Liang Wenfeng
China's AI disruptor
Liang Wenfeng is the founder of DeepSeek, China's disruptive AI bot developed at a fraction of the cost of others, shaking up the world of AI.
Liang Wenfeng was born in 1985 in Zhanjiang, Guangdong, China. His parents were both primary school teachers, fostering an environment that valued education. He pursued his higher education at Zhejiang University, one of China’s premier institutions for science and technology. He obtained a Bachelor of Engineering in Electronic Information Engineering and later a Master of Engineering in Information and Communication Engineering. His master’s dissertation focused on target tracking algorithms based on low-cost PTZ cameras.
High-Flyer Hedge Fund
After university, Liang initially ventured into finance. In 2015, he co-founded the quantitative hedge fund High-Flyer with classmates from Zhejiang University. High-Flyer quickly became one of China’s leading quant funds, using AI to make smart investment decisions. By 2021, it had over 100 billion yuan ($14 billion) in assets under management.
At the same time, Liang began scooping up thousands of GPUs from Nvidia while running High-Flyer, Financial Times reported, with one of his business partners describing him to the publication as a “very nerdy guy with a terrible hairstyle talking about building a 10,000-chip cluster to train his own models.”
DeepSeek AI
In 2023, with the global AI boom, Liang announced a pivot to AI and founded DeepSeek. DeepSeek is known for its cost-effective open-source large language models (LLMs). The company’s flagship model, DeepSeek-R1, rivals leading models like those from OpenAI but at a fraction of the cost. DeepSeek’s approach involves using legacy chips and innovative engineering designs to achieve high performance without the need for cutting-edge hardware.
In a paper released in late December 2024, DeepSeek researchers estimated they built and trained the model using 2,000 Nvidia H800 chips at a cost of under $6 million, significantly less than many of its AI competitors.
X owner Elon Musk, for example, has said his platform’s AI chatbot Grok 3 is training on 100,000 of Nvidia’s H100 GPUs. Meta CEO Mark Zuckerberg said last January that the company would purchase 350,000 Nvidia H100 GPUs by the end of 2024.
DeepSeek then stunned Silicon Valley again with the launch of its R1 model on January 20, 2025.
DeepSeek’s breakthrough in developing AI models at low cost has been a game-changer. The company claims to have built its AI model for less than $6 million, significantly lower than the costs incurred by competitors. This has been achieved through techniques like pruning, quantization, and knowledge distillation, which help in reducing the computational resources required.
DeepSeek has garnered attention from the Chinese government, which sees the company as a key player in advancing China’s AI capabilities. The company aligns with China’s policy of fostering innovative companies under the “new productive forces” initiative. While most Chinese tech companies receive heavy subsidies, DeepSeek is funded by Liang’s hedge fund, High-Flyer.
DeepSeek aims to commercialize its AI models by making them available to various industries. The company has already seen adoption by companies like Lenovo, UBTech, and Geely, which have integrated DeepSeek’s models into their products. DeepSeek’s open-source commitment means that its AI models, algorithms, and training details are freely available for use, modification, and further development by the global community.
Liang Wenfeng’s journey from a student at Zhejiang University to the founder of a groundbreaking AI company is inspiring. His innovative approach to AI development and commitment to open-source principles have made DeepSeek a significant player in the global AI landscape.
Interviews
Liang Wenfeng has remained low-profile, granting interviews only to Anyong, a sub-brand of China’s commercial tech media 36Kr, in 2023 and 2024. Below are translated excerpts from these interviews, offering a glimpse into his philosophy and vision.
DeepSeek’s ‘long-termism’ …
For Liang, DeepSeek is more like a side project or hobby, driven by deep curiosity and a commitment to foundational research. He acknowledges that basic research often yields low immediate returns on investment, yet he is captivated by the challenge of exploring complex fields like finance and the potential of artificial general intelligence (AGI). Liang’s focus is on understanding the essence of human intelligence and the processes that underlie it, believing that such exploration is crucial despite the lack of immediate commercial incentives.
「人类智能本质可能就是语言，人的思维可能就是一个语言的过程。你以为你在思考，其实可能是你在脑子里编织语言。这意味着，在语言大模型上可能诞生出类人的人工智能（AGI）。」
“The essence of human intelligence might be language; human thought could be a linguistic process. You think you’re thinking, but you might actually be weaving language in your mind. This implies that human-like artificial intelligence (AGI) could emerge from large language models.”
「当时我们尝试了很多场景，最终切入了足够复杂的金融，而通用人工智能可能是下一个最难的事之一，所以对我们来说，这是一个怎么做的问题，而不是为什么做的问题」
「如果一定要找一个商业上的理由，它可能是找不到的，因为划不来。从商业角度来讲，基础研究就是投入回报比很低的。」
Talent and team building …
DeepSeek’s LinkedIn profile shows that the company has a team of fewer than 10 people. One member was reportedly poached by Xiaomi’s Lei Jun to work on AI development in December, 2024. Liang believes in discovering talent within China.
「如果追求短期目标，找现成有经验的人是对的。但如果看长远，经验就没那么重要，基础能力、创造性、热爱等更重要。从这个角度看，国内合适的候选人就不少。」
「因为我们在做最难的事。对顶级人才吸引最大的，肯定是去解决世界上最难的问题。其实，顶尖人才在中国是被低估的。因为整个社会层面的硬核创新太少了，使得他们没有机会被识别出来。我们在做最难的事，对他们就是有吸引力的。」
On Innovation …
Innovation requires freedom and room for trial and error. He noted that innovation often emerges naturally, rather than being planned or taught.
「我们的总结是，创新需要尽可能少的干预和管理，让每个人有自由发挥的空间和试错机会。创新往往都是自己产生的，不是刻意安排的，更不是教出来的。」
「创新就是昂贵且低效的，有时候伴随着浪费。所以经济发展到一定程度之后，才能够出现创新。很穷的时候，或者不是创新驱动的行业，成本和效率非常关键。看OpenAI也是烧了很多钱才出来。」
On China’s role in AI development …
Liang believes that China cannot remain a follower in AI forever. In the interviews, he emphasizes the need for China to shift from imitation to originality and build its own technological ecosystem.
「我们看到的是中国AI不可能永远处在跟随的位置。我们经常说中国AI和美国有一两年差距，但真实的gap是原创和模仿之差。如果这个不改变，中国永远只能是追随者，所以有些探索也是逃不掉的。」
「英伟达的领先，不只是一个公司的努力，而是整个西方技术社区和产业共同努力的结果。他们能看到下一代的技术趋势，手里有路线图。中国AI的发展，同样需要这样的生态。很多国产芯片发展不起来，也是因为缺乏配套的技术社区，只有第二手消息，所以中国必然需要有人站到技术的前沿。」