Who is Liang Wenfeng, the quant who owns 81.6% of DeepSeek

If the $15 billion round reported this week closes anywhere near the valuations under discussion, Liang Wenfeng will be one of the richest people in China, and most people outside the industry would not recognise him in a lift. He does not give many interviews. He did not respond to this week’s reports. What is known about him comes from filings, a few long conversations with Chinese media, and the company he built.
Liang co-founded High-Flyer, a quantitative hedge fund that used machine learning to trade Chinese equities, and it was High-Flyer’s money and GPUs that financed DeepSeek’s early years. The AI company was set up in 2023 in Hangzhou, the same city as Alibaba, as what looked at the time like a research side project. According to WireScreen data cited by TechCrunch, Liang held an estimated 81.6 percent of the Chinese operating entity after an earlier funding round. That is an unusual degree of founder control for a company this large, and it means the investors now writing cheques, Tencent and CATL among them, are buying into his decisions rather than a board’s.
The moment that made the company famous came in early 2025, when DeepSeek released a free, open-weight reasoning model that matched the output of far more expensive American systems. Nvidia’s share price fell on the news, Washington began asking questions, and “DeepSeek moment” became shorthand for the realisation that China’s AI gap was months, not years.
Since then Liang has done something rarer than a viral model: he has built a business. The Information reported in September that he told investors the company’s annualised revenue run rate had reached $1 billion, more than double the level a few months earlier, almost all of it from businesses and developers paying for API access. He said raising prices had not “driven customers away.” DeepSeek has kept publishing its methods openly, including, in recent papers, the economics of serving its models, which is not what a founder protecting a moat normally does.
The next test is the hardest. The Information reported this week that Huawei engineers are embedded in DeepSeek’s offices and that the company plans to pre-train on Huawei’s Ascend chips by year end, a year after its first attempt to do so failed and it fell back on Nvidia. On October 20 the US government ordered federal agencies to drop DeepSeek’s software. Liang’s response, as far as anyone can tell, was to raise 100 billion yuan.
What to make of him: a quant who treats AI research the way he treated markets, as a problem of compute, data and patience, and who has so far declined every opportunity to become a public figure. In a sector built on founders who talk, that may be the strategy.
