DeepSeek

DeepSeek is a China-based artificial intelligence initiative and a technology company that develops large language models (LLM). The company has gained attention for its open-source artificial intelligence models that require low cost and minimal chip resources for training. DeepSeek’s models stand out in the field of artificial intelligence by offering lower hardware requirements and cost advantages compared to existing Western competitors.

DeepSeek logo - DeepSeek

The company’s most well-known product is the DeepSeek-R1 artificial intelligence model. This model was trained using fewer chips and at a lower cost compared to other large language models. DeepSeek’s models demonstrate strong performance in areas such as mathematics and coding like. They have been widely adopted by a broad user base globally World.

DeepSeek’s released AI models are viewed as an alternative to the current dominance of large technology companies based in USA.

History

Founding of DeepSeek

DeepSeek was founded in China in December 2023. Its founder, Liang Wenfeng, previously managed a quantitative investment fund called High-Flyer Capital Management and specialized in AI-based financial analysis. Inspired by the potential of AI technology in finance and data analysis, he established DeepSeek to develop large language models (LLM). During its founding phase, DeepSeek was influenced by China’s strategic ambitions in AI and efforts to create alternatives in response to U.S. chip export restrictions.

Development Phases

DeepSeek’s development process has been guided by principles of high efficiency and low cost. The company adopted a strategy of training large language models using significantly fewer chips than major technology firms. Core research and development efforts have been carried out by teams in Beijing and Hangzhou. DeepSeek has sought to optimize large-scale AI training processes by operating its own data centers.

Initial Releases and Market Launch

DeepSeek launched its first AI model in late 2024. The DeepSeek-V3 model was introduced in December 2024 and announced as the company’s first large language model. However, the major impact came with the announcement of the DeepSeek-R1 model on 20 January 2025. This model was trained using 2,000 Nvidia chips at a cost of approximately $5.6 million.

Introduction of the DeepSeek-R1 Model

The model was designed to serve as a opponent to OpenAI’s ChatGPT and Google’s AI systems. It attracted attention for its strong performance in mathematics and coding. It became the most downloaded AI assistant in application stores, surpassing U.S.-based ChatGPT in download rankings. The release of DeepSeek-R1 triggered significant echo in the global technology sector and intensified competition in the AI field.

Impact on Global Technology Stocks

The market launch of DeepSeek’s R1 model caused important fluctuations in global technology stocks. Its development using fewer chips and at lower cost was perceived as a major competition advantage in the AI market.

Following the release of DeepSeek’s R1 AI model, sharp declines occurred in the stock prices of U.S.-based technology companies. The Nasdaq index fell by more than 3 percent after the announcement. Significant value losses were recorded especially in the shares of chip manufacturers.

Nvidia experienced a drop of up to 17 percent in its stock price after DeepSeek’s release, with the company’s market value declining by over $50 billion. This decline was recorded as the largest single-day market value loss in U.S. stock market history.

Broadcom shares fell by more than 12 percent, Micron Technology by 8 percent, Advanced Micro Devices (AMD) by 5 percent, Applied Materials by 5 percent, and Qualcomm by 2 percent. Taiwanese Taiwan Semiconductor Manufacturing Company (TSMC) shares dropped by over 10 percent, while Arm lost 8 percent of its value. U.S. AI server manufacturers Dell Technologies and Super Micro Computer saw their shares decline by more than 7 percent.

Censorship and Content Moderation

Autonomous censorship mechanisms have been identified in DeepSeek’s AI models regarding certain political topics. Observations show that the models self-censor when responding to user queries on specific political issues.

It has been reported that DeepSeek’s public-facing versions implement content filtering in accordance with the policies of the Chinese government. West media have interpreted DeepSeek’s censorship mechanism as being developed in alignment with China’s information control framework, although it has been noted that such restrictions are absent in the model’s open source versions.

DeepSeek’s open-source version can be used by researchers with its censorship mechanisms disabled. The versions released in China include features that limit or alter responses on specific topics.