DeepSeek’s First Fundraise: $100B Valuation and $300M Target Amid Talent Exodus

According to reports from The Information and Reuters, DeepSeek – a domestic AI startup known for its high‑performance, low‑cost inference model DeepSeek‑R1 – is in talks with investors to raise at least $300 million at a valuation of roughly $100 billion.

DeepSeek has never taken external venture capital; it was spun out of the top‑tier quantitative hedge fund Fantasia Quant, with early compute resources and R&D funding fully provided by its parent company and internal capital. The firm previously turned down multiple financing offers from leading Chinese VCs and tech giants.

The current financing move reflects the growing capital demands of developing and operating cutting‑edge large‑model AI systems, especially as advanced inference capabilities and AI agents become central to the field. Recent job postings for a data‑center operations engineer in Inner Mongolia suggest DeepSeek is transitioning to a “heavy‑asset” model, building its own compute infrastructure to lower long‑term costs – a shift that requires substantial cash reserves.

Simultaneously, the industry is watching the upcoming release of DeepSeek‑V4, the next‑generation foundational model, which will further increase funding needs. DeepSeek’s API pricing remains a fraction of that of U.S. competitors such as OpenAI and Anthropic, and external capital would help the company maintain its aggressive pricing stance in the global AI price war.

If the financing proceeds at the reported valuation, the equity dilution would be minimal, allowing the parent Fantasia Quant and the founding team to retain full control and making a move to a closed‑source model unlikely.

DeepSeek’s success stems from a focused strategy emphasizing compute efficiency and core base models, including MoE architectures and reinforcement‑learning‑driven R1. However, recent talent departures raise concerns: core researcher Luo Fuli has joined Xiaomi to lead AI efforts, and 95‑born researcher Guo Dayi, a key contributor to DeepSeekMath, DeepSeek‑V3, and DeepSeek‑R1, has moved to ByteDance’s Seed team as an agent lead (L8). Speculation about Guo’s compensation suggests a near‑¥100 million annual package, which ByteDance’s VP Li Liang clarified as a combination of cash, company stock, and ByteDance‑issued options, with potential multi‑billion‑yuan payouts after four years for top performers.

These talent shifts illustrate the broader industry trend of AI expertise moving from pure research toward engineering validation and ecosystem construction. Future financing could provide DeepSeek with the resources needed for broader commercial deployment of its models.