DeepSeek Seeks $300M Funding at $10B Valuation
Key Highlights
– First external funding round for DeepSeek — targeting $300M+ at >$10B valuation (~¥6.8B RMB)
– Strategic shift from self-funding to institutional capital amid rising AI R&D costs
– Critical hires lost to Xiaomi and ByteDance; V4 launch delayed due to Huawei Ascend porting
– U.S. investor caution noted amid export controls on Blackwell GPUs
Breaking: DeepSeek Enters External Fundraising Phase
According to The Information, DeepSeek — the China-based AI research lab founded by Liang Wenfeng and incubated by hedge fund Hunyuan Quant — has officially commenced its first external fundraising round, aiming for a valuation exceeding $10 billion (approximately ¥6.818 billion RMB). The company plans to raise at least $300 million, with proceeds earmarked to bolster infrastructure, accelerate next-gen model development, and retain top-tier talent.
This marks a pivotal strategic inflection point: DeepSeek had long maintained a closed, self-sustaining financial model — rejecting multiple investment overtures from leading Chinese VCs and tech giants. Its prior operations were fully funded by Hunyuan Quant’s proprietary trading revenues.
Why Now? Pressures Mount on All Fronts
🔹 Capital Intensity of Frontier AI
Training state-of-the-art large language models now demands exponentially higher compute budgets — especially as competition intensifies across the U.S.-China AI axis. DeepSeek’s reliance on legacy A100/H800 clusters (originally deployed for quant finance) is no longer sufficient for competitive scaling.
🔹 Talent Exodus Accelerates
Since early 2025, key researchers have departed:
– Luo Fuli, core contributor to DeepSeek-R1 and architect of V3, joined Xiaomi’s AI division.
– Guo Daya, senior researcher in model optimization, moved to ByteDance with a significant compensation uplift.
Funding will directly support competitive compensation packages and expanded research teams — critical to reversing attrition.
🔹 Product Delays & Architectural Shifts
DeepSeek originally scheduled the release of its flagship DeepSeek V4 for February 2026. However, the launch has been postponed multiple times due to:
– A major engineering pivot toward native compatibility with Huawei Ascend chips, reducing dependence on restricted NVIDIA hardware.
– Ongoing training on NVIDIA Blackwell architecture, constrained by U.S. export controls — limiting chip availability and increasing procurement complexity.
Market Implications & Investor Sentiment
While demand for DeepSeek’s capital remains strong domestically, several U.S.-based investors are adopting a cautious stance, citing regulatory scrutiny — including prior inquiries by the U.S. Congress. Notably, DeepSeek remains outside all U.S. trade restriction lists, preserving access to global cloud infrastructure and enterprise partnerships.
Still, the move underscores a broader industry trend: even elite, well-resourced AI startups can no longer outpace hyperscalers on capital alone. Strategic capital infusion may be essential not just for survival — but for sovereign AI stack independence.
Background: From R1 Breakthrough to V4 Ambitions
DeepSeek-R1 stunned the AI community in early 2025 by delivering near-SOTA reasoning performance at a fraction of the cost — trained efficiently on older-generation GPUs. Its success triggered renewed investor interest in AI efficiency economics, reshaping assumptions about chip demand cycles.
With this new round, DeepSeek aims to:
– Scale training capacity using hybrid GPU/Ascend infrastructure,
– Launch V4 with enhanced multimodal capabilities and real-time inference optimization,
– Expand commercial licensing and enterprise API offerings globally.
Source: The Information, reported via “Zhixi Dong” (SmartThings News)