On January 20, the same day as Donald Trump’s inauguration in Washington, Chinese Premier Li Qiang met with experts to discuss the government’s policies for the coming year. The low-profile gathering received little international attention, but one of the few speakers was Liang Wenfeng, a hedge fund founder and AI entrepreneur who was relatively unknown outside China. That day also marked the launch of DeepSeek’s latest AI model, R1, which the company claimed could rival OpenAI’s most advanced reasoning model.
Within a week, DeepSeek’s app soared to the top of U.S. app stores, surpassing competitors like OpenAI’s ChatGPT and Anthropic’s Claude. This rapid rise brought both Liang and his AI firm into the global spotlight.
Leading AI startups in the U.S. are valued in the tens of billions—ranging from $50 billion for Elon Musk’s xAI to $157 billion for OpenAI—backed by major investors like Microsoft, Amazon, and Silicon Valley’s top venture capital firms. In contrast, DeepSeek appears to have no external investors beyond Liang and his three cofounders. Chinese corporate records indicate that Liang owns about 84% of the Hangzhou-based company, which he founded in 2023 using funds from High-Flyer Capital Management, the quantitative trading hedge fund he cofounded in 2015.
The value of Liang’s open-source AI model remains uncertain. Forbes consulted five analysts and investors, three of whom estimated DeepSeek to be worth at least $1 billion—possibly much more—despite its limited revenue. Currently, its only paid product is developer access to its models, priced at $2.19 per million output tokens (roughly 750,000 words), significantly lower than OpenAI’s $60. This pricing strategy may be an attempt to undercut larger competitors and capture market share. Based on user token consumption estimates from investment firm D.A. Davidson, Forbes calculates that DeepSeek generates about $6 million in annualized revenue per million paying users.
Applying a revenue multiple of 65—falling between those of Anthropic and Chinese open-source AI startup 01.AI—DeepSeek would need approximately 3 million paying users to reach a $1 billion valuation. Appfigures data shows that DeepSeek’s app was downloaded over 3.6 million times in its first two weeks. While William Blair partner and software analyst Arjun Bhatia estimates that fewer than 10% of these users are paying customers, he suggests that valuation should be based on DeepSeek’s user base in a year or two.
Regardless of its exact revenue, Forbes estimates DeepSeek is worth at least $1 billion, in part because cutting-edge AI models are often valued based on their long-term potential rather than immediate earnings. D.A. Davidson analyst Alexander Platt considers DeepSeek one of the “top five AI labs in the world,” arguing that its research capabilities alone justify a higher valuation.
Bhatia believes $1 billion is a conservative estimate, even with a “China discount” due to geopolitical risks. Meanwhile, tech investor and Corpora.ai CEO Mel Morris values DeepSeek at $10 billion, suggesting that some companies might even be willing to acquire it simply to remove it from the competition.
DeepSeek’s Chinese rivals have all been valued at $1 billion or more in funding rounds, despite not matching its performance. The company has avoided venture capital backing partly because Liang believes VCs prioritize quick commercialization, which did not align with DeepSeek’s research-driven approach, as he stated in a 2023 interview with Chinese tech outlet 36Kr.
Beyond DeepSeek, Liang also owns at least 76% of High-Flyer, which manages $8 billion in assets, according to financial data provider Preqin. Forbes values High-Flyer at $240 million, making Liang’s stake worth approximately $180 million. Additional Chinese corporate records indicate he owns 85% of another High-Flyer entity, overseeing 65 of the firm’s 503 active funds, suggesting his equity in the company may be even greater. Combining these holdings, Liang’s net worth is likely at least $1 billion, making him the latest entrepreneur to amass a fortune in AI.
However, wealth doesn’t seem to be his primary focus. Speaking to 36Kr, Liang emphasized that DeepSeek’s value isn’t purely financial, likening AI development to buying a piano—not just because one can afford it, but because a group of people is eager to create music with it. DeepSeek and Liang, likely on holiday for Chinese New Year, have yet to respond to Forbes’ requests for comment.
Born in 1985 in the port city of Zhanjiang in southern China, Liang is the son of a primary school teacher. He studied artificial intelligence at Zhejiang University, earning a bachelor’s degree in electronic engineering in 2006 and a master’s in information and communication engineering in 2010.
During graduate school, he began exploring ways to fully automate trading in China’s stock market. In 2013, he founded his first investment firm, Hangzhou Jacobi—named after mathematician Carl Jacobi—with his college classmate Xu Jin. Two years later, they co-founded High-Flyer with another classmate, leveraging AI and mathematical models to build a hedge fund.
