Why Is DeepSeek Raising $300M Despite Its $10B Valuation?

Core data

Financing type: First external round

Target valuation: > $10 billion (≈ 680 billion CNY)

Funding amount: ≥ $300 million (≈ 20 billion CNY)

Current status: In talks, not yet announced

Expected closing: May‑June 2026

Why raise capital despite abundant cash?

Reason 1 – Compute cost explosion

DeepSeek’s R1 model was trained for $5.576 million and approached OpenAI o1 performance, relying on reuse of existing chips. New models are scaling from hundreds of billions to trillions of parameters, causing training cost to grow exponentially. Chip supply is constrained: Nvidia Blackwell GPUs are limited, and Huawei Ascend chips are still being adapted.

模型参数：从千亿级 → 万亿级
训练成本：指数级增长
芯片获取：英伟达 Blackwell 受限，华为昇腾适配中

Reason 2 – Talent poaching

Key staff have left DeepSeek:

罗福莉：V2 模型重要贡献者 → 跳槽小米负责 AI 业务
郭达雅：核心研究员 → 跳槽字节跳动

Salary ranges for AI talent are high (large‑model engineers ¥900k‑2 M+, AI trainers ¥300k‑800k, AI ethics reviewers ¥400k‑1 M). External capital is needed to remain competitive with tech giants.

Reason 3 – Strong competitors

OpenAI – valuation $852 billion, financing $40 billion

Anthropic – valuation $380 billion, financing $30 billion

DeepSeek – valuation $10 billion, planned financing $300 million

Stanford’s 2026 AI Index reports a performance gap of only 2.7 percentage points between the top U.S. models and China’s best, but closing that gap requires increasingly costly compute.

Reason 4 – V4 model launch

The next flagship V4, originally scheduled for February 2026, has been delayed to the end of April 2026. V4 will abandon Nvidia GPUs and adopt Huawei Ascend chips, a shift that demands substantial funding. Nvidia CEO Jensen Huang warned that this move would be a “bad result” for the United States.

Valuation context

OpenAI: $852 billion
Anthropic: $380 billion
DeepSeek: $10 billion

Given DeepSeek’s lack of prior external financing and the market success of the R1 model, the $10 billion valuation is considered attractive relative to U.S. peers.

Planned use of funds

Compute expansion – 40%

V4 model development – 30%

Talent compensation – 20%

Commercialization rollout – 10%

IPO timeline

2026 May‑June: financing closes
End 2026‑2027: IPO filing (preferred STAR Market, Hong Kong secondary option)

Conclusion

The raise is not driven by a cash shortage but by a strategic pivot: DeepSeek moves from a “small‑and‑beautiful” independent model to competing with global AI giants. The influx of capital will increase pressure from investors, accelerate profit expectations, and potentially speed up R&D cycles, marking the start of a capital‑driven competition era for Chinese large‑model AI.