Executive Summary
Chinese AI models delivered 4.12 trillion tokens in one February week versus 2.94 trillion from US models, establishing a structural cost advantage that threatens American AI dominance. Chinese token production costs $2-3 per million outputs compared to $15 for US models like Anthropic's Claude, creating a six-fold pricing gap that's driving Silicon Valley startups to adopt Chinese AI infrastructure. This cost differential stems from cheaper electricity and China's mixture-of-experts architecture that requires less compute power—ironically developed as a response to US chip restrictions. The shift is accelerating as agentic AI systems consume significantly more tokens than traditional chatbots, making cost efficiency critical. US commentators are raising strategic and regulatory concerns as Chinese AI becomes embedded in American tech infrastructure, with the algorithm, staff, and data processing occurring in China beyond US regulatory reach. However, this dynamic positions NVIDIA as the primary beneficiary, as both Chinese and US companies will compete for the most powerful Blackwell chips to maximize token generation efficiency. The geopolitical implications are profound: China has developed what amounts to 'digital oil' production capacity that could reshape global AI economics, while export control battles intensify across rare earths, solar technology, and now potentially AI services.
Key Insights
what Alice Han and James Kynge said“Minimax and Moonshot, these are two Chinese AI models. Their cost per million output tokens is about $2 US to $3 US. Conversely, Anthropix Claude Sonnet 4.5, that's a big American LLM, they cost about $15 US dollars per million output token.”
This is a preview. Log in to see the full analysis including investment opportunities, risks, catalysts, and detailed insights.