DeepSeek Seeks First Outside Funding at $10 Billion Valuation

DeepSeek, the Chinese AI lab that shook the industry with its low-cost R1 reasoning model in January 2025, is in talks to raise at least $300 million in outside capital for the first time, according to a report by The Information. The round would value the Hangzhou-based company at $10 billion or more.

The fundraise marks a sharp reversal for founder and CEO Liang Wenfeng, who built DeepSeek as a side project of his quantitative hedge fund, High-Flyer Capital Management, and repeatedly turned down offers from major Chinese venture capital firms and tech giants. High-Flyer, which holds 99% ownership of DeepSeek, had until now been the company’s sole source of funding – a deliberate strategy Liang framed as insulating the lab from commercial pressure.

That independence is now under strain from two directions: rising costs and a thinning talent bench.

Key Engineers Defect to Rivals

DeepSeek has lost several high-profile researchers in recent months. Luo Fuli, a core developer of the V2 model, left to join Xiaomi in November 2025 to lead the company’s MiMo spatial intelligence team. More recently, Guo Daya – a lead author on the R1 model – departed for ByteDance’s Seed AI division, according to the South China Morning Post. ByteDance denied reports that Guo’s compensation package was worth up to 100 million yuan (roughly $14 million) annually, but confirmed it follows a standard framework that includes equity.

The departures reflect an intensifying talent war in China’s AI market, where well-funded rivals like Alibaba, ByteDance, and Tencent are aggressively poaching from smaller labs. Wang Bingxuan, another DeepSeek engineer, reportedly left for Tencent during the same period.

V4 Delays and the Huawei Chip Gamble

DeepSeek’s next flagship model, V4, has been pushed back multiple times. The delays stem partly from efforts to make V4 compatible with Huawei’s Ascend AI chips – a technically challenging migration that involves overcoming slower inter-chip connectivity and less mature software tooling compared to Nvidia’s CUDA ecosystem.

The V4 model is being engineered to run on Huawei chips as part of Beijing’s push to reduce China’s reliance on American semiconductors. If DeepSeek succeeds, the move would represent a meaningful milestone for domestic chip development. But the technical hurdles are real – Huawei engineers have reportedly been embedded with DeepSeek’s team to troubleshoot stability problems during training runs.

Meanwhile, competitors are not waiting. Alibaba’s Qwen family of open-source large language models now commands over 50% of global open-source model downloads, with more than 940 million cumulative downloads as of March 2026. ByteDance’s Doubao chatbot crossed 100 million daily active users by the end of 2025, and the company has stated its aim to become at least the third-largest player in generative AI globally. The latest Stanford AI Index shows the global AI development race accelerating, with China narrowing the gap to roughly three to six months behind leading U.S. companies.

At $10 billion, DeepSeek’s proposed valuation sits well below the stratospheric figures seen elsewhere in AI – Anthropic recently attracted offers at $800 billion, and even infrastructure plays like Databricks have reached $134 billion. But it would be a significant step for a company that trained its V3 base model for a reported $5.6 million on 2,048 Nvidia H800 GPUs – a fraction of what Western labs spend.

The investors are expected to be primarily domestic Chinese firms, as U.S. venture capital faces regulatory and national security barriers to investing in Chinese AI. No deal has closed, and it remains unclear whether Liang will accept terms that dilute High-Flyer’s near-total control. What is clear is that the days of running a frontier AI lab on a hedge fund’s balance sheet alone appear to be ending – even for the company that proved it was possible.