WHEN DEEPSEEK, the Chinese generative artificial intelligence (GenAI) model, debuted in late January, few could have predicted the impact it would have on the global AI landscape. Developed at a fraction of the cost of its U.S. counterparts, DeepSeek-R1 has not only surpassed OpenAI’s ChatGPT in downloads but also sent Wall Street into a tailspin. On January 27, the tech-heavy Nasdaq Composite plummeted by 3.1%, the S&P 500 slid 1.8%, and tech giant Nvidia, the dominant force in AI hardware, took a major hit. By January 29, the Nvidia stock had shed a whopping $130 billion in market value.

What makes DeepSeek’s rise so extraordinary? For the longest time, it was believed that to develop a state-of-the-art GenAI model, one would require high-end GPUs — that can perform complex calculations on massive data sets simultaneously and help ‘train’ the software — manufactured by the likes of Nvidia, which led development costs shooting up to billions of dollars. And Nvidia’s stock skyrocketed. But DeepSeek showed that one can develop a GenAI model through a combination of cost-efficiency, open-source accessibility, and technological prowess, under $6 million.

At the heart of DeepSeek’s rise is Liang Wenfeng, its founder. A computer scientist-turned-AI-visionary, Liang’s ambition is not rooted in financial gain but in the pursuit of knowledge. “I wouldn’t be able to find a commercial reason [for founding DeepSeek] even if you asked me to,” he confessed to Chinese tech publication 36Kr.

DeepSeek’s origins can be traced to High-Flyer, one of China’s most successful quantitative hedge funds. Founded in 2015, High-Flyer stockpiled GPUs for financial data analysis. Recognising the opportunity, Liang repurposed these for AI research. Rather than hiring seasoned engineers to develop consumer-facing products, Liang sought out the brightest PhD students from China’s top institutions, fostering a research-driven culture. While many AI firms guard their models closely, DeepSeek has chosen transparency, allowing researchers to access and build upon its innovations.

DeepSeek faced other obstacles. U.S. export controls imposed in 2022 restricted Chinese firms’ access to advanced GPUs. But, DeepSeek’s engineers devised ingenious methods to optimise their models, and managed to create a system requiring only one-tenth the computing power needed by Meta’s Llama 3.1. “It (DeepSeek) has shown that instead of solely relying on high-end computing, one can think differently and focus on efficiency,” says Varun Mishra, senior analyst, Counterpoint Research.

DeepSeek’s ascent has sent ripples across financial markets. U.S. tech investors have started reallocating assets; Treasury yields have dropped; and the dollar has weakened against safe-haven currencies like the yen.

DeepSeek-R1

How DeepSeek is challenging the dominance of U.S. AI companies