Liang Wenfeng: DeepSeek Founder in Nature’s 2025 Top 10
Liang Wenfeng, the 40-year-old founder and CEO of Chinese AI startup DeepSeek, landed a spot on Nature’s annual “Top 10: People who shaped science in 2025” list. Nature called him a “Chinese finance whizz” whose work shook up the AI world. Yahoo News, drawing from South China Morning Post, detailed how his models challenged ideas about U.S. dominance in AI.
From Finance to AI
Liang started in finance. He co-founded High-Flyer Quantitative Fund, one of China’s biggest quant trading funds. DeepSeek spun out from there as its AI lab. The company stays low-profile, funded internally by High-Flyer, and partners with Chinese hardware makers like Moore Threads to build AI chips despite U.S. export curbs.
High-Flyer scored big lately too. It bought into Moore Threads’ Shanghai IPO and saw shares jump over fivefold, netting about 40 million yuan ($5.6 million) profit after two days of trading, per calculations in Yahoo Finance.
Reasons for Nature’s Pick
Nature spotlighted DeepSeek’s January 2025 release of R1, a reasoning model. It hit markets hard: On January 27, tech stocks dropped, erasing nearly $1 trillion total and $600 billion from Nvidia alone. R1 proved powerful AI could come cheap, rivaling OpenAI while costing far less to train. For comparison, Meta’s Llama 3 405B took over 10 times more.
DeepSeek’s open-source approach helped researchers tweak models for their work. That pushed companies in China and the U.S. to release their own open models. Adina Yakefu, a Hugging Face researcher, told Nature DeepSeek proved “hugely influential.”
In September, DeepSeek engineers published R1’s training details in Nature itself—the first major large language model to get peer review there. Nature said this “recipe” showed others how to build reasoning models. Xinhua and CGTN both covered Liang alongside geoscientist Du Mengran, who found the planet’s deepest animal ecosystem.
Main Achievements
- DeepSeek-R1 (Jan 2025): Leading reasoning at low cost, sparked market chaos and AI rethink.
- DeepSeek-V3 (Dec 2025): Matches OpenAI and Google models in tests.
- Open-source releases: Boosted global research, inspired copycats.
- Nature paper (Sept 2025): Peer-reviewed LLM training method, first of its kind.
Liang shares the list with figures like a CDC director, a pandemic treaty negotiator, and a baby saved by CRISPR. Nature’s editors pick based on big stories, not rankings.
