DeepSeek seeks outside capital while keeping its open-source bet

DeepSeek is taking outside money for the first time, and the pitch is unusually blunt: the company wants investors to back AGI research and open models, not a quick hit of product revenue. According to Bloomberg, the lab is seeking at least $300 million, with a target valuation that could reach about 70 billion yuan, or around $10 billion.

That matters because DeepSeek built its reputation by doing the opposite of many AI rivals: staying lean, relying on internal capital, and pushing research-first releases that made more noise than sales decks ever could. Now the math is catching up with the mission. Training frontier models is expensive enough that even a profitable quant shop can start to look small.

A first external round for DeepSeek

The company had been funded entirely through High-Flyer Quant, the investment firm founded by Liang Wenfeng. That arrangement helped DeepSeek keep tight control over its roadmap, but it also meant the lab was effectively spending from one pocket while trying to build one of the world’s most ambitious AI programs.

Some sources suggest the raise could eventually scale much higher, potentially reaching $7 billion over the long term if DeepSeek shifts toward a more durable revenue model. For now, though, the headline number is the smaller one: a first institutional round that signals the research lab is entering the adult section of AI financing.

Recommended reading

Sam Altman ridicules Anthropic ad as backlash grows

Open-source models stay central to the pitch

Liang Wenfeng has reportedly told potential investors that AGI development will stay the priority, with monetization taking a back seat. DeepSeek is also sticking with open-source models, which is the kind of statement that plays well with researchers and less well with anyone hoping for a neat enterprise software business.

That stance is not new. In April, DeepSeek released its V4-Pro and V4-Flash family, mixture-of-experts systems with trillion-scale parameters tuned for both Chinese and overseas hardware, including Huawei Ascend, Cambricon, and Nvidia. The company already made a splash earlier with R1, which helped reset expectations around how much it should cost to train competitive frontier models.

What investors get, and what they don’t

• Minimum target raise: $300 million
• Target valuation: about 70 billion yuan, or around $10 billion
• Funding history: first external round; previously backed only by High-Flyer Quant
• Product stance: continued release of open-source models and emphasis on fundamental research

There is, predictably, a catch. The deal’s investor list, final valuation, and closing timeline are still undisclosed, and any raise of this size will invite scrutiny from Chinese regulators, who have been tightening oversight of foundation-model developers. The likely outcome is a more institutional DeepSeek, not a softer one.

The real question now is whether DeepSeek can keep its research identity while accepting the slower, messier logic of outside capital. If it can, it may become the model for how China’s most ambitious AI labs scale without surrendering their open-source reflex. If it cannot, the market will get another reminder that AGI talk is easy; paying for it is the hard part.