does china use more tokens for AI than the us?
AI GeopoliticsAI Infrastructure & Compute
According to recent analysis, Chinese AI models have surpassed their US counterparts in token consumption, indicating that China is using more tokens for AI processing overall [1]. This shift is driven by the rapid adoption of cost-effective Chinese models from companies like DeepSeek and MiniMax, which