Three years ago, Liang Wenfeng’s quantitative Hedge Fund apologized to investors for losing money during a tumultuous period for China’s stock market.
It was a surprising drop for Zhejiang High-Flyer Asset Management, which used artificial intelligence to choose actions and had rapidly grown to become one of the largest quantitative funds in the country. While the company sailed for this crisis and its assets shrunk over a third of a peak of over $ 12 billion behind the scenes, Liang was preparing the ground for a new AI startup, Deepseek.
Deepseek, which has emerged from High-Flyer, now threatens to destabilize the Global Artificial Intelligence Supply Chain and challenge the seemingly unshakable leadership of the US in state-of-the-art critical technologies. The sudden popularity of the company’s innovative technology, which is 20 months old, and its homonymous application caused a major drop in stocks in the US and Europe on Monday (27), eliminating nearly $ 1 trillion in the giant’s combined market value of NVIDIA chips and other competitors.
Read more:
This also generated shock and admiration about how Liang, an engineer who never studied or worked outside the mainland China, got such a feat. He demonstrated that with local artificial intelligence engineers, restricted access to the latest semiconductor technologies and limited resources, it is possible to match – and even overcome – the best in the countryside.
“Every country in the world could have this kind of project in progress, if you can get talent and work on it, of course. The rest of the industry will learn from it, ”said Shuman Ghosemajumder, co -founder and CEO of Reken, a St. Francis -based AI startup.
Continues after advertising
The issue that now occupies investors, companies and policy training is whether artificial intelligence requires hundreds of billions of dollars in capital expenses to generate the latest avant -garde innovations and models – and whether export controls may contain competition Chinese.
Liang was compared to the founder of OpenAi, Sam Altman, but the Chinese citizen maintains a much more discreet profile and rarely speaks in public. “Openai is not a god and it cannot always be in the forefront,” Liang told Chinese media 36kr in July 2024.
In the previous year, Liang stated that more investment does not necessarily lead the more innovation. He also opined on how Chinese companies have been mostly followers rather than innovative technology. The problem has been a “lack of confidence and not knowing how to organize high density talents to achieve effective innovations,” was quoted.
Continues after advertising
Exception
Liang was born in 1985 in Zhanjiang, a poor city in the province of Guangdong, southern China. His father was a teacher of elementary school. He studied electronic engineering at the University of Zhejiang, a prestigious college in the city of Hangzhou, and also obtained a master’s degree in Information and Communication Engineering there.
High-Flyer was as unusual in China’s quantitative industry as DeepSeek is for the global AI industry.
Liang and two of their former university colleagues began to venture into domestic actions in 2008. Unlike the founders of most Chinese quantitative funds, none of them had experience in outside or institutional negotiations.
Continues after advertising
The trio tried different strategies, from discretionary negotiation to arbitration, before settling in a systematic approach to implement negotiation ideas in 2015, the year they founded High-Flyer. Initially, they built a model based on price and volume factors before trying to learn machine in 2016.
The new tool allowed the firm to deepen the search for new factors and identify “non-linear” connections between factors, his CEO Simon Lu said in an interview in 2020. The founders joined machine learning to High-Flyer products in 2018.
AI has allowed High-Flyer to reach “many innovations” and develop a multi-steraction and multi-cycle investment model to “accumulate” returns from different sources, according to a 2020 firm leaflet. Its main product, which was referenced by the CSI 500 index, was part of low-risk strategies such as intra-dia negotiation, allowing it to overcome the index in 120 combined percentage points in the previous three years, he showed.
Continues after advertising
High-Flyer grew rapidly in assets, reaching over 90 billion yuans in 2021, before stumbled later that year.
In December 2021, after experiencing record losses in some funds, High-Flyer said its artificial intelligence touched some negotiations and had a bad performance during periods of major stock market oscillations. “We feel a deep blame,” he told investors. The firm also stopped accepting new contributions and said it would reduce its assets under management and adjust its strategies.
Three months later, his marketing boss warned that certain clients sensitive to volatility should redeem their money – a highly unusual measure.
Last year, High-Flyer said it would end products that were making two-sided bets on the markets and focused on “long-only” strategies, which only assumed optimistic action positions. His assets under management fell to about 60 billion yuans.
Financing
DeepSek’s survey was funded by High-Flyer’s R&D budget, Liang said earlier. It obtained computational funds from the quantitative fund, which had accumulated 10,000 NVIDIA GPUs in 2021, before the US exports of sophisticated Nvidia chips and other graphic processing units.
Liang recruited engineering talents almost exclusively from China. Many were newly graduated from prestigious universities, interns in their final doctoral phase and Olympics medalists.
“He is a nerd, but nerd in this context is not negative,” said Zihan Wang, a doctoral student at Northwestern University who did a six -month stage at Deepseek in 2024.
Wang said Liang conducted many experiments on his own, and DeepSeek worked a lot as a research laboratory. “It started small, but as they started to have real progress, they started to get excited,” he said.
The startup began to launch periodically, seemingly immune – even stimulated – by the US ban on state -of -the -art exports.
DeepSeek launched its advanced model of reasoning in AI R1 on January 20, the same day Donald Trump was sworn in as the 47th President of America.
Earlier on Monday, Liang participated in a closed business symposium in Beijing, which was organized by Chinese Prime Minister Li Qiang. There, experts in technology, science, education and other fields offered their opinions and suggestions for a scratch of government work report, according to the official Xinhua agency. Filming on YouTube show Liang sitting at the table in front of Li and speaking, with the Chinese leader waving closely.
Significantly, DeepSeek has made its open source R1, allowing researchers and developers to use, modify and freely market the model. This has sent a sign that he wants to collaborate and innovate with others in the Global AI community.
Liang stands out among Chinese entrepreneurs for this non -commercial objective, their laser focus on research and the realization of general artificial intelligence, said Thomas Qitong Cao, assistant professor of technology policy at Tufts University in Medford, Massachusetts.
Liang is believed to have 51% of High-Flyer. This would give you a participation valued at $ 71 million based on a comparative analysis, according to Bloomberg Billionaires Index. If DeepSeek reaches the same potential as OpenAi, valued at about $ 150 billion, the founder could potentially be on the line for a huge profit.
Some questioned if Liang’s Deepseek is as promising as it seems. Disabilities include startup infrastructure capacity to deal with global traffic that expects to experience its service, or the management of the application on sensitive issues, such as the 1989 protests at Tiananmen Square and Chinese leader Xi Jinping.
Experts also questioned the assumption that DeepSeek was building with 10,000 Nvidia A100 chips, with analysts like Dylan Patel speculating that Deepseek needs at least 50,000 H100 chips, much more powerful from Nvidia. The goal, for example, operates the equivalent of 600,000 H100s from Nvidia.
Still, Liang is causing a reassessment and recalibration in the global AI ecosystem. It is clear that “the AI race will not be won by the creation of the most sophisticated model; It will be won by incorporating AI into business systems to generate tangible economic value, ”said Mike Capone, CEO of Qlik, a data analysis platform and artificial intelligence.
© 2025 Bloomberg L.P.
