China’s AI Landscape: A Free-For-All, Not a Central Plan
The dominant narrative about China’s AI race often frames it as a government-backed sprint toward AGI capabilities, competing head-to-head with the US frontier. However, an examination of over 6000 records of generative AI models filed through China’s registry system reveals a different story.
The AI Registry System
Since 2023, all public-facing AI models must be filed with regulators before launch, creating an unprecedented window into China’s actual ecosystem. This registry categorizes multiple datasets by service type and regulatory concern, including:
- Internet Information Service Algorithm (IISA)
- Deep Synthesis Algorithms (DSA)
- Generative AI Services (AIGC)
This analysis primarily draws on the AIGC and DSA datasets, focusing on generative AI development.
Key Findings
The analysis reveals several crucial points:
- Private companies, rather than the state, drive development.
- Frontier developers are pursuing specialized models rather than converging on a single path for scaling Large Language Models (LLMs).
- Geographic concentration indicates local governments actively shaping innovation clusters through fiscal competition.
Understanding the Data
The AIGC dataset tracks all new public-facing AI models developed in China, capturing:
- Models being developed (training from scratch or fine-tuning open-source models)
- Models being deployed (using APIs of China’s models or locally installed open-source models)
These datasets provide insights into the landscape and speed at which models reach actual users.
Caveats
Three important caveats must be considered:
- Both datasets track general model families rather than individual versions.
- The registry system captures only part of the ecosystem, omitting internal corporate AI deployments and overseas operations.
- The filing-to-registration process typically takes 2-5 months, meaning dates may not align perfectly with actual development or deployment.
Private Sector Leadership with Accelerating Deployment
The total number of AI models in China has been steadily growing, with a noticeable increase in deployment from 2023 to 2025. Although no significant surge in the number of models filed has been observed, it is essential to remember that the registry does not account for model updates.
Private companies dominate both model development and deployment, including major players like Alibaba and TAL Education Group. Even as state actors such as telecommunications companies become more active, they remain secondary participants in overall model development.
State Involvement
State-affiliated actors are increasingly visible in the registry but primarily focus on building infrastructure and application layers. For example, major telecom operators like China Mobile and China Telecom are among the most active government participants, developing models to support public services.
State institutions are deploying AI for specific use cases, particularly in customer service and healthcare. However, their integration of AI remains narrow, indicating limited efforts toward comprehensive sectoral digitalization.
Frontier Developer Strategies
While generalist developers like DeepSeek and Moonshot focus on large general-purpose models, most frontier developers are optimizing for specific commercial applications rather than pursuing AGI. Companies like Alibaba and ByteDance have integrated AI into their existing platforms, enhancing their services across various sectors.
Smaller startups are more focused on specific verticals, such as healthcare and finance, often achieving success in their niche markets.
Geographic Concentration and Innovation
A significant pattern in the data indicates that five provinces account for over 80% of all model development and deployment. These provinces—Beijing, Shanghai, Guangdong, Jiangsu, and Zhejiang—are dominant due to greater fiscal capacities and talent concentrations.
Local Government Incentives
Local governments are competing to attract AI development through fiscal incentives, establishing subsidy programs to encourage model filings. For example, Shanghai’s Xuhui District increased rewards significantly from 2 million RMB to 5 million RMB between 2023 and 2025.
Conclusion
China’s AI ecosystem operates under a different logic than the typical “US-China AI race” narrative suggests. Development and deployment remain private-led, with the state filling infrastructure gaps. The data indicates a trend toward specialization and regional policy competition shaping innovation.
This decentralized approach fosters a dynamic landscape where private companies lead, while state actors contribute in a complementary manner, forming a unique AI development environment that diverges from centralized planning.
