The Innovation Mechanism and Effects: Case Study of DeepSeek-R1 Model
深度求索公司创新机制及其效应分析———以DeepSeek - R1 大模型为例
DOI:
https://doi.org/10.65967/siss.v44i5.154Keywords:
DeepSeek, Artificial Intelligence Innovation, Demand-Driven Innovation, Flywheel Effect, Technological Self-RelianceAbstract
DeepSeek-R1, developed by DeepSeek, has garnered global attention for its low-cost, high-performance, and open-source model. However, while existing research extensively discusses its innovative impact, it fails to reveal its underlying innovation mechanism. This paper employs a longitudinal single-case study, systematically detailing its innovation process, "business model," strategic management, and organizational structure to reveal its unique innovation mechanism. The study finds that DeepSeek's founders and team not only demonstrate a dual entrepreneurial spirit of arbitrage and innovation but also a strong sense of technological self-reliance, both contributing to outstanding achievements in independent and integrated innovation. Originating from its parent company, Magic Square Quantitative AI team, the typical application scenarios of quantitative trading, data science, artificial intelligence, and large-scale models enabled DeepSeek to establish a de facto demand-driven "business model" from its inception. Furthermore, the repeated reuse and optimization of single modules during the development of the DeepSeek-R1 model fostered internal economies of scale, while the synergy between multiple modules further realized an internal "flywheel effect." Its open-source model and application diffusion, while creating external economies of scale, also feed back into itself, forming a complementary innovation effect. This study fills a gap in existing research and provides a practical explanation and theoretical foundation for subsequent theoretical research and industrial policy and governance.
Downloads
Published
Issue
Section
License
Copyright (c) 2026 Studies in Science of Science

This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.


