Liang Wenfeng: Building Open AI from Quantitative Roots

Liang Wenfeng, an entrepreneur with a background in applying machine learning to financial markets, is the driving force behind DeepSeek, an artificial intelligence company specializing in open-weight large language models.

Foundations in Finance and AI

Liang’s professional path began with a strong emphasis on integrating advanced computational methods into finance. After completing his studies at Zhejiang University, he dedicated his early career to exploring how machine learning could enhance quantitative finance. This pursuit led him to co-found Hangzhou Yakebi Investment Management Co Ltd in 2013, focusing on the synergy between artificial intelligence and quantitative trading. He further expanded his involvement in the sector by co-founding Hangzhou Huanfang Technology Co Ltd in 2015, which later became Zhejiang Jiuzhang Asset Management Co Ltd.

A significant milestone in his career was the co-founding of Ningbo High-Flyer Quantitative Investment Management Partnership in 2016. This firm utilized mathematics and AI to inform investment strategies. By 2019, Liang established High-Flyer AI, an entity dedicated to fundamental research in AI algorithms and their practical applications. During this period, High-Flyer managed substantial assets. Liang articulated his perspective on quantitative investment, emphasizing that decisions should be driven by quantitative methods rather than human discretion, with a stated aim to enhance the efficiency of China’s secondary market. He also drew inspiration from the work of prominent figures in quantitative finance, reflecting on their philosophies when facing professional challenges.

The Genesis of DeepSeek

The computational infrastructure developed at High-Flyer for its financial operations proved instrumental in a pivotal strategic shift. In 2021, Liang began investing in thousands of advanced graphics processing units, envisioning a transformative AI project. This initiative culminated in May 2023 with the launch of DeepSeek, an endeavor by High-Flyer to advance artificial general intelligence. Prior to international restrictions on AI chips, DeepSeek had already secured a significant number of high-performance GPUs, establishing a robust foundation for its large language model development.

DeepSeek operates with a distinctive funding model, receiving financial backing from High-Flyer. This structure provides the company with a unique advantage, shielding it from typical commercial pressures and rigid performance metrics often imposed by venture capital firms. This financial independence allows DeepSeek the flexibility to explore and deviate from conventional model architectures, fostering an environment for innovative development. Liang’s personal stake in DeepSeek is a small percentage, with the majority held by High-Flyer, underscoring the strategic alignment between the two entities.

Shaping the AI Landscape

DeepSeek’s impact on the AI sector became evident with the release of its models. The DeepSeek-R1 model, for instance, garnered international attention for its impressive performance while utilizing relatively modest computational resources. A subsequent release, DeepSeek-V2, unexpectedly ignited a pricing competition within China’s AI market. Liang’s aggressive pricing strategy prompted other major domestic technology companies to significantly reduce their own model rates.

Liang holds a perspective that China’s economic progression necessitates a transition from simply adopting innovations to actively contributing to them. He believes the challenge lies not in capital availability, but in cultivating the confidence and expertise to effectively organize talent for innovation. DeepSeek’s hiring philosophy reflects this, prioritizing individuals with inherent ability and passion, often recruiting locally educated talent, rather than focusing solely on prior experience. The company fosters a flat organizational structure and project-based teams, aiming to attract and retain skilled professionals by offering competitive compensation and an environment that encourages individual initiative. This bottom-up approach, where employees are encouraged to bring their unique insights, directly led to breakthroughs like the multi-head latent attention training architecture, which substantially reduced the training costs for the DeepSeek-V3 model.

Industry Recognition and Influence

Liang Wenfeng’s contributions to the field have earned him significant recognition. In 2025, he was acknowledged by Time magazine on its list of the 100 Most Influential People in AI and was also named to Fortune’s list of the Most Powerful People in Business. His expertise is valued at a high level, as evidenced by his invitations to participate in important national symposiums in Beijing in 2025. During these gatherings, he provided insights and suggestions on government work reports and discussed matters pertaining to private-sector enterprises. Observers often note his distinct capacity for learning and skill development, alongside a personality frequently described as geeky.