How DeepSeek overtook ChatGPT on the market
DeepSeek made headlines in the spring of 2025 when it launched a direct competitor to ChatGPT. The Artificial Intelligence developed by the Chinese startup demonstrated new capabilities. At the same time, the model proved to be much cheaper to train than its American counterpart.
DeepSeek is one of the leading developers of large-scale LLM language models. The Hedge Fund High-Flyer finances the company.
The founder of the company is Liang Wenfeng. He became interested in trading while still a student. This period coincided with the 2008 financial crisis. Liang expected to be successful in this field, and in 2016, he became a co-founder of High-Flyer. It is worth noting that the fund initially focused on innovation and Artificial Intelligence. A significant moment for the company was launching a GPU-based deep learning solution. This system was later used by the fund to trade stocks. Since then, the company has undergone a number of important changes:
– Artificial Intelligence took over full trader operations in 2017;
– High-Flyer evolved into a full-fledged hedge fund in 2019;
– the company focused solely on developing algorithms for trading;
– from 2019, all financial processes of the fund are fully controlled by Artificial Intelligence. At the same time, the system uses Nvidia processors.
Not content to stop there, the company began to scale up. However, in order to further improve Artificial Intelligence, a more powerful infrastructure was needed. In 2019, Fire-Flyer started to build its first computing cluster at a cost of 200 million yuan.
In 2021, Liang launched a new project and actively procured Nvidia processors. He managed to buy 10,000 chips before the US sanctions were imposed. The cost of the new computing platform was 1 billion yuan. He supported scientific development, including projects outside the company.
Founding a startup
In 2023, a new research division called DeepSeek was launched. Its activities focused on the creation and development of artificial intelligence models.
The division operated as an independent company, with High-Flyer as the main investor. The first product was DeepSeek Coder, followed by a set of LLM language models. The startup released product updates at a rapid pace. These included DeepSeek-Coder V2, DeepSeek V2.5 and DeepSeek-Math, which included three AI models.
In November 2024, the company unveiled R1-Lite-Preview, available via API and chat. A month later, the startup released the V3-Base model and a chatbot-formatted version. However, the company’s real success came with the release of the DeepSeek-R1 chatbot, available for iOS and Android. The AI solution appeared in January 2025. Just a week later, it became the most downloaded app in the US, overtaking ChatGPT.
