DeepSeek Announces V4 Model and Initiates Funding: Strategy to Retain 97% of Employees

DeepSeek Unveils V4 and Launches Initial Funding: The “Unconventional” AI Company Led by Liang Wenfeng

In late April 2026, Chinese AI company DeepSeek announced its latest large-scale language model, “DeepSeek V4.” Simultaneously, the company revealed that it was initiating external funding for the first time, drawing significant attention from the industry. What stands out is the company’s remarkable ability to maintain stability within its R&D team, despite fierce competition for talent.

Rumors of Talent Drain and the Hidden Truth

Since the debut of DeepSeek’s R1 in early 2025, the company has been in the spotlight of the Chinese AI industry. However, between 2025 and 2026, there were reports of key members, including Luo Fuli, Wang Bingxuan, and Guo Daya, leaving for Xiaomi, Tencent, and ByteDance. This led to widespread concerns about a serious talent drain.

Yet, the truth revealed in V4’s technical report tells a different story. According to the acknowledgments section of the report, only 10 out of the approximately 270 researchers and engineers left during the development phase. This translates to a turnover rate of less than 4%, a stark contrast to OpenAI, which has reportedly lost over 25% of its key research talent in the past two years. CEO Liang Wenfeng has managed to retain the majority of the company’s top talent.

Why Funding Now?—Providing “Confirmed Valuation”

In early 2025, when R1 was at its peak, Liang Wenfeng rejected all investment proposals. Now, in 2026, amid intensified competition, the company has opened the door to external funding for the very first time. The rationale behind this “unconventional” move lies in the company’s intent to provide a confirmed valuation to its core members who have contributed significantly over the years.

Specific developments include an increase in the company’s registered capital in April and a rise in Liang Wenfeng’s direct shareholding ratio from 1% to 34%. As a result, Liang Wenfeng now owns approximately 84% of the company’s shares, either directly or indirectly, solidifying his control. Sources say that only select investors will be allowed to participate in this funding round. Although the company collaborates regularly with Tencent and Alibaba, no substantial investment negotiations are reportedly underway. Rumors suggest a pre-money valuation of 300 billion yuan (approximately 6 trillion yen), but the company has not officially commented on this.

Technological Advancements and Ecosystem Impact

The V4 model includes a Pro version with up to 1.6 trillion parameters and an expanded context length of 1 million tokens. Furthermore, its pricing is highly competitive, with the Pro version costing 1 yuan (approximately 20 yen) per 1 million tokens for input during cache hits and 24 yuan (approximately 480 yen) for output.

Technologically, one of the most significant developments in the V4 report is the mention of Huawei Ascend and NVIDIA as verification platforms. This demonstrates compatibility with China’s indigenous AI chips alongside NVIDIA’s CUDA ecosystem, which is considered the industry standard. If Huawei Ascend 950 ultra-nodes enter mass production, the Pro version’s price could drop even further. This development holds substantial implications for China’s domestic AI chip industry, potentially driving stock price increases and new order opportunities for related companies.

Ripples from the “Slow” Strategy

DeepSeek’s approach has diverged from that of typical startups. While conventional startups prioritize rapid growth, fundraising, and going public, DeepSeek has chosen a deliberately slower path. This “slow” strategy is now beginning to exert a broad impact on the industry. The announcement of V4 is not merely a model upgrade but a catalyst for accelerating the formation of a domestic AI ecosystem in China.

Liang Wenfeng has cultivated an organizational culture that prioritizes fundamental skills and creativity over short-term results. This has enabled the company to minimize turnover rates and establish a solid foundation for long-term technological development. Now, building upon this foundation, DeepSeek is integrating funding and ecosystem strategies, setting the stage for a new chapter in China’s AI industry.

FAQ

Q: Is DeepSeek V4’s turnover rate truly low? How does it compare to other companies?
A: According to V4’s technical report, the turnover rate among DeepSeek’s 270-member R&D team is less than 4%. This is exceptionally low, especially given the competitive nature of the AI industry. In comparison, OpenAI has lost over 25% of its key research talent in the past two years, underscoring DeepSeek’s organizational stability.

Q: Why has DeepSeek decided to start funding now?
A: Although all funding proposals were rejected when R1 was launched in early 2025, DeepSeek has initiated funding in 2026 amid fierce competition. The move is intended to provide a confirmed valuation to long-standing core members, boosting morale and securing future development. Additionally, it aims to raise capital for V4’s development and the company’s ecosystem expansion.

Q: What are the key technological implications of DeepSeek V4?
A: The most significant aspect is its demonstration of compatibility with China’s domestic AI chips, such as Huawei Ascend, alongside NVIDIA’s CUDA ecosystem. This could diversify the AI development foundation within China and potentially lower costs for consumers. Furthermore, the Pro version’s scale of 1.6 trillion parameters and competitive pricing could accelerate the adoption of AI technology.