DeepSeek’s innovative breakthroughs and choice between open versus closed-source

Research methodology This case is solely based on published sources. These secondary sources include company websites and news releases of major artificial intelligence (AI) firms, AI industry reports such as Stanford Artificial Intelligence Index Report, news reports that have special coverage of tech news, such as the Wired, Hardware Corner, Venture Beat and Tech Radar, popular AI platforms at GitHub and Hugging Face and AI research paper repository in AI communities at Cornell arXiv. Case overview/synopsis DeepSeek was an AI startup founded in 2023 in Hangzhou, China. The company was fully funded by a hedge fund HighFlyer, whose co-founder and CEO, Wenfeng Liang, also served as DeepSeek’s CEO. The company aimed to develop large language models (LLMs) and ultimately to build Artificial General Intelligence (AGI). Due to US sanctions and embargo on advanced hardware exports to China, however, DeepSeek had to invent novel algorithms and model structures to develop stronger model capability with limited hardware resources. As such, the key to success for DeepSeek was to develop potentially game-changing architectural and algorithmic innovations. On December 26, 2024, DeepSeek launched a chat model, DeepSeek V3, with high performance at very low training costs. The initial benchmark tests indicated that DeepSeek V3 model outperformed Llama 3.1 and was comparable to GPT-4o and Claude 3.5 Sonnet. Yet, the company claimed to have trained its models in just two months at a total cost of merely $5.6m. That was the annual salary for one of those AI experts working at Meta. Initializing from the V3 model and sharing the V3 architecture, DeepSeek was planning to release its first chatbot application, the DeepSeek-R1 reasoning model, for iOS, Android, Web and application programming interface on January 20, 2025. Up to the release of the V3 model, DeepSeek’s algorithms, models and training details had been open-source, allowing its code to be used, viewed and modified by others. For the planned release of DeepSeek-R1, the company had to decide whether to continue the open-source policy or to choose closed-source, like most other firms were doing. Complexity academic level This case may be used in upper-level undergraduate and graduate classes in the fields of strategic management, innovation or business ethics. In strategic management classes, it can help students to analyze a firm’s external environment and internal strengths and weaknesses and then learn how to take advantage of the strengths to neutralize the weaknesses to cross industry entry barriers. In innovation classes, the case can help students understand how innovation takes place and how innovation can help to break industry entry barriers. In business ethics classes, students can learn how to balance firm performance and corporate social responsibility in decision-making.