The release of the generative artificial intelligence (AI) model DeepSeek-R on January 20th, 2025, set off a wave of more or less sincere panic in Silicon Valley. With performances surpassing those of OpenAI’s ChatGPT4, the small Chinese start-up, 143 employees strong at the time, reaffirmed its talents, first glimpsed a month earlier with the publication of DeepSeek v3 in late 2024. Founded and led since 2023 by Liang Wenfeng, head of the High-Flyer hedge fund of which DeepSeek is a subsidiary, the company has benefited from substantial initial funding, allowing it to devote itself to research rather than chasing profitability. Its ties to Chinese authorities remain unclear, and its sudden success may have come as a pleasant surprise for Beijing, which has hailed it as a national champion in its technological competition with Washington.

Rather than a “Sputnik moment”, the U.S. leaders of this rapidly changing industry are now experiencing a “DeepSeek moment”, as so-called “reasoning” models mark a significant step in the evolution of large language models (LLMs). DeepSeek-R1, drawing inspiration from its competitors, also reflects the growing popularity of distillation, a process whereby the capabilities and “knowledge” of another AI model can be extracted by training based on its responses, resulting in a credible iteration of the model. The freely accessible model reflects China’s strategic positioning in open-source AI, and challenges Europe’s role in a game that appears to be less rigid and more open.

This Memo is available in French and English.