Multiple people familiar with the matter revealed that Chinese artificial intelligence startup DeepSeek is preparing for its first external financing since its establishment. This round of valuation is expected to be as high as 50 billion US dollars, marking that this large language model research and development company will reverse its strategy of rejecting external capital for many years.
One of the sources close to the transaction said that China’s National Artificial Intelligence Fund, which was established in January last year and has a scale of 60 billion yuan, is negotiating with Deepin to lead this round of financing. Multiple people familiar with the matter said that the scale of this round of financing is expected to be between US$3 billion and US$4 billion, and the funds will mainly be used to expand computing resources and improve employee compensation and benefits to cope with fierce industry competition. People familiar with the matter also said that Tencent Holdings is also in contact with Shenzhen Search for potential investment.
Depth Search has not yet responded to Reuters for relevant news. China Integrated Circuit Industry Investment Fund, the main financier of the National Artificial Intelligence Fund, declined to comment, while Tencent also declined to comment. The Financial Times previously reported, citing sources, that the China Integrated Circuit Industry Investment Fund was negotiating to lead an in-depth search for first-round financing, with the deal valuing the cutting-edge artificial intelligence laboratory at approximately US$45 billion.
At the time of this first round of external financing, Deep Search is facing pressure from a group of well-funded competitors. These competitors include model teams under large technology groups such as ByteDance and Alibaba, as well as cutting-edge startups that have raised billions of dollars in public and private markets, such as MiniMax and Dark Side of the Moon.
Informed sources said that Liang Wenfeng, the founder of Deep Search, was directly involved in this round of financing negotiations. Over the years, he has been quite unique in China's artificial intelligence industry - choosing to rely on his own quantitative hedge fund Gaofei Capital to provide financial and operational support to the company, rather than relying on large technology groups or raising funds through initial public offerings like many of his peers, preferring to manage the company as a research laboratory model. Liang Wenfeng is currently unavailable.
In the context of fierce competition, some deep-search researchers have been poached by their opponents in recent years. For example, Luo Fuli, the former depth search researcher, resigned last year and became the head of Xiaomi's large model team MiMo. At the same time, the rapidly evolving artificial intelligence industry has largely replicated and surpassed the technical path that Deep Search relied on earlier when the low-cost, high-efficiency open source chatbot model became globally popular.
Industry focus is shifting from chatbots to “agent” systems that rely less on human intervention to complete complex tasks but require significantly more computing power. Deep Search claimed last month that its new generation V4 model for agent scenarios "redefines" the latest level of open source models. However, third-party evaluation results show that the overall performance of the model still lags behind some of the top products from leading Chinese and American companies. It's worth noting that the V4 launch didn't repeat the global sell-off in tech stocks that triggered last year's launch of its predecessor models, the V3 and R1.