UK Media Reveals More Details About DeepSeek's External Funding: Not Short on Funds, Aiming to Retain Talent
According to a report by the Financial Times on April 24th, AI startup DeepSeek is conducting its first external funding round to retain employees who are compensated with stock options, preventing them from being poached by competitors.
[Image Caption: DeepSeek]
According to two people familiar with the matter, DeepSeek is in talks with a small group of strategic investors, and this round of funding could value the company at over $20 billion.
Sources said that while DeepSeek's free cash flow is not as abundant as that of larger competitors, the company does not have an urgent or significant need for external funding. Instead, this round of funding is mainly to appease the company's researchers, some of whom have already jumped ship to competitors whose valuations have risen sharply over the past year.
In the salary structure of AI researchers, stock options often account for a large proportion, even a core component. However, options require an underlying valuation.
Sources reveal that the amount of this funding is expected to be a "symbolic" amount, in the low hundreds of millions of dollars, significantly lower than the billions of dollars typically raised by peers. In other words, a primary purpose of this funding is to establish a valuation.
Sources said that DeepSeek founder Liang Wenfeng is also considering other ways to determine the company's valuation, including stock buybacks or valuation methods based on the company's performance, in case financing terms cannot be reached.
Last year, DeepSeek released the R1 model, which shocked the world. The model claims to be close in performance to US peer models such as ChatGPT, but at a much lower training and usage cost.
According to a report by the Financial Times last year, despite its overnight fame, Liang Wenfeng rejected external investors, choosing to focus on research rather than commercialization. He has been supporting DeepSeek with funds from his own quantitative trading company.
For financial institutions, DeepSeek's lack of a business model may hinder its financing. Strategic investors typically seek synergies with their own businesses in transactions, but according to someone familiar with the negotiations, Liang Wenfeng is unwilling to disclose information beyond minimum financial data, which also poses a challenge.
Reasons for Talent Loss
Over the past few months, DeepSeek has lost several key researchers. According to people familiar with the personnel changes, Guo Daya, the main author of the R1 paper, has joined ByteDance, and Wang Bingxuan, a senior member of the DeepSeek model training team, has resigned and joined Tencent.
Although DeepSeek's cash compensation is competitive, the lack of a clear valuation puts it at a disadvantage when competing with tech giants such as Alibaba and Tencent, as well as peer startups.
Sources reveal that Kimi AI model operator Moonshot AI was valued at $18 billion in its last round of funding. Two AI developers listed in Hong Kong, MiniMax and Zhipu AI, are valued at $34 billion and $58 billion respectively.
Nevertheless, people familiar with DeepSeek's financing say that only a small number of researchers have left, and most employees remain focused on Liang Wenfeng's idealistic path. The company offers a flexible work environment and a flat reporting structure similar to a research lab, which is hard to find elsewhere.
“If the valuation can be more certain, DeepSeek will be able to continue doing what it does best: pursuing ultimate AI amidst the chaos,” said a source.
As of press time, DeepSeek has not commented on the matter.