Deepseek, the Chinese -based Artificial Intelligence Laboratory, went to the top of the Apple App Store lists with the chat application and attracted great interest internationally. Artificial intelligence models trained with efficient calculation techniques of the company led to the US leadership in the artificial intelligence race and the sustainability of the demand for artificial intelligence chips among the Wall Street analysts and technology experts.
When and by whom Deepseek was founded?
Deepseek is supported by the Chinese Quantitative Hedge Fund, High-Flyer Capital Management, which takes artificial intelligence-supported trade decisions. High-Flyer was founded in 2015 by Liang Wenfeng, who started to be interested in trade when he was a student at the University of Zhejiang. In 2019, the company started to operate as a Hedge Fund that focuses on the development and implementation of artificial intelligence algorithms. In 2023, High-Flyer founded Deepseek, a laboratory dedicated to investigating artificial intelligence vehicles apart from his financial affairs.
Technical Developments and Models
Deepseek started model training by creating its own data center clusters. However, like other Chinese artificial intelligence companies, it was influenced by the US equipment export bans. The company had to use the Nvidia H800 chips, a less powerful version of the H100 chips offered to US companies to train one of its latest models. Deepseek’s technical team consists of young researchers, and the company hires artificial intelligence researchers from China’s leading universities at a doctoral level. It also employs people who do not have any computer science history to better understand a wide range of technology.
In November 2023, Deepseek introduced the first model set – Deepseek Coder, Deepseek LLM and Deepseek Chat -. However, the company’s main attention was in the last spring when the next generation Deepseek-V2 family released. Deepseek-V2 performed well in various artificial intelligence criteria as a general purpose text and image analysis system and worked at a lower cost than similar models at that time. This forced local competitors, such as ByTedance and Alibaba, to reduce the use prices of some models and make some of them completely free.
Deepseek-V3, released in December 2024, further increased the company’s reputation. According to Deepseek’s internal tests, Deepseek V3 left meta’s models of Llama and OpenAI’s GPT-4O. In addition, Deepseek’s R1 “reasoning” model, which was released in January, performs similarly to OpenAI’s O1. R1 stands out as a self -controller model that produces more reliable results in fields such as physics, science and mathematics.
Market Impact and Future Perspectives
Deepseek’s pricing well below the market value and offers some of them free of charge leads to uncertainties about the business model. The company states that it maintains excessive cost competitiveness thanks to productivity breakthroughs. However, some experts discuss the company’s figures. In spite of everything, developers are interested in Deepseek’s models. Although the company’s models are not accepted as open source, they are presented with flexible licenses that allow commercial use. For example, according to Huging Face CEO Clem Dengue, developers on the Huging Face platform have created more than 500 “derivatives” models of the R1 model and these models have reached a total number of 2.5 million downloads.
Deepseek’s success against larger and long -established competitors is defined as the beginning of a “new era ında in the field of artificial intelligence. The success of the company caused Nvidia’s share price to fall 18 %on Monday and led to a public response from OpenAI CEO Sam Altman. The future of Deepseek remains uncertain. Developed models are expected to be launched, but the US government is cautious against the situation it perceives as a harmful foreign impact.