Who is Liang Wenfeng, the Designer of 'DeepSeek' That Surpassed Nvidia?

Liang Wenfeng, a Chinese engineer born in 1985
Released his first AI model in 2023... Unveiled R1 just over a year later
Nvidia's corporate value drops by $600 billion due to the DeepSeek shock

"DeepSeek R1 is the Sputnik moment in the field of artificial intelligence (AI)"

Mark Andreessen, a leading venture capitalist in Silicon Valley, described the emergence of the Chinese startup DeepSeek, which shocked Nvidia?dominating the AI semiconductor market with low-cost, high-efficiency AI models?on the social networking service X (formerly Twitter). The term "Sputnik moment" originates from the space race when the Soviet Union launched the Sputnik 1 satellite before the United States, which had been confident in its technological superiority. It refers to the moment when a country confident in its technological dominance is shocked by the advanced technology of a latecomer.

DeepSeek application screen. Photo by Reuters, Yonhap News

Liang Wenfeng, who managed 20 trillion won with an AI-based hedge fund

Liang Wenfeng (梁文鋒), the founder of DeepSeek who designed the AI model R1, was born in 1985 in Zhanjiang City, Guangdong Province. In 2002, he ranked first in the university entrance exam at Wuchuan No. 1 Middle School in Zhanjiang City and entered the prestigious engineering university Zhejiang University’s Department of Electronic Information Engineering. He loved mathematics so much that he self-studied college-level math since middle school, and this inclination naturally led to an interest in finance. In 2013, he founded "Jacobi Investment Management Co., Ltd." in Hangzhou with his university alumnus Xu Jin.

In 2015, Liang and his university classmates created the AI-based hedge fund "Highflyer," which used a method combining computer trading with deep learning. By 2021, Highflyer’s assets under management reached 100 billion yuan (approximately 20 trillion won).

Using the investment capital, Liang established an AI deep learning platform department within the company in 2019. There, they secured 10,000 Nvidia chips to build a cluster for training large language models (LLM). The cost was about 1 billion yuan (200 billion won). Subsequently, Highflyer spun off DeepSeek and began full-scale AI model development, releasing the first AI model, DeepSeek Coder, developed in an open-source manner, in November 2023. They then launched models such as DeepSeek Math, Math, DeepSeek, VL, V2, Coder2, V3, and last week, the R1 model.

Ryang Won-pung, founder of DeepSeek (front row, right) Photo by Yonhap News

The remarkable aspect of DeepSeek is that it achieved technology close to the performance of major American AI models at a low cost. DeepSeek revealed that the total development cost for R1 was $5.6 million (8.2 billion won), which is less than one-tenth of the typical $100 million (145 billion won) used by major U.S. tech companies for AI model training. DeepSeek’s research and development (R&D) staff is also about 180 people, roughly one-sixth the size of OpenAI’s development team.

DeepSeek’s corporate value

So, what is DeepSeek’s corporate value? Currently, the exact corporate value of DeepSeek is not publicly known. As an unlisted stock and a startup only one year old, it is difficult to assess its corporate value. However, some speculate that DeepSeek is preparing for a Series B funding round and that its corporate value could reach approximately $10 billion (14.54 trillion won).

If DeepSeek grows like OpenAI, its corporate value could increase exponentially. Recently, there have been projections that OpenAI, the developer of ChatGPT, could reach a corporate value of $340 billion (494.16 trillion won). OpenAI’s corporate value has more than doubled from $157 billion three months ago to $400 billion as it prepares to raise $40 billion in investment.

On the 28th, traders are working at the New York Stock Exchange on Wall Street in New York. Photo by Yonhap News.

The "DeepSeek shock" was significant. The New York stock market once fluctuated due to DeepSeek’s emergence. DeepSeek differentiated itself from Nvidia, which has supplied high-performance GPUs, by building AI with low cost and high efficiency. DeepSeek reportedly combined Nvidia’s stockpiled chips with chips made in China to create this AI model.

As a result, Nvidia’s corporate value dropped sharply. On the 27th of last month (local time), the day DeepSeek’s R1 was unveiled, Nvidia’s market value evaporated by about $600 billion (871.68 trillion won) in a single day, marking the largest single-day loss in U.S. corporate history. According to Forbes, Nvidia’s market value shrank from $3.5 trillion to $2.9 trillion. Its stock price plunged 17% in one day.