Produced by Leida Finance | Text by Ding Yu | Edited by Meng Shuai

Amidst fierce competition among AI giants, has DeepSeek, once known for its 'maverick' stance and longstanding resistance to external capital, also found itself 'short on cash'?

According to foreign media reports on April 17, DeepSeek is in discussions to introduce external capital for the first time, aiming to bolster its financial reserves and secure a more favorable position in the high-stakes competition to develop top-tier artificial intelligence models.

Reports indicate that DeepSeek plans to raise no less than $300 million (approximately RMB 2 billion), with a company valuation of at least $10 billion (approximately RMB 68.2 billion). However, this valuation is notably lower than the peak of $150 billion suggested by foreign media last year.

Amid news of DeepSeek's financing, its brain drain has also garnered external attention. Recently, there have been reports of post-95s researcher Guo Daya being poached by ByteDance with a lucrative salary. Earlier, there were also departures such as 'AI prodigy' Luo Fuli, who left for Xiaomi, and Wang Bingxuan, the core author of DeepSeek's LLM, who was recruited by Tencent.

Notably, in late March this year, DeepSeek experienced large-scale service disruptions for three consecutive days. Meanwhile, DeepSeek's new flagship large model, DeepSeek V4, is slated for an official release in late April. Additionally, recent job postings by DeepSeek suggest that the company intends to build its own data center in Inner Mongolia.

Some believe that completing this round of financing at this critical juncture will help DeepSeek advance model iteration, talent acquisition, and global market expansion.

Opening Doors to External Capital: DeepSeek Rumored to Be Seeking Financing

Recently, news of DeepSeek's rumored financing has stirred up both the investment and AI circles.

Reports indicate that DeepSeek is planning to raise no less than $300 million (approximately RMB 2 billion) at a valuation exceeding $10 billion (approximately RMB 68.2 billion).

According to Tianyancha, DeepSeek is affiliated with Hangzhou DeepSeek Artificial Intelligence Foundation Technology Research Co., Ltd., established in July 2023.

For a long time, DeepSeek has been known in the industry for 'not seeking financing,' with its operational funds primarily supported by High-Flyer Quantitative, a company under founder Liang Wenfeng's name.

According to LatePost, in 2023, Liang Wenfeng met with a select group of investors. However, Liang hoped that investors would accept a cap on returns, which deterred many investment institutions.

In the following years, despite DeepSeek causing a sensation in the industry with its low-cost models, Liang did not engage with other investors again or even establish new contacts.

According to The Information, before DeepSeek sought external financing in this round, the company had repeatedly rejected approaches from well-known domestic venture capital firms and large tech companies.

DeepSeek's decision to break with convention and seek external funding reflects the common challenges faced by the AI industry: as competition intensifies, players require massive capital investments to maintain their leading positions in the field.

Notably, in early April this year, new developments emerged in DeepSeek's recruitment, with job locations appearing for the first time in Ulanqab, Inner Mongolia, in addition to the company's previous offices primarily in Beijing and Hangzhou.

The two positions DeepSeek is recruiting for are Senior Data Center Operations Engineer and Senior Data Center Delivery Manager, both with salaries ranging from RMB 15,000 to RMB 30,000.

This marks DeepSeek's first recruitment for physical infrastructure positions, as the company has previously relied more on renting cloud services. This is widely interpreted as DeepSeek potentially building its own data center in Inner Mongolia.

Furthermore, this comes at a critical juncture as DeepSeek V4 is about to be released. Reports suggest that this version of the model will have parameters in the trillions, support context lengths in the millions, and will abandon its long-term reliance on NVIDIA chips, achieving deep compatibility and optimization with domestic AI chips such as Huawei's Ascend for the first time.

To complete the ecological migration from NVIDIA's CUDA to Huawei's CANN architecture, DeepSeek's engineers spent a significant amount of time addressing V4's compatibility with Huawei chips and rewriting core code.

