[News] DeepSeek Rumored to Raise USD 300 Million at Over USD 10 Billion Valuation
Per a report by The Information citing multiple sources, Chinese large language model (LLM) contender DeepSeek is initiating its first external equity financing since inception. The company aims to raise at least USD 300 million at a valuation of no less than USD 10 billion. As of press time, DeepSeek has not issued an official response.
DeepSeek’s transition from a capital-averse stance to active fundraising reflects structural pressures within the AI sector. As the industry enters a capital-intensive phase—characterized by soaring compute costs and intensifying competition for talent—external financing has become increasingly necessary.
Historically, DeepSeek has maintained a high degree of independence, supported by its parent firm High-Flyer. With strong performance in 2025 and assets under management exceeding CNY 70 billion, High-Flyer provided sufficient backing for early-stage R&D. However, as model development advances into the trillion-parameter era, the exponential rise in compute and development costs is outpacing the financial capacity of a single quantitative fund.
Talent retention has also emerged as a pressing challenge. Following the instant success of the DeepSeek-R1 model in early 2025, key team members have reportedly been recruited by major technology firms. Notably, contributor Luo Fuli joined Xiaomi, while core researcher Guo Daya moved to ByteDance’s Seed team. Additional talent departures to companies such as Tencent and DeepRoute.ai have also been reported. Without external funding and a clear market valuation, DeepSeek has faced difficulties in offering competitive equity incentives, weakening its ability to retain key personnel.
Establishing a formal valuation through this financing round is therefore seen as critical to unlocking employee stock option value, stabilizing the core team, and strengthening competitiveness in the ongoing talent war.
Notably, sources indicate that V4 will feature a trillion-parameter scale, adopt a Mixture-of-Experts (MoE) architecture, and support context lengths at the million-token level. The important point is that V4 is expected to advance DeepSeek’s strategic goal of reducing reliance on NVIDIA by enabling full compatibility with domestic AI chips such as Huawei Ascend.
However, the transition from the CUDA ecosystem to alternative compute architectures has proven complex. As a result, V4’s release has been delayed multiple times, shifting from the originally planned February timeline to late April. A 13-hour service outage on the DeepSeek platform in late March has also been speculated to be linked to V4’s gray-scale testing and underlying infrastructure adjustments.
The planned USD 300 million in funding is expected to be allocated toward expanding compute capacity, ensuring smooth deployment of V4 on domestic chips, and accelerating commercialization efforts.
The global LLM funding landscape has become increasingly competitive. Industry leaders such as OpenAI and Anthropic continue to break fundraising records, with valuations reaching into the hundreds of billions of dollars.
In contrast, while China’s LLM financing activity saw a slowdown in Q1 2026, capital is increasingly concentrating among leading players. DeepSeek’s entry into this arena with a USD 10 billion valuation, yet far lower than global frontrunners, places it firmly within the top tier of domestic AI companies. That said, geopolitical factors remain a potential source of uncertainty for the deal.
(Photo credit: DeepSeek)
