China's AI Industry Shifts Focus from Models to Infrastructure and Specialized Data

The competitive landscape of China's AI industry is undergoing a fundamental structural shift, moving from model development toward infrastructure and specialized data. As high-performance models like DeepSeek V4 are released as open source, the value of general-purpose models is declining, while the importance of foundational technologies for deploying AI agents in practice is growing.

Tencent Cloud recently open-sourced 'Cube Sandbox', a core infrastructure layer for running AI agents. This tool provides a hardware-isolated execution environment and is compatible with OpenAI's and AI agent startup Manus's SDKs. Notably, it touts a sub-100ms startup time as its key strength, aiming to become a major utility provider for the era of autonomous AI agents.

Moreover, the commoditization of models is stimulating the high-value-added market for specialized data. Data governance company Xunce Technology announced that its annual recurring revenue (ARR) from token calls surged 300% following the release of DeepSeek V4. This is driven by explosive demand for 'scenario-specific tokens' optimized for particular use cases such as finance, manufacturing, and robotics.

As a result, value in China's AI market is shifting away from the models themselves and toward the isolated environments (Sandboxes) that reliably run them and the high-quality data (Tokens) that maximize their efficiency. This suggests that, even amid U.S. semiconductor export restrictions, Chinese companies are building their own independent ecosystem through efficiency and vertical integration.