Therefore, some believe that the multiple financial pressures brought about by these moves may be a key reason for DeepSeek's current pursuit of financing.

Frequent Departures of Core Personnel: Perhaps to Retain Talent Through Option Pricing

DeepSeek's shift in attitude toward external capital may also hide an intention to retain talent through option pricing.

Since its rise to prominence in early 2025, major tech giants have set their sights on DeepSeek, a rising star in the AI industry. As a result, the company has faced challenges with core personnel being poached across multiple technical directions, from LLM infrastructure, multimodality, and OCR to reasoning and agents.

According to Huxiu in June last year, half a year ago, a core executive at DeepSeek quietly left to start a business and secured financing from IDG Capital, planning to release an agent product around Christmas 2025.

A source told Huxiu that the executive was the former CTO of DeepSeek. However, another insider revealed that there is no clear CTO position within DeepSeek, but there are individuals responsible for corresponding work.

At the end of 2024, there were also rumors that Xiaomi founder Lei Jun personally offered Luo Fuli, the 'AI prodigy,' a multimillion-dollar annual salary to join Xiaomi.

A year later, in November 2025, this 'AI prodigy' who participated in the development of DeepSeek-V2 officially announced on her social media that she had joined Xiaomi as the head of the Xiaomi MiMo large model.

At the end of last year, Yao Shunyu, the chief AI scientist in Tencent's 'CEO/President Office,' poached Wang Bingxuan. It is reported that Wang was the core author of DeepSeek's first-generation large language model (LLM) and subsequently participated in the training of subsequent models.

Around the Spring Festival this year, Wei Haoran, the core author of the DeepSeek-OCR series, also left the company and is expected to join a major company.

On April 12 this year, Zhou Guang, CEO of the autonomous driving technology company Yuanrong Qixing, confirmed that Ruan Chong, a core researcher in DeepSeek's multimodal technology, had joined the company several months earlier as chief scientist.

Recently, there have been rumors that Guo Daya, a post-95s researcher at DeepSeek, has been poached by ByteDance with an annual salary of RMB 100 million.

Leida Finance learned that Guo Daya is a top talent in code intelligence and large model reasoning. After completing his Ph.D. in 2023, he joined DeepSeek and deeply participated in research on DeepSeek's code and reasoning directions. He was also a core contributor to models such as DeepSeek V3, R1, Coder, and Math.

After news of his departure broke in March this year, this 'genius teenager' became a highly sought-after target for major companies. According to China Fund News, ByteDance, Baidu, and Alibaba all made offers to him.

According to LatePost, Guo Daya will be responsible for the large model research and development organization Seed at ByteDance and serve as one of the heads of the agent team, with a job level of L8.

However, on the evening of April 16, Li Liang, Vice President of Douyin Group, posted on Weibo that the company had not recently hired any employees with an annual salary of nearly RMB 100 million.

Li also pointed out that ByteDance's salary system for all Seed team technical personnel is consistent, including three parts: cash, ByteDance options, and Doubao options. Among them, the options are fully vested over four years, with no situation where 'certain conditions must be met to receive the full amount.' However, Li also emphasized that if the business performs well, it is not ruled out that some Seed technical personnel may earn hundreds of millions of yuan in revenue after four years.

An investor who has invested in large models told Yicai that even if DeepSeek opens up for financing, it is not a game for most people, and according to Liang Wenfeng's ideas, the terms are likely to be exceptionally stringent. Regarding this shift in financing, the investor judged that it is probably for employee option pricing and vesting, and 'it's being done too late.'

No Longer Solely Reliant on High-Flyer Quantitative for 'Blood Transfusion': How Will Its 'Uniqueness' Be Maintained?

As the founder and leader, Liang Wenfeng can be considered the 'soul figure' of DeepSeek.

Public information shows that Liang Wenfeng was born in Zhanjiang, Guangdong, in 1985. At the age of 17, he was admitted as the top scorer in the college entrance examination at Wuyuan No.1 High School to Zhejiang University's undergraduate program in Electronic and Information Engineering, and later continued to pursue a graduate degree in Information and Communication Engineering at Zhejiang University.

