Liang Wenfeng

Early Life and Education

Liang Wenfeng was born in 1985 in Mililing village, Qinba town, Wuchuan, in Zhanjiang, Guangdong; both his parents were local primary school teachers. He attended Wuchuan No. 1 Middle School for junior and senior high school and in 2002 entered Zhejiang University to study electronic information engineering. After graduating in 2007 he stayed for graduate study in information and communication engineering, earning a master's degree in 2010 with a thesis on object-tracking algorithms for low-cost PTZ cameras.

Quantitative Investing and High-Flyer

From 2008 Liang led a team exploring fully automated quantitative trading using machine learning. In 2013 he co-founded Hangzhou Yakebi Investment Management with his Zhejiang University classmate Xu Jin, and in 2015 the quantitative hedge fund High-Flyer was formally established. In October 2016 High-Flyer launched its first AI model, executing trading positions generated by deep learning, and by 2017 it said its investment strategies were fully AI-driven. In 2019, when its assets under management passed 10 billion yuan, Liang gave a rare public speech, "The future of quantitative investment in China through a programmer's eyes", saying High-Flyer's mission was to improve the efficiency of China's secondary market.

High-Flyer went on to build its own AI computing capacity: the Fire-Flyer 1 supercomputer, which entered service in 2020, cost more than 100 million yuan, and Fire-Flyer 2, built with an investment of 1 billion yuan in 2021, was equipped with 10,000 Nvidia A100 chips. In 2021 High-Flyer became the first Chinese quantitative fund to manage more than 100 billion yuan, but after a sharp drawdown that year it closed to new investors and issued a public apology at the end of December. In 2022 a High-Flyer employee using the online name "an ordinary little pig" donated 138 million yuan to charity, and reports at the time said the donor was Liang.

Founding DeepSeek

In July 2023 High-Flyer announced the founding of DeepSeek to pursue artificial general intelligence, with Liang as chief executive. In May 2024 DeepSeek-V2 was released at 1 yuan per million input tokens and 2 yuan per million output tokens, prompting ByteDance, Alibaba, Baidu and others to cut their own model prices. "China should gradually become a contributor rather than always free-riding," Liang said in an interview.

In December 2024 DeepSeek released its V3 model, whose technical report put the full training cost at about $5.576 million. On 20 January 2025 it released the reasoning model DeepSeek-R1 as open source under the MIT licence; that afternoon Liang spoke as an entrepreneur representative at a symposium chaired by Premier Li Qiang on the draft government work report. A week later DeepSeek's app topped Apple's US App Store and Nvidia lost a record $593 billion in market value in a single day. Liang himself rarely appears in public or gives interviews.