Chinese AI start-up DeepSeek has made waves in the US stock market after showcasing groundbreaking artificial intelligence models that deliver performance comparable to the leading chatbots at a significantly lower cost. Founded just a year ago, DeepSeek has quickly risen to become the top app in the Australian app store, challenging the prevailing notion that the future of AI will necessitate ever-increasing power and energy consumption.
As excitement around DeepSeek’s innovation grew, global technology stocks experienced a sharp decline, with investors beginning to assess the potential impact on US-based competitors and their hardware suppliers. Unlike other chatbots, such as OpenAI’s ChatGPT, DeepSeek’s app stands out by explaining its reasoning before providing a response.
Shares of AI giant Nvidia plummeted by 17%, resulting in a staggering loss of $589 billion in market capitalization, while CEO Jensen Huang’s net worth dropped by approximately $20 billion within hours. Oracle’s shares also saw a significant decline.
DeepSeek was established in 2023 by Liang Wenfeng, who leads the AI-driven quant hedge fund High-Flyer. The company focuses on developing open-source AI models, allowing the broader developer community to inspect and enhance the software. Following its launch in early January, the mobile app quickly ascended to the top of the iPhone download charts in the US.
DeepSeek claims that its R1 release matches the performance of OpenAI’s latest models and has provided licenses for individuals interested in creating chatbots using its technology. Although the company has not fully disclosed details, the cost of training and developing DeepSeek’s models appears to be significantly lower than that of OpenAI or Meta’s top products. This efficiency raises questions about the necessity of substantial investments in advanced AI accelerators from companies like Nvidia.
DeepSeek asserts that R1 performs on par or better than competing models in several key benchmarks, including AIME 2024 for mathematical tasks, MMLU for general knowledge, and AlpacaEval 2.0 for question-and-answer performance. It also ranks among the top performers on a UC Berkeley-affiliated leaderboard known as Chatbot Arena.
In the US, there are growing concerns as Washington has prohibited the export of high-end technologies, such as GPU semiconductors, to China in an effort to hinder the country’s advancements in AI—an essential battleground in the US-China tech rivalry. However, DeepSeek’s progress indicates that Chinese AI engineers are making significant strides despite these restrictions.