Even then, Liang was investing in AI infrastructure, equipping High-Flyer with 100 graphics processing units (GPUs) to optimize investment decisions. By 2019, High-Flyer had become one of China’s top-performing quantitative trading firms, prompting Liang to invest nearly $30 million to expand its computing capacity to 1,100 GPUs and establish dedicated facilities. He also founded Hangzhou High-Flyer AI to advance AI research and extended High-Flyer’s reach overseas by launching a Hong Kong-registered fund.
High-Flyer follows a conventional hedge fund model, charging a 2% annual management fee on assets and a 20% performance fee for its “enhanced funds.” Its quantitative hedging and Hong Kong-based funds command an even higher profit percentage.
In 2021, as High-Flyer’s assets under management peaked at approximately $14 billion—generating an estimated $200 million in management fees—Liang invested $155 million to acquire 10,000 Nvidia A100 chips. A pitch deck from that year, reviewed by Forbes, revealed that the firm allocated 60% of its research budget to its AI lab. This followed one of High-Flyer’s most successful years in 2020, when its flagship fund, focused on the Chinese CSI 500 stock index, outperformed the index by 50%, achieving a 71% annual return. The success was attributed to an AI-powered prediction model that identified high-performing stocks. A timeline in the deck featured an upward-trending graph with a rocket illustration next to 2020, symbolizing the firm’s vision for the future.
However, performance began to decline in late 2021, partly due to AI-driven trade timing errors. In response, High-Flyer halted new fund subscriptions in November, and a month later, an executive publicly apologized for the disappointing returns. To stabilize performance, the firm injected $55 million of its own capital into its funds in January 2022. By the end of the year, it had rebounded, with its three main funds delivering over 15% returns, outperforming the broader market, which had declined by more than 20% in 2022.
Then came DeepSeek. In April 2023, High-Flyer announced its new venture in a WeChat post titled “High-Flyer’s New Journey,” revealing the establishment of DeepSeek to develop artificial general intelligence (AGI). Liang personally financed the project, partly using High-Flyer’s proceeds, and assembled a team primarily composed of top graduates from Chinese universities. His motivation, as he told 36Kr in 2023, stemmed from a desire to explore AI’s full potential rather than purely for profit.
As DeepSeek’s prospects became clearer—and amid a regulatory crackdown on quantitative trading in China—Liang began scaling down High-Flyer to shift focus to the new venture. In October 2023, a High-Flyer representative told financial outlet Cailianshe that the firm would split into two divisions: AI research and the hedge fund, both managed by separate but parallel companies under Liang’s control. Around the same time, High-Flyer reduced its assets under management to approximately $6 billion.
Throughout 2023, as DeepSeek refined its AI models, High-Flyer adjusted its investment strategies, abandoning market-neutral products in favor of long-only positions. The hedge fund remains operational, investing $35 million of its own capital into its funds in February 2024. However, performance has been mixed. Of the 65 High-Flyer funds reporting financials, 36 recorded losses of up to 6% in 2024, while 29 posted gains of nearly 18%, according to Chinese financial data firm Tonghuashun.
Now, the spotlight is on DeepSeek, largely due to its claim that it trained its V3 model—released in December—on a fraction of the resources used by OpenAI. It reportedly required just $6 million, compared to an estimated $100 million for GPT-4, while utilizing less advanced hardware. (Some of OpenAI’s costs likely include personnel and infrastructure.) DeepSeek’s approach involves optimizing model efficiency by using fewer parameters without significant accuracy loss—similar to rounding numbers to fewer decimal places, processing entire phrases instead of individual words, and leveraging a specialized system where only certain “expert” components are activated as needed.
Additionally, DeepSeek claims that its reasoning model, R1, launched in January, can compete with OpenAI’s “o1” model in complex problem-solving and coding tasks. Unlike OpenAI, DeepSeek maintains an open-source approach, allowing anyone to access its models for free.
DeepSeek has significantly lowered the cost of AI model training, once considered an extremely capital-intensive process, says William Blair analyst Arjun Bhatia.
Despite the excitement, DeepSeek’s valuation remains uncertain and will depend on its ability to scale efficiently while maintaining competitive performance. Another key factor is how much of the global market it can capture, given growing concerns over national security. Chris Franzek, who leads financial firm Stout’s valuation services, notes that user data input into DeepSeek could be “more sensitive, from a national security standpoint, than anything that ever will appear on TikTok.” This concern is underscored by the recent U.S. Department of Commerce decision to place Zhipu AI—one of DeepSeek’s Chinese competitors—on a list of foreign entities considered national security risks, subjecting it to export restrictions.
Another question is whether Liang, who initially launched DeepSeek without a strong commercialization strategy, will aggressively pursue monetization.
“Our current thinking is to share most of our training results publicly, integrating them with commercialization,” Liang said in 2023. “We want more people, even small apps, to access large AI models at a low cost—rather than having this technology monopolized by a handful of companies.”
