A small AI lab from China, DeepSeek, has created a global buzz by revealing the technical blueprint of its cutting-edge AI model, DeepSeek R1. This move has turned the company’s low-profile founder, Liang Wenfeng, into a national icon, showcasing China’s resilience against U.S. efforts to curb its high-tech advancements.
DeepSeek, founded by hedge fund manager Liang Wenfeng, released its R1 model on Monday, explaining in a detailed paper how to build a large language model on a bootstrapped budget that can automatically learn and improve itself without human supervision. This sparked a global discussion in the AI community about whether US companies like OpenAI and Google can maintain their dominance in the field.
What makes DeepSeek’s R1 model a game-changer in AI?
On Monday, DeepSeek published a detailed paper explaining how it built its AI model on a modest budget. Unlike traditional AI models, R1 can learn and improve itself automatically, pushing boundaries in AI development. U.S. companies like OpenAI and Google DeepMind had been considered leaders in advanced AI models, but they’ve kept their methods closely guarded. DeepSeek’s transparency with R1 has set off a heated debate in Silicon Valley, challenging U.S. giants like Meta and Anthropic to defend their edge.
Liang, originally a hedge fund manager, is now being celebrated as a symbol of China’s innovative potential. This week, he attended a high-profile meeting with China’s second-most powerful leader, Li Qiang, alongside other entrepreneurs tasked with driving breakthroughs in core technologies.
Liang’s foray into AI began in 2021 as an eccentric side project when he purchased thousands of Nvidia GPUs while running his High-Flyer quant trading fund. At the time, industry insiders dismissed him as a hobbyist billionaire. But Liang had a vision—to develop human-level AI, a goal he made public when launching DeepSeek in 2023.
His unconventional background proved an asset. Liang’s hedge fund experience helped his team master GPU optimization, a skill they leveraged to maximize computing power despite U.S. export restrictions on advanced Nvidia chips. “DeepSeek’s engineers know how to squeeze the most out of these GPUs, even if they aren’t cutting-edge,” said one AI researcher.
Unlike most AI companies chasing commercial gains, DeepSeek operates more like an academic institution. Liang reinvests his hedge fund profits into the company, paying top-tier salaries to recruit the best AI talent from Chinese universities like Peking and Tsinghua. DeepSeek’s purely Chinese identity—eschewing overseas experts—has garnered admiration at home, emphasizing local ingenuity and independence.
DeepSeek reportedly trained its 671-billion-parameter model using only 2,048 Nvidia H800 GPUs and a $5.6 million budget—an achievement that rivals OpenAI’s and Google’s far costlier models. This efficiency highlights the advantage of being a “second mover,” according to Ritwik Gupta, an AI policy researcher at UC Berkeley. By learning from others’ breakthroughs, DeepSeek avoided the heavy R&D costs incurred by pioneers.
Despite its accomplishments, questions remain about DeepSeek’s ability to keep pace with global competitors. High-Flyer’s financial returns lagged in 2024, partly due to Liang’s focus on DeepSeek, raising concerns about long-term sustainability. Meanwhile, U.S. players like OpenAI and Elon Musk’s xAI are doubling down on next-gen AI infrastructure. OpenAI, for instance, plans to spend at least $100 billion through its Stargate joint venture with Japan’s SoftBank, while xAI is building a supercomputer with over 1 million GPUs.
DeepSeek has built one of China’s largest advanced computing clusters, but as rivals scale up their efforts, the race is far from over. Still, Liang and his team are driven by a bold ambition: to prove that Chinese innovation can thrive on the world stage, even in the fiercely competitive AI landscape.