Dreams can come true!
This is the story of Liang Wenfeng, the founder of DeepSec.
Liang Wenfeng’s story is the realization of a dream that once began in a modest family and has now become a symbol of a revolution in China’s artificial intelligence race. Born in 1985 in Zhanjiang, Guangdong, Liang’s father was a primary school teacher, but his son’s extraordinary mathematical abilities led him to study electronic engineering and AI at Zhejiang University and earn a master’s degree in electronic engineering and AI. In 2015, Liang founded a hedge fund called Highflyer Quant, which used AI for algorithmic trading. In just four years, he managed $8-10 billion in assets, but Liang’s heart still beat in basic AI research.
In 2021, when the United States announced restrictions on chip exports, Liang took a leap of faith and stockpiled 10,000 Nvidia A100 GPU chips. It was a time when the world was suffering from a hardware shortage for AI, but Liang not only pulled China out of the crisis but also spun off Highflyer’s research wing in 2023 to found DeepSec. The goal was not just to build a chatbot, but to take China “from replication to creation”—a country that was self-reliant in technology.
DeepSec’s first major achievement was its R1 model, which was open source and allowed free commercial use under the MIT license. Liang believed that the walls of closed systems were temporary, while real progress was possible only through open collaboration. This philosophy came to fruition when DeepSec’s chatbot launched in the US in January 2025 and became the most downloaded app in 50 countries in just ten days. Its performance not only surpassed OpenAI’s GPT-4 but also caused earthquakes, such as an 18% drop in Nvidia shares, shaking the global tech industry.
The success was not just about resources, but also about Liang’s unique approach. He recruited fresh graduates, eliminated KPIs and hierarchies, and gave researchers unfettered access to GPU clusters. It was this model of “bottom-up innovation” that made it possible to develop models like DeepSec V3 for just $5.58 million—1/20th the cost of OpenAI’s GPT-4. When they lowered the price of their API to 1/53 of that of Cloud 3.5 Sonnet in May 2024, it shook up the entire industry landscape, prompting companies like Alibaba and Tencent to also lower their prices.
Liang’s vision now is to reach AGI (Artificial General Intelligence), for which he sees language, mathematics, and multi-modality as key pillars. He believes AGI is possible in our lifetimes within 2 to 10 years, especially as young Chinese scientists “show their brilliance.” While there are challenges like US chip restrictions, DeepSec claims it can train its models on just 2,000 chips, a revolutionary step compared to competitors’ 16,000 chips.
The most interesting aspect of this journey is that Liang eschewed VC funding altogether. Highflyer's financial resources and philosophy of "long-term research" have kept DeepSec free from market pressure. That is why when experts around the world expressed surprise at the performance of DeepSec V3, Nvidia scientist Jim Fen called it "the biggest dark horse in the AI world
."
Today, DeepSec has become not just a company, but a metaphor for China's technological sovereignty. As Professor Marina Zhang of the University of Sydney said, "Compensating for hardware deficiencies with software expertise—this is DeepSec's real success." This story of Liang Wenfeng proves that when curiosity, courage, and vision come together, even a lack of resources cannot be an obstacle. Liang insists that as long as China continues to copy the United States, it will remain number two. Now is the time for us to produce the original and the world to make copies of us. Let's see what happens next.
