China's AI Boom Hits Reality Check: Intelligence Meets Cash Flow Crunch

The Chinese AI industry is reaching a critical juncture as models demonstrate technical progress but struggle to convert intelligence into cash flow. According to Goldman Sachs analyst Ronald Keung, China's AI race has entered its most expensive phase yet.

Compute demand is high, cloud growth is accelerating, and token usage is exploding, propelling Chinese developers to the top of global rankings. However, falling free cash flow estimates indicate that investors are no longer willing to overlook the cost of infrastructure development. Keung's team cautions that the industry is shifting from focusing solely on model intelligence to considering funding strength, distribution, and control of the agent layer.

The speed of technical progress is evident in models like DeepSeek's 437-fold inference efficiency improvement. Nevertheless, narrowing differences between competing models are making it increasingly challenging for standalone AI models to justify their investment. Goldman Sachs has adjusted its preferences, elevating Games and Entertainment to its top China Internet preference while relegating standalone AI models to fifth place.

The research suggests that the most valuable AI moat may emerge in workplace agents, where distribution, workflow data, and third-party ecosystems can transform model intelligence into recurring revenue. Keung's team emphasizes that the next phase of the industry will be decided by factors beyond model intelligence alone, such as funding strength and control of the agent layer.