DeepSeek, the Chinese lab that rattled Silicon Valley, is raising outside money for the first time at a $10B valuation

After two years of turning down offers, the Hangzhou AI lab behind the $6 million R1 model is finally opening up – targeting at least $300 million.

DeepSeek, the Hangzhou-based AI lab whose R1 model rattled Silicon Valley last year, is raising outside money for the first time since it was founded in 2023.

According to a Reuters report cited by Tech Funding News, the Chinese startup is targeting at least $300 million at a valuation of more than $10 billion – its first external round after two years of turning down venture capital firms and major tech companies.

The pitch is simple. DeepSeek has spent the past two years proving that a small Chinese lab can match the frontier labs in the United States on a fraction of the budget – and it has done so while refusing every outside check that came its way.

Until now.

The lab that ran on hedge fund money

DeepSeek was spun out of High-Flyer, a Chinese quantitative hedge fund co-founded by Liang Wenfeng, in July 2023. High-Flyer still owns 99% of the company, and it has single-handedly bankrolled DeepSeek’s research since day one.

That arrangement has held up better than most investors might have guessed. High-Flyer reportedly posted a 56.6% return in 2025, according to Bloomberg, and those gains have funded DeepSeek’s compute, talent, and training runs without a single outside dollar.

It is an unusual setup for an AI lab of this scale – one that effectively turned a hedge fund’s trading profits into frontier AI research, and then gave the results away for free.

DeepSeek’s models are fully open-source, released under the MIT license.

The $6 million model that moved the market

The reason investors are now circling is R1.

Released in January 2025, DeepSeek’s R1 reasoning model matched OpenAI’s top model on key benchmarks but reportedly cost only $6 million to train. Rather than the expensive supervised fine-tuning favored by Western labs, R1 leans on a mixture-of-experts design and reinforcement learning – a setup that usually activates only 37 billion of the model’s 671 billion parameters at any one time.

The economics that fall out of that design are hard to ignore. Running the model costs about $0.55 per million input tokens, roughly 96% less than OpenAI’s equivalent.

The launch reshaped the conversation in both directions. Since R1 shipped, OpenAI, Anthropic, Google, and Meta have all accelerated their own reasoning model projects in response, while the Western narrative that frontier AI requires hundreds of millions in training compute has quietly softened.

Why raise now?

For a company that has been funded entirely by its parent’s trading book, the shift to outside capital is a real pivot – not a curtain call for the High-Flyer era, but a new chapter.

Liang has previously said that China’s AI industry needs “confidence as much as capital,” and now he is looking for both from the market. The $10 billion-plus valuation being floated would put DeepSeek well below the Western frontier labs it competes with on performance, but well above nearly every other Chinese AI startup.

The funding plan, first reported by Reuters, has since been picked up by Bloomberg, Yahoo Finance, Unite.AI, and other outlets. DeepSeek has not publicly confirmed the round.

For anyone watching the AI race, the signal matters more than the number. A lab that built one of the most influential models of the last two years on hedge fund profits is now willing to sit across the table from outside investors – and the market, for the first time, gets to decide what that work is worth.