On January 20, DeepSeek announced two outstanding AI models, including the free version V3 and the R1 model, trained on Nvidia's H800 chips at a cost of only 3.6% compared to OpenAI o1 but still ensuring equivalent performance.
The entire development process of the R1 model took only two months and cost less than $6 million. The V3 version quickly achieved great success, surpassing ChatGPT to become the most downloaded application on the App Store on January 27.
The impact of DeepSeek’s AI models has been far more than just a technological one; it has also sent shockwaves through financial markets. Nvidia, one of the world’s two most valuable AI companies, saw its market capitalization drop by nearly $600 billion as its stock fell nearly 17% on January 27 alone. Investors are starting to question Nvidia’s future as its expensive GPUs now compete with a much cheaper AI model.
DeepSeek has also forced other major companies to adjust their strategies. Meta responded by setting up four “war rooms,” bringing together top engineers to study how DeepSeek can reduce costs and optimize its AI models. Two rooms focused on analyzing training and operating costs, a third looked at DeepSeek’s training data sources, and the last room studied how to improve the Llama model based on the characteristics of AI from China.
DeepSeek is funded by High-Flyer Ventures and has no plans to raise additional capital. The company is based in Hangzhou, Zhejiang, and is focused on innovative research in AI architecture and algorithms. With a mission to “unlock the mystery of AGI through curiosity,” DeepSeek has made a new step forward in the technology race.
DeepSeek’s use of open source is a breakthrough in the industry, experts say, raising questions about the efficiency of massive investments in proprietary models. DeepSeek’s product manager, Shubham Saboo, said that R1’s cost per million output tokens is just $2.19, while OpenAI o1 requires $60.
The impact of DeepSeek’s development was not limited to Nvidia or Meta. Top tech billionaires also suffered significant losses in their fortunes. Nvidia CEO Jensen Huang lost $20.1 billion, Oracle founder Larry Ellison lost $22.6 billion, while CEOs Michael Dell and Changpeng Zhao lost $13 billion and $12.1 billion, respectively.
Experts say DeepSeek's success is a testament to strong innovation. Microsoft CEO Satya Nadella said: "We need to keep a close eye on developments in China. Models like DeepSeek are demonstrating that computational efficiency can be achieved at a much lower cost."
