang Wenfeng said: “From this perspective, there are many suitable candidates in China.”
In addition, Liang Wenfeng is quite heroic in terms of salary level. A quantitative fund manager who was familiar with Liang Wenfeng revealed that in magic square quantification, it is not uncommon for senior data scientists to make an annual salary of 1.5 million yuan, while the annual salary of competitors rarely exceeds 800,000 yuan.
It can be seen from the current recruitment information disclosed by DeepSeek that the deep learning researcher it is recruiting-AGI has a monthly salary of 80,000 to 110,000 yuan, with an annual salary of up to 1.54 million yuan, and the monthly salary of core system R & D engineers (school recruitment) has a monthly salary of 60,000 yuan to 90,000 yuan. After an estimated salary of 14 a year, the annual salary can reach up to 1.26 million yuan, and the monthly salary of large model full-stack engineers is 50,000 to 80,000 yuan, with an annual salary of up to 1.12 million yuan.
04.
Take root in the “rich mine” of AI talents
The government’s 100 billion investment fund has been implemented
In the AI industry, a “battlefield” where top talents compete fiercely, talent is undoubtedly a key factor determining the destiny of a company. From this perspective, it is inevitable for DeepSeek to focus its research and development on Beijing.
There is still a certain gap in the construction of AI-related disciplines in China compared with developed countries. In 2022, the Ministry of Education will release the “Catalog of Postgraduate Education Subjects”, establishing intelligent science and technology as a first-level discipline in the cross-disciplinary category.
Zheng Qing said that in fact, the current top domestic AI talents are still scarce in my country’s entire job market, making the independent training of talents even more urgent. Talent training is a long-term process. It can start from the aspects of curriculum design, teachers and application practice, and is committed to cultivating top talents with cross-field and composite characteristics.
Against the background of talent scarcity and accelerating development of the AI industry, Beijing’s leading edge in AI talents has attracted more attention.
Beijing has always been a gathering place for AI talents. Data from the Beijing City Science and Technology Commission and the Zhongguancun Science and Technology Park Management Committee show that Beijing AI scholars have exceeded15,000 people, the national proportion is 30.6%,ranked first。
According to statistics, Beijing has Peking University, Tsinghua University, Beihang University, etc.More than 30 universities have been approved for undergraduate majors in artificial intelligence, yesMore than 20 universities have established artificial intelligence colleges or research institutes, 32 universities have doctoral and master’s degrees authorizations in disciplines directly related to artificial intelligence,There are 44,000 graduate students and 12,000 teachers and scientific researchers。
Talent and financial support are undoubtedly necessary conditions for the current development of AI, and Beijing City is even more vigorous in investing in government industrial funds.
In the past two years, Beijing’s investment in the AI industry has continued to increase.
Since 2019, Beijing’s Haidian District, Economic and Technological Development Zone, and Shunyi District government investment funds have been continuously expanded, including the release of Beijing Haidian Zhongguancun Science and Technology Development Fund Phase III, and the total fund size has increased to20 billion yuan; The second phase of the Beijing Economic and Technological Development Zone Government Investment Guidance Fund announced that it will be officially established, with the total scale10 billion yuan; Beijing Shunyi Equity Investment Guidance Fund (Limited Partnership) officially released, long-term scale10 billion yuan, the first phase is 1 billion yuan……
In addition, in 2024, Beijing City will be implemented in one go8 industry funds, the capital scale is 10 billion yuan or 20 billion yuan, covering multiple fields such as robots, AI, advanced manufacturing and intelligent equipment.
Just last month, the special press conference of the “Beijing Specific Intelligent Technology Innovation and Industry Cultivation Action Plan (2025-2027)” held in Beijing clearly mentioned that Beijing has established a total scale1000 billion Yuan, a government investment fund with a duration of 15 years, focusing on supporting future industrial fields such as AI and robotics.
At present, the venture capital industry in several major first-tier cities in China is booming. According to data from the Zero 2 IPO Research Center,Fundraising scale in 2024Beijing, Zhejiang and Jiangsu rank among the top three, with scales of 189.279 billion yuan, 182.664 billion yuan and 171.780 billion yuan respectively;National investment scaleThe first place is Shanghai, with 110.145 billion yuan, Beijing ranks second with 104.272 billion yuan, and the third place is Shenzhen, with 40.914 billion yuan. among themBeijing ranks first in the country in terms of fundraising scale and second in terms of investment scale, is an out-and-out venture capital highland.
05.
Conclusion: Creating an AI super ecosystem
Stimulating the emergence of “DeepSeek” companies
From the explosion of the big model industry to the birth of DeepSeek, a phenomenal company in the AI industry, the competition among cities across the country in emerging industries has become more intense.
For the AI industry, its development has a strong overall perspective. Its technology research and development involves multiple subject fields such as mathematics, computer science, statistics, and neuroscience. Application scenarios cover medical care, transportation, finance, education, and manufacturing, etc. Almost all industries, plus the length of the upstream and downstream industrial chains of the AI industry, from basic algorithm research and chip manufacturing, to model development and data annotation in the middle, to downstream application product development and service provision, all links are closely connected.
Therefore, from the government level, Beijing, Hangzhou, Shanghai, Shenzhen and other cities have successively issued various support policies, strengthened capital investment, talent introduction and other measures to accelerate the development of AI around computing power clusters, data supply, scenario applications, etc. in the AI industry. Development and popularization.
On this basis, urban power is relying on a complete AI development ecosystem to form strong results transformation capabilities, making it possible for more “DeepSeek” to emerge.
