China's AI models are nearly as good as America's. Its AI companies make a tenth of the money. Chinese models like Kimi K3, DeepSeek, and GLM now rank close to the US frontier on independent benchmarks, often trailing by months rather than years. But combined annual revenue for major Chinese AI firms sits around $10.7 billion. OpenAI alone is estimated at $40-50 billion. Anthropic around $65 billion. Here's the gap underneath the gap. Chinese consumer AI apps barely convert free users into paying ones, sometimes under 6% of monthly users. Doubao has hundreds of millions of active users and only a few hundred thousand paying. Price sensitivity runs deep, and putting up a paywall triggers real user backlash. So who's actually funding this race? More than 60% of equity investment in Chinese AI chips and servers comes from state-affiliated sources. Government spending covers infrastructure and strategic direction, private companies drive the model competition itself, and ordinary users show up in huge numbers but mostly for the free tier. The result is a strange split: China can build models that compete with the best in the world, but can't yet get its own citizens to pay for them the way American users pay for ChatGPT or Claude. Matching capability was supposed to be the hard part. Turning that into revenue might be harder.