China's state-directed competition commoditizes AI models

China's AI sector is characterized by state-directed competition, where the government sets strategic goals and provides cheap energy and data centers, allowing regional entities to compete aggressively. This has led to a commoditization of models, with the Mainland China AI index dropping 40% from its peak as 480 billion dollars in value evaporated. Thomson Reuters fine-tuned the Chinese Quen model for legal and tax use cases, achieving performance superior to Google and OpenAI models for those specific tasks using 40 million dollars in investment. This demonstrates the viability of open-source models fine-tuned with proprietary data. While China lacks the best chips, its low energy costs allow for brute-force training. The transcript suggests that US regulations may eventually block Chinese models, but their low cost could force global adoption or drive further consolidation in the US market. The approach of initiating competition rather than pure state dirigism is considered effective, and open-source models from China are becoming competitive through fine-tuning.