DeepSeek appears to be following a markedly different strategy from leading U.S. AI companies. Rather than competing aggressively for consumers, developers, and enterprise customers, CEO Liang Wenfeng is focused primarily on artificial general intelligence, according to an alleged leaked transcript of a May investor call analyzed by The Strange Review.
Liang reportedly views today’s chatbot and API markets as secondary to the much larger opportunity he expects AGI to create. That philosophy also shapes DeepSeek’s pricing. The company maintains relatively thin API margins even though Liang believes it could charge considerably more without significantly reducing demand. Lower prices discourage competitors from hosting DeepSeek’s open models more cheaply, effectively turning thin margins into a competitive defense.
Compute remains DeepSeek’s biggest constraint. Liang estimates the company is 1–2 years behind frontier AI labs while operating with roughly one-twentieth of their computing resources. His goal is not to eliminate that compute disadvantage but to shorten the performance gap to 3–6 months. He also argues that U.S. export restrictions have accelerated China’s move toward domestic AI chips.
DeepSeek’s research priorities further distinguish it from competitors. Liang describes a progression from language models and chain-of-thought reasoning to agents, continual learning, self-iteration, and eventually physical AI. Continual learning is considered the next major bottleneck because current models cannot continuously absorb knowledge and improve themselves effectively.
DeepSeek is deliberately avoiding areas such as video generation and world models despite their commercial potential because Liang does not consider them central to advancing intelligence.
The company’s unconventional philosophy extends internally. It has operated without traditional organizational charts, key performance indicators, or a written corporate vision, although Liang acknowledges that greater structure will become necessary as DeepSeek expands.
The strategy represents a concentrated wager: sacrificing immediate commercial opportunities to remain focused on the longer-term pursuit of increasingly autonomous and capable AI.