During the 2008 financial crisis, Liang, still a graduate student, saw hidden opportunities and began researching how to use machine learning and other technologies to explore fully automated quantitative trading.

In 2013, Liang Wenfeng and his Zhejiang University alumnus, Xu Jin, among others, founded Hangzhou Jacobi Investment Management Co., Ltd., an investment company named after the German mathematician Carl Jacobi.

Two years later, Liang founded High-Flyer Quantitative, dedicated to quantitative investment through mathematics and artificial intelligence.

In October 2016, High-Flyer's first stock position generated by a deep learning algorithm model went live for real-time trading, using GPUs for computation. By the end of 2017, nearly all of the company's quantitative strategies had adopted AI model calculations.

In 2018, High-Flyer established AI as its main development direction. In 2019, High-Flyer AI (High-Flyer Artificial Intelligence Foundation Research Co., Ltd.) was registered and established, dedicated to AI algorithm and basic application research. That same year, High-Flyer Quantitative's management scale exceeded RMB 10 billion.

Earlier this year, media outlets cited data from Simuwang, stating that High-Flyer Quantitative's average return in 2025 reached 56.55%, ranking second in China's performance rankings for quantitative private equity firms with management scales exceeding RMB 10 billion, second only to Lingjun Investment, which topped the list with an average return of 73.51%.

Data shows that High-Flyer Quantitative's average return over the past three years is 85.15%, and over the past five years, it is 114.35%. Currently, High-Flyer Quantitative's management scale exceeds RMB 70 billion.

Some analysts believe that High-Flyer Quantitative's strong performance has provided DeepSeek, under Liang Wenfeng's leadership, with sufficient research and development funds.

However, according to LatePost, although DeepSeek's absolute salaries are not low, external offers are even higher. Some headhunters stated that DeepSeek's competitors have offered 'irresistible figures,' with 'doubling or tripling the salary being no problem,' and 'other companies offering eight-figure (including stock or option) total packages.' Faced with massive offers, although some have left DeepSeek, many more have chosen to stay. This is because they recognize Liang Wenfeng's approach to pursuing AGI and are willing to engage in exploration that is not competition-driven; they are also accustomed to DeepSeek's relatively relaxed and unhurried research atmosphere.

Many who have interacted with Liang Wenfeng describe him as follows: He is a rare figure in China's current AI circle who 'possesses both strong infra engineering capabilities and model research capabilities, can mobilize resources,' has 'terrifying learning abilities,' and at the same time 'does not resemble a boss at all, but more like a geek.' As the 'glue' between teams within the company, Liang Wenfeng enables the entire company to spontaneously form close collaboration. He devotes almost all his time to the few things he believes should be the focus, 'organizing research and development, coordinating resources, and conducting specific research,' making him a mentor-like boss.

Liang Wenfeng also believes that it is difficult for a person to maintain high-quality output for more than 6 to 8 hours a day. Fatigue-induced poor judgment from overtime work can waste valuable computational resources, making it not worth the cost.

Therefore, DeepSeek has no explicit performance evaluations or DDLs (deadlines). The company does not require overtime, and employees do not clock in. Most members leave the office around 6 to 7 pm on weekdays.

This contrasts sharply with the work hours of core AI development personnel at many domestic and foreign AI giants, who often work 70 to 80 hours a week. However, under this somewhat 'special' work environment, DeepSeek previously gained fame for its unprecedented cost-effectiveness.

However, some believe that as AI exploration gradually reaches diminishing marginal returns, the cost for companies to gain a leading edge is increasing. Maintaining DeepSeek's previous work mode may become increasingly difficult.

If DeepSeek successfully completes its first round of financing and receives a financial 'blood transfusion,' will it be able to recreate the 'AI myth' from early 2025? Leida Finance will continue to follow the story.