Per Woofun AI, J.P. Morgan disclosed on June 22nd that AI usage is transitioning from casual 'chatting' to structured workflows, accelerating token consumption across code, agents, and content generation. OpenRouter data indicates platform-wide token usage rose 4.7% week-over-week to 46.7 trillion, with Chinese models capturing 18.8 trillion tokens compared to 5.8 trillion for US counterparts.

DeepSeek V4 Flash topped the rankings with 4.94 trillion tokens, followed by Xiaomi MiMo-V2.5, MiniMax M3, and Tencent Hy3 preview. This dominance stems from superior cost-performance ratios, as Chinese models leverage MoE architectures and attention mechanism optimizations to offer API calls at a fraction of US prices. Consequently, while call volumes surge, the Silicon Data LLM Token Expenditure Index remained subdued between 1.64 and 1.68, reflecting a market preference for cheaper, efficient models. J.P. Morgan projects this trend will bolster cloud providers like Baidu, Alibaba, and Tencent, with the China Big Model Market expected to exceed 100 billion RMB by 2030.