This article is part of Forbes’ exploration of China’s Wealthiest in 2025. Discover the complete list here.
In January, the tech landscape was shaken by Chinese AI innovator DeepSeek, which unveiled its budget-friendly AI model, R1. Developed at a significantly lower expense than competitors like OpenAI’s GPT, this move propelled the company’s founder, Liang Wenfeng, 40, into the billionaire echelon. Making his debut on China’s top 100 wealthiest list, Liang’s fortune is estimated at $11.5 billion, largely attributed to his share in the privately managed DeepSeek, valued by Forbes at approximately $15 billion.
DeepSeek has yet to seek outside investment, relying primarily on funding from Liang’s earlier enterprise, the quantitative hedge fund High-Flyer, established a decade ago. Analysts from both China and the U.S. speculate that DeepSeek’s worth could range from 5% to 10% of OpenAI’s $300 billion valuation achieved in a March funding round spearheaded by Japan’s SoftBank. While OpenAI’s valuation reportedly soared to $500 billion in October through a secondary share sale to investors like Thrive Capital and T. Rowe Price, Alex Platt, a research analyst with D.A. Davidson in Portland, suggests the March valuation offers a more accurate comparison for DeepSeek.
Although OpenAI introduced its sophisticated GPT-5 model in August, DeepSeek has only rolled out minor enhancements to its existing models and has yet to unveil a next-generation version. Commercially, the Chinese firm’s progress has been sluggish, with sales figures trailing far behind OpenAI’s reported $12 billion annual revenue last year, observes Platt. (DeepSeek and High-Flyer did not respond to requests for comment.) OpenAI’s revenue streams include monthly subscription fees, ranging from $20 to $200 for advanced ChatGPT versions, and developer charges for model access, serving over 700 million weekly users.
In China, intense competition has led many tech companies to offer chatbots free of charge. DeepSeek’s application had more than 73 million active users by September, based on data from research firm aicpb.com. Although DeepSeek imposes a fee on business developers for model usage, it is minimal, starting at $0.03 per million tokens, a metric for data processing volume. In contrast, OpenAI’s latest GPT-5 charges significantly more, at a minimum of $0.125 per million tokens.
Liang, the mind behind DeepSeek, has not set financial goals for his self-funded enterprise. In a rare interview with local media outlet 36Kr last year, Liang, who tends to avoid the limelight, expressed that his main aim is to foster innovation. He noted that China has long been a technology follower, and DeepSeek’s mission is to expand the boundaries of technological advancement.
Amid fierce domestic competition, Chinese companies have mostly been offering their chatbots for free. DeepSeek’s app had over 73 million active users in September, according to Chinese research firm aicpb.com. While the company charges business developers for using its models, the fee is modest, starting from $0.03 per million tokens, which is a measurement of the amount of data processed. OpenAI’s latest GPT-5 offering costs almost five times more, at least $0.125 per million tokens.
The DeepSeek founder has yet to outline any financial targets for his bootstrapped venture but in a rare interview with local media firm 36Kr last year, the reclusive Liang said his primary motive is driving innovation. China has been a follower for too long, he was quoted as saying, and DeepSeek’s goal is pushing the boundaries of technology.
The son of a primary school teacher, Liang traces his interest in AI back to his college days as a student of computer vision at Zhejiang University. After graduating with a master’s degree in information and communication engineering in 2010, he set up High-Flyer, using proprietary AI algorithms to pick stocks. The company manages about $8 billion in assets, according to financial data provider Preqin.
A research paper by Liang and his team, which made the cover of academic journal Nature in September, provided details of DeepSeek’s training methods, including how the firm spent only $294,000 to train one of its AI models. That compares with $500 million reportedly spent by rival OpenAI in one six-month training run of GPT-5. But DeepSeek’s self-declared cost doesn’t include an estimated $1.6 billion outlay on hardware over the years, including acquiring Nvidia chips, according to a January report by SemiAnalysis, a Florida-based research firm. Now with U.S. restrictions on semiconductor exports, the company has bought chips from Chinese telecoms giant Huawei to support its models, according to SemiAnalysis analyst AJ Kourabi.
Meanwhile at home, DeepSeek has to now contend with strong rivals, notably Chinese e-commerce giant Alibaba, which is investing over $50 billion in AI over the next three years as it aims for breakthroughs in artificial super intelligence, says Charlie Chai, a Shanghai-based analyst at 86Research, who notes that developers are starting to lean toward Alibaba’s Qwen models. These are “just as good as DeepSeek’s models, if not better,” says D.A. Davidson’s Platt, adding that the world is waiting to see what’s next from DeepSeek.
