An AI company based in Hangzhou
DeepSeek is a private Chinese artificial intelligence company headquartered in Hangzhou, Zhejiang. Founded in July 2023 by Liang Wenfeng, it develops large language models (LLMs)—systems trained to process and generate text—as well as a chatbot app that shares the company’s name.
The company is owned and funded by High-Flyer, a quantitative investment fund co-founded by Liang. DeepSeek is therefore a technology company working on AI research and development, not an investment institution.
Open-weight models
A defining feature of DeepSeek is that it releases models with open weights. Weights are the numerical parameters that shape how a model processes information; making them available can let others examine or adapt a model, subject to its license. This does not, by itself, mean that the training data is open or that every component of a system is available.
DeepSeek gained international attention in January 2025, when it released the reasoning model DeepSeek-R1 alongside its chatbot. Reports of R1’s competitive performance compared with OpenAI models, together with the company’s stated training-cost figures, drew attention to the efficiency of its methods. Techniques associated with its models include mixture-of-experts architectures and sparse attention mechanisms; implementation and results can differ across models.
Why it became a subject of debate
In practical terms, DeepSeek can refer to the Chinese AI company, the models it develops, and, by extension, its chatbot. Its prominence reflects a combination of open-weight releases, claims of competitive performance, and a focus on using fewer resources to train large-scale systems. This has prompted discussion about China’s place in international AI competition and about chip export controls.
Making model weights available does not resolve questions about use and governance. Controversies associated with the company include censorship in China, protection of user data, and the use of chips subject to export restrictions. These are separate issues: one concerns content and political context, while the others involve privacy and access to technology. The existence of these debates does not, on its own, establish how every model or service handles each issue.