On January 27, 2025, according to S&P Global Market Intelligence, the chip-making giant Nvidia Corporation lost nearly $600 billion in its market value. Additionally, semiconductor manufacturing company Broadcom’s value had plummeted by $194.9 million, and over a trillion dollars were lost in the Nasdaq Composite. The tech sector of the stock market had experienced a massive sell-off, due to the rise of a Chinese AI startup: DeepSeek.
DeepSeek is an AI research company based in Hangzhou, Zhejiang, China. Specifically, their focus is on the development of large language models (LLMs), which are a classification of machine learning models designed for language processing. For example, OpenAI’s chatbot ChatGPT currently utilizes the GPT-4o large language model to handle the bot’s conversational abilities. In particular, the January 2025 model DeepSeek R-1 has stirred attention because of its competitiveness with other well-known large language models such as OpenAI’s GPT-4 and o1. Furthermore, the company reportedly trained its V3 model for $6 million, which is much less than the $100 million cost for GPT-4 in 2023. In addition to cost efficiency, the V3 model’s documentation reveals that it was trained on midrange hardware: Nvidia’s H800 chips. The competitive edge DeepSeek has over other leaders in artificial intelligence, combined with its open-source policies, have evidently changed the future of the artificial intelligence market.
The company’s aims toward research rather than commercialization are reflected in their open-source models, meaning the source code is freely accessible. DeepSeek-V3’s source code is licensed under the MIT license. Consequently, more researchers are now able to contribute to the development of artificial intelligence through these transparents models. Also, the company launched a technical report on the V3 model, publicizing its signature optimization methods necessary for a strong artificial intelligence model. DeepSeek’s powerful contributions to machine learning research have led to a gain in support from researchers and smaller organizations.
While DeepSeek’s revolutionary approach to the artificial intelligence industry offers new opportunities in research, there are potential problems with the freedom the business grants to users with its models. Moreover, there are some security and data concerns regarding DeepSeek’s data collection policy. In the wrong hands, such technology could be used for misinformation, deep fakes, and AI-focused cyberattacks. Despite these recurring issues in machine learning advancement, DeepSeek proves to give new hope to making research in artificial intelligence more accessible.
