Rise of DeepSeek
How a Trading Firm Unlocked Affordable Innovation, Disrupting the Billion-Dollar AI club
In the tech-transformed landscape of modern Guangdong, Liang Wenfeng’s story begins in the 1980s, when the region was still finding its technological footing. Born to an elementary school teacher in what would later become China’s silicon heartland, Liang’s path would mirror his homeland’s evolution from industrial center to tech powerhouse. After graduating from Zhejiang University, Liang ventured into an arena that would combine his passion for technology with the complexity of financial markets.
In 2015, along with two fellow Zhejiang University engineers, Liang founded High-Flyer. What started as a traditional quantitative hedge fund would evolve into something far more ambitious. The firm’s growth tells a compelling story of rapid innovation and success: from managing 1 billion yuan in 2016, High-Flyer expanded to oversee more than 10 billion yuan by 2019.
High-Flyer distinguished itself by integrating AI into every aspect of their operation, from data collection to trade execution. The firm’s early days of multi-factor price-volume models gave way to a broader embrace of machine learning by 2017, marking the beginning of a technological transformation that would ripple beyond China’s financial markets. They capitalized on market inefficiencies in China’s domestic stock market, particularly those resulting from emotional trading by retail investors, which they exploited through sophisticated quantitative strategies. It was here, at the intersection of technology and finance, that Liang would make his first significant mark — not as an AI researcher, but as an innovator who saw the transformative potential of AI in traditional industries.
But it was Liang’s bold move in 2022 that would truly set the stage for what was to come. In a prescient decision, High-Flyer acquired 10,000 Nvidia A100 GPUs, an investment of 1 billion Yuan that would power their Fire-Flyer II supercomputer. This massive computational backbone, secured just before U.S. chip restrictions took effect, would prove crucial for what followed. In April 2023, High-Flyer announced DeepSeek, a research initiative with ambitions reaching far beyond financial markets into the realm of artificial general intelligence. The venture quickly turned heads in the global AI community with its R1 language model, developed in just two months at a fraction of the cost typically associated with such projects — under $6 million, a figure that challenged conventional wisdom about the resources needed for AI innovation.
DeepSeek’s R1 model, with its 671 billion parameters, has shown remarkable capabilities, particularly in mathematical reasoning and software engineering tasks. It achieved a 97.3% score on MATH-500, surpassing OpenAI’s leading models, while maintaining competitive performance across various benchmarks. More importantly, its API costs are 96.4% cheaper than competitors, making advanced AI capabilities accessible to a broader range of users and developers. To put this achievement in perspective, traditional AI model development costs are staggering — GPT-3’s training alone cost $4.6 million, while Google’s PaLM required approximately $12.4 million. Most large-scale AI projects involve teams of 50–100 engineers working for months or years, with infrastructure costs running into millions.
DeepSeek’s success represents a paradigm shift in AI development, demonstrating that innovation isn’t solely about computational power but about efficiency and creative resource utilization. Their approach suggests a future where AI development becomes more decentralized, with significant contributions possible from outside traditional tech giants. This could reshape the global AI landscape, particularly benefiting smaller companies and developing nations by lowering the barriers to entry in advanced AI research and development.
