Liang Wenfeng, the controlling shareholder of DeepSeek, has been pursuing AI for years and has consistently pushed for the application of this technology in business.
On January 20, the Chinese startup DeepSeek announced the release of its free V3 model and the R1 model, which is “trained on less advanced Nvidia chips, 100% open-source, and 96.4% cheaper than OpenAI’s o1 while delivering similar performance.” This announcement caused panic among U.S. tech companies and the stock market, drawing attention to the man behind DeepSeek.
Chinese media reports indicate that this AI startup is headquartered in Hangzhou. The controlling shareholder of the company is Liang Wenfeng, a 40-year-old graduate with a degree in AI from Zhejiang University.
Liang co-founded the quantitative hedge fund High-Flyer in 2015, which gained fame for its innovative application of AI-based trading strategies. In 2023, High-Flyer Quant announced its focus on building a specialized branch to research artificial general intelligence (AGI). The fund also holds numerous patents related to chip clusters used for training AI models.
Liang is said to have started purchasing thousands of Nvidia graphics chips as early as 2021 for AI projects, before the Biden administration imposed export restrictions on advanced chips to China. In 2022, High-Flyer’s AI division claimed to own and operate approximately 10,000 Nvidia A100 chips.
In an interview in July 2024, Liang revealed his ambitions for DeepSeek and China’s broader AI strategy. “Chinese businesses have long been accustomed to using technological innovations developed abroad and profiting from their applications. This is not sustainable. Our goal is not quick profits but to expand the boundaries of technology and drive the development of the ecosystem,” he said.
One of Liang’s business partners told the Financial Times: “The first time we met, he seemed like a very bookish guy with a terrible haircut, talking about building a 10,000-chip cluster to train his own model. We didn’t pay much attention. He couldn’t articulate his vision beyond saying, ‘I want to build this, and it will be a game-changer.’ We thought such things could only be done by giants like ByteDance and Alibaba.”
Chinese leaders are also paying attention to DeepSeek. Before the launch of the R1 model, Liang was invited to a closed-door industry seminar chaired by Chinese Premier Li Qiang on January 20.
Experts believe DeepSeek is not currently at risk of being targeted by U.S. government sanctions. “They haven’t released a commercial version or a paid plan yet, so they wouldn’t face direct financial impact if restricted by the U.S. They haven’t taken external investment, and all funding comes from the founder’s hedge fund, so there’s no pressure from shareholders to exit,” said Phelix Lee, an analyst at Morningstar.
Since DeepSeek uses open-source technology, the most likely scenario is that U.S. companies could use the publicly available code from the Chinese firm to refine their own models. “This could help reduce global AI computing costs,” Lee added.
