The Chinese startup DeepSeek is striving to capitalize on its market lead. Last month, the company made a massive impact, valued at $1 trillion, in global stock markets with its low-cost AI model for logical reasoning, outperforming many Western competitors.
اضافة اعلان
Now, the company, based in Hangzhou, is fast-tracking the release of its next-generation AI model, R2, a successor to the R1 model launched in January, according to three sources familiar with the company’s plans.
Initially, R2 was set to launch in early May, but sources say the company is now aiming to release it as soon as possible, though no further details have been provided.
The goal of the new model is to enhance programming capabilities and support languages beyond English. Previously, the timeline for the accelerated R2 launch had not been disclosed. DeepSeek did not respond to a request for comment on the report.
Competitors Trying to Catch Up
Rival companies are still grappling with the impact of DeepSeek's R1 model, which was developed using less powerful Nvidia chips, but remains competitive with models developed by American tech companies at a cost of billions of dollars.
"Launching the DeepSeek R2 model could be a game-changer in the AI industry," said Vijay Simha Alilojata, COO of Indian tech services company Zensar. He added that DeepSeek’s success in developing low-cost AI models could prompt companies worldwide to speed up their own efforts, potentially breaking the monopoly of dominant players in the field.
The launch of the new model is likely to raise concerns for the U.S. government, which prioritizes AI leadership as a national goal. It may also encourage Chinese authorities and companies, as dozens of companies have already started integrating DeepSeek’s models into their products.
Liang Wenfeng: The Quiet Billionaire Behind DeepSeek
Little is known about DeepSeek, but its founder Liang Wenfeng has become a billionaire through the High-Flyer quantitative hedge fund he manages. Liang, described by a former colleague as "quiet and introverted," has not spoken to any media outlets since July 2024.
Reuters conducted interviews with 12 former employees and hedge fund professionals familiar with DeepSeek’s operations and its parent company, High-Flyer.
Reports and social media posts from companies, as well as research papers dating back to 2019, describe the company as more of a research lab than a profit-driven entity, unbound by the strict administrative traditions of the tech industry in China, which is known for high-pressure work environments. Despite this, DeepSeek has been credited with driving the latest boom in AI, according to many investors.
A Different Path
Liang was born in 1985 in a rural village in Guangdong Province, southern China. Later, he obtained degrees in communications engineering from the prestigious Zhejiang University.
One of his early roles involved managing the research department at a smart imaging company in Shanghai. On February 9, his former boss Zhou Chaowen told state media that Liang had hired award-winning algorithm engineers and managed his team in a flat organizational structure, unlike the traditional hierarchical model.
At DeepSeek and High-Flyer, Liang continued this approach, distancing himself from the practices of China's tech giants, which often rely on top-down management, low wages for young employees, and the grueling "996" work system (9 AM to 9 PM, six days a week).
Liang set up his office in Beijing, just a short walk from Tsinghua University and Peking University, China’s top two institutions. Former employees said that he regularly engaged in technical details and was happy to work alongside Generation Z interns and recent graduates, who made up the majority of the company's workforce.
Ex-employees described the work environment as collaborative, with daily working hours typically extending to just eight hours.
Artificial Intelligence Focus and Salary
Liang, in contrast to many other tech companies, intentionally avoided heavy investments in developing consumer-facing applications. Instead, he focused on improving the quality of the AI model itself.
According to three sources familiar with compensation policies at DeepSeek and High-Flyer, both organizations are known for paying high salaries compared to others in the sector.
In High-Flyer, a senior data scientist could earn up to 1.5 million yuan annually, while competitors rarely pay more than 800,000 yuan.
These generous salaries are funded by High-Flyer, one of China’s most successful quantitative hedge funds. Even after the regulatory crackdown on the sector, the fund continues to manage billions of yuan, according to industry sources.
Computing Power: A Long-Term Investment in AI
DeepSeek’s success in developing its low-cost AI model stems from over a decade of heavy investment in research and computing power by High-Flyer, according to three sources.
High-Flyer was one of the early pioneers in AI-based trading. In 2020, one of the company’s senior executives revealed that High-Flyer was betting heavily on AI, reinvesting 70% of its revenues into AI research.
The hedge fund spent 1.2 billion yuan in 2020 and 2021 to build two supercomputing clusters for AI. The second cluster, named Fire-Flyer II, consisted of around 10,000 Nvidia A100 chips, which are used to train AI models.
At that time, DeepSeek had not yet been established, so the accumulation of this computing power drew the attention of Chinese financial regulators.
Ultimately, the authorities decided not to intervene, a decision that became pivotal for DeepSeek’s future. In 2022, the U.S. imposed a ban on exporting A100 chips to China, but by that time, Fire-Flyer II was already up and running, giving DeepSeek a significant technological advantage.
State Embrace: DeepSeek Gains Beijing’s Support
Even before the launch of R1, there were signs that DeepSeek had earned Beijing's favor. In January, state media reported that Liang Wenfeng attended a meeting with Chinese Premier Li Qiang in Beijing, where he was selected as a representative for the AI sector, ahead of more well-known industry leaders.
The success of DeepSeek’s low-cost models has bolstered Beijing's confidence in China’s ability to innovate and surpass the U.S. This has led to an unprecedented rate of adoption by Chinese companies and government agencies, a level of support not enjoyed by other firms as quickly.
At least 13 local Chinese governments and 10 state-owned energy companies have announced that they’ve integrated DeepSeek models into their systems. Major Chinese tech companies like Lenovo, Baidu, and Tencent have also incorporated DeepSeek models into their products, including the WeChat app, the largest social media app in China.
Alfred Wu, an expert in Chinese policies at the Lee Kuan Yew School of Public Policy in Singapore, said, “Chinese leaders, including Xi Jinping and Li Qiang, have made it clear that they support DeepSeek. Now, everyone is adopting it.”
This embrace of DeepSeek by China comes at a time when governments from South Korea to Italy are removing DeepSeek from their national app stores, citing privacy concerns.
Stephen Wu, an AI expert and founder of the hedge fund Carthage Capital, said, “If DeepSeek becomes the default AI model for Chinese government agencies, it may give Western regulators more reason to tighten restrictions on exporting AI chips or cooperating on software development with China.”
Chip Shortages: A Key Challenge for DeepSeek
One of the biggest challenges for DeepSeek is the shortage of high-performance chips, something Liang himself acknowledged.
In an interview with Waves last July, Liang said, “Our problem has never been a lack of funding. The real issue is the ban on advanced chips.”
