- DeepSeek is raising at least $300 million in its first-ever outside funding round at a valuation exceeding $10 billion, The Information reported Wednesday.
- The Chinese AI lab has been entirely funded by parent hedge fund High-Flyer Capital Management since its 2023 founding—after turning down multiple offers from China’s top VCs.
- The fundraise comes as reasoning models and agentic AI drive capital requirements higher across the industry, even for the company once celebrated for doing more with less.
DeepSeek—the Chinese AI lab that wiped $600 billion off Nvidia’s market cap with a $6 million model—is finally asking for outside money. The company is in talks to raise at least $300 million at a valuation starting at $10 billion, according to The Information, citing four people familiar with the negotiations.
Until now, DeepSeek has operated entirely on capital from High-Flyer Capital Management, the Chinese quantitative hedge fund that birthed it. Founded in July 2023 by Liang Wenfeng—a quant trader who’d been running AI-driven strategies since the 2008 financial crisis—DeepSeek had consistently declined investment offers from China’s most prominent venture capital firms and technology conglomerates.
The timing is telling. When DeepSeek launched its R1 model in January 2025, the narrative was simple: a scrappy Chinese outfit built a system competitive with OpenAI’s GPT-4 for roughly $6 million—a fraction of the $100 million GPT-4 reportedly cost. The release triggered a selloff in US tech stocks and sparked an uncomfortable debate about whether American AI companies were overcapitalized. One year later, DeepSeek is raising capital like everyone else.
Why DeepSeek Needs Outside Cash Now
The math has changed. Building frontier AI models in 2026 costs dramatically more than it did in 2024, even with DeepSeek’s efficiency-first approach. The rise of reasoning models—think chain-of-thought systems that burn compute on every query—and agentic AI applications that run autonomously for extended periods have pushed capital requirements industry-wide. Training a competitive model is one thing; running inference at scale for millions of users is another.
Reuters reported earlier this year that DeepSeek trained one of its newest models on Nvidia’s most advanced chips—despite those chips being subject to US export restrictions on sales to China. The company also declined to share its flagship model with US chipmakers for performance optimization. China’s push for domestic chip adoption adds another layer of cost complexity.
There’s also the geopolitics problem. Because DeepSeek is a Chinese company, US venture capital firms are expected to approach the round cautiously, The Information noted. That leaves the pool of willing investors narrower than a $10 billion valuation might suggest. Chinese VCs and sovereign wealth funds are the most likely participants—but DeepSeek already turned those parties down once.
The Open-Source AI Darling Tries to Monetize
DeepSeek’s position in the AI ecosystem is unusual. Its models are open-weight—meaning the parameters are freely shared, though training data isn’t openly licensed—and the company has released new models under the MIT License since R1. That openness earned it a devoted following among researchers and developers who see it as a counterweight to the increasingly closed strategies of OpenAI and Anthropic.
But open-weight and free don’t pay for GPU clusters. The fundraise suggests DeepSeek is preparing for a phase where it needs to do more than publish impressive benchmarks. Running a competitive API service, scaling infrastructure, and retaining top talent in a cutthroat market all require capital that a hedge fund parent—even one as successful as High-Flyer—may not want to indefinitely bankroll.
Liang Wenfeng reportedly recruited AI researchers from top Chinese universities and amassed 10,000 Nvidia A100 GPUs before US export restrictions took effect. That hardware stockpile gave DeepSeek a head start, but GPUs depreciate, and the next generation of training runs will demand more. The $300 million floor is almost certainly just the opening ask.
High-Flyer’s own trajectory offers context. By 2021, the fund was using AI exclusively for trading. Its Fire-Flyer 2 computing cluster logged 56.74 million GPU hours in 2022, with 27% allocated to external scientific computing. The hedge fund’s appetite for AI infrastructure has been voracious—and expensive.
DeepSeek’s R1 model is available on Hugging Face.
