Chinese AI startups revamp their business models after DeepSeek's rise; sources: Zhipu eyes an IPO, 01.ai drops pretraining to sell apps with DeepSeek, and more
Context & Ripple Effects
DeepSeek’s rise was already broadening from model development into deployment: Chinese automakers, phone sellers, and other companies were building services on top of it in a rapid wave of adoption. At the same time, major Chinese technology groups were releasing low-cost AI services and models, raising the pressure on smaller AI companies to distinguish themselves.
This report captures the startup-level response: Zhipu is said to be considering an IPO, while 01.ai is shifting away from pretraining toward applications built with DeepSeek.
First-order effects
- 01.ai’s reported exit from pretraining would redirect its effort from owning a foundation-model training stack to selling applications that use DeepSeek.
- Zhipu’s reported IPO exploration would put financing and public-market readiness alongside product strategy as immediate priorities for a Chinese AI startup.
Second-order effects
- Startups that continue training their own models face a sharper burden to show why their models or economics outperform readily available low-cost alternatives.
- Application developers can treat DeepSeek as a shared model layer, shifting near-term competition toward distribution, product integration, and customers rather than pretraining alone.
Third-order effects
- If this pattern persists, China’s AI market could separate more clearly into a smaller set of model providers and a larger application layer built on their models.
- The reported move toward an IPO suggests that AI-company differentiation may increasingly be tested by commercial traction and financing access, not only by model-building capability.
The trend: DeepSeek’s rise is accelerating the commercialization of Chinese AI by making application-focused and capital-efficient business models more viable than standalone pretraining for some startups.