Founded in 2023, DeepSeek has quickly emerged as a challenger to U.S. global AI leadership despite stringent American sanctions. An in-depth investigation by RFA Cantonese reveals that DeepSeek had stockpiled a large number of Nvidia chips before the U.S. semiconductor ban came into effect, and utilized Meta’s free “Llama” large language model for research and training. The report also found its funding company, which recruited a team of young tech elites from top universities in China, may have received government grants. Its founder, Liang Wenfeng, was a guest at a forum hosted by Chinese Premier Li Qiang, signaling DeepSeek’s close ties with the ruling Chinese Communist Party.
Just as U.S. President Donald Trump is preparing for a new round of chip sanctions against China and making significant investments in the U.S. AI sector, China’s DeepSeek made a major breakthrough and sent shock waves through Silicon Valley.
Founded in 2023, DeepSeek has earned an official “reputation” in China as the “Price Chopper,” known for its low cost yet high efficiency. DeepSeek spent $6 million in development and training, less than 5% of that of the other larger AI models, according to public data. Using merely 2,000 Nvidia chips, DeepSeek’s performance rivals that of ChatGPT in two months. DeepSeek tops the App Store charts in the U.S. and China, threatening America’s edge in advanced AI technology.
How did this startup rise so rapidly while its Chinese peers struggle under U.S. chip and high-tech sanctions?
From engineer to fund manager
DeepSeek’s 40-year old founder Liang Wenfeng set up a hedge fund applying quantitative trading strategies and is dubbed “King of Quantitative Analysis” in China. With a master’s degree in Information and Communication Engineering from Zhejiang University, Liang co-founded a Hangzhou-based asset management company with his classmate Xu Jin in 2013. The two went on to establish “High-Flyer Quant,” a AI-driven hedge fund and the primary backer of DeepSeek, in 2015. By 2021, High-Flyer managed assets worth more than 100 billion Chinese yuan and became one of the top four algorithm-driven quant funds in the country.
According to reports from state media including Xinhua and People’s Daily, Liang shared his suggestions on a government work report when attending a forum hosted by Premier Li Qiang on January 21. According to Mainland China Business Verification Network, High-Flyer’s subsidiary “Hangzhou High-Flyer Technology Co.” is recognized as a national high-tech enterprise, making it eligible for significant tax incentives and R&D subsidies.
Significant funds raised in 2023
High-Flyer apologized for its poor performance and stopped raising funds in 2021. However, two years later, Liang established DeepSeek focusing on AI research. In the same year, Liang and High-Flyer returned to the public spotlight after a Chinese state media report on a “mystery” philanthropist who donated 580 million yuan to charities. Over the past two years, High-Flyer has collaborated with organizations such as the China Youth Development Foundation and the Chinese Communist Youth League to launch a scholarship program, donating hundreds of millions of yuan and revealing its strong finances at the same time.
Hoarding chips before U.S. sanction
Liang’s biggest obstacle in AI development was not funding but chips. Chinese media 36Kr reported that before the U.S. imposed chip ban against China in 2023, Liang’s company had already purchased over 10,000 Nvidia A100 chips, though AI research consultancy SemiAnalysis founder Dylan Patel estimated the figure to be at least 50,000. This significant purchase enabled DeepSeek to develop models by integrating low-power chips.
Built on Meta’s model
Both Meta AI Chief Scientist Yann LeCun and Open AI CEO Sam Altman have said DeepSeek’s success does not imply that China has surpassed the U.S. in the AI sector. The Chinese company simply leveraged existing open-source large language models, including Meta’s Llama, showing “open source models are surpassing proprietary ones.” In addition to the chip war, the sharing of AI big data will become another worldwide battleground, analysts say.
To read the original story in Chinese, click here.
