Key Highlights

  1. Chinese AI Models Now 20 of World's 50 Most Used — Up 400% Since 2025: The extraordinary increase in Chinese AI model representation among the world's most-used models reflects a rapid shift in competitive dynamics.
  2. Monthly Token Usage of Chinese Models Surged 113% to 98 Trillion — vs 43% US Growth to 53 Trillion: The dramatically different growth rates between Chinese and US AI model token usage signal an accelerating Chinese competitive advantage in deployed AI scale.
  3. Chinese Token Usage Now 85% Higher Than US Models — Up From 24% Gap in May: The rapid closure and reversal of the Chinese-US AI usage gap within a single month illustrates the speed at which competitive dynamics can shift in the AI model market.

The AI race between the United States and China is intensifying rapidly according to Apollo data: 20 of the world's 50 most-used AI models now come from China, up 400% since 2025, while the number of US models in the top 50 has declined to 28 from 33 over the same period. The competitive gap in deployed AI model usage is shifting at a pace that is surprising even well-informed observers of the US-China technology competition.

The monthly token usage data is the most striking specific metric. Chinese model token usage surged 113% month-over-month to 98 trillion tokens in June, while US model token usage grew 43% month-over-month to 53 trillion tokens. The difference in growth rates — 113% versus 43% — reflects a compounding dynamic where faster Chinese growth rate is widening an already substantial absolute usage advantage.

As a direct result of these growth rate differentials, token usage for Chinese AI models is now 85% higher than for US models — a dramatic shift from just May, when the gap was only 24%. A competitive metric that moves from 24% Chinese advantage to 85% Chinese advantage in a single month reveals the speed at which the deployment dynamics of AI models can shift when growth rates diverge significantly.

For investors in US AI infrastructure companies including Nvidia (NASDAQ: NVDA), which supplies chips used in AI model training and inference, the competitive data raises questions about the geographic distribution of future AI demand. If Chinese AI models are experiencing faster token usage growth, the supporting AI infrastructure investment may increasingly be captured by Chinese chip designers and infrastructure providers rather than US-based suppliers.

The data also has implications for the thesis that AI model efficiency gains — like those reported by DeepSeek — reduce semiconductor demand growth. A world where Chinese models are processing 98 trillion tokens monthly with faster growth rates is not one where AI compute demand is contracting — it is one where the geographic distribution of that demand may be shifting in ways that affect US semiconductor company revenue assumptions.