On January 27, 2025, a major shock wave rippled through financial markets and the artificial intelligence industry. Chinese startup DeepSeek unveiled a new low-cost, open-source model that, despite skepticism from some analysts, could threaten hundreds of billions invested in AI infrastructure.
A disruptive approach to AI
DeepSeek has just introduced DeepSeek-V3, an intelligent assistant capable of competing with the best market solutions, including ChatGPT. Its distinguishing feature lies in its exceptionally low development cost, estimated between 5 and 6 million dollars – a trivial sum compared to the massive investments made by American giants like OpenAI or Meta. The stock market impact was immediate, affecting semiconductor leaders like Nvidia (-17%, erasing nearly 600 billion dollars in market capitalization) and Broadcom (-17%), as well as service providers like Oracle (-14%) and infrastructure providers like Cisco (-5%).
Open-source as a transformation catalyst
DeepSeek-V3 stands out for its open nature. Unlike its competitors, the company has opted for a particularly permissive open-source license. This strategy could radically transform the AI ecosystem by allowing numerous players, beyond China, to benefit from this cost-effective approach. History shows that open architectures, such as IBM PC in the 1980s, Firefox in the 2000s, or Android in the 2010s, have often stimulated innovation while lowering entry barriers.
A new performance paradigm
DeepSeek's innovation challenges the established principle that AI model performance increases proportionally with parameters, training data, and deployed computing power.
If this frugal approach were to become widespread and meet generative AI needs at lower costs, the significant computing capabilities under development could become surplus, leading to price drops and investment strategy revisions. For semiconductor, cloud, and data center industries, the stakes are critical. A decrease in high-end equipment demand could trigger an overcapacity crisis, affecting the entire AI value chain.
Market impact and geopolitical perspectives
In the short term, markets expect a period of increased volatility as investors reassess sector companies' values. Tech giants like Nvidia, Broadcom, and ASML could see their valuations decrease significantly. Venture capital funds and institutional investors heavily exposed to AI are also under pressure. However, this correction could ultimately foster the emergence of more efficient and targeted AI models.
This innovation is part of the Sino-American technological rivalry. Founder Liang Wenfeng, based in Hangzhou, recently met with Premier Li Qiang, highlighting the project's strategic dimension. DeepSeek's approach appears as a response to American restrictions on advanced technology access. Less expensive models would reduce these restrictions' impact on the Chinese ecosystem and could benefit regions like Europe, eager for AI applications but limited in infrastructure.
Limitations and questions
However, experts point out that the announced costs are likely underestimated. They don't include expenses related to data cleaning, network equipment, energy, engineer salaries, and open-source resources. The actual development cost would therefore be significantly higher than the figures presented.
For more on the China-US technology war, consult our detailed economic study.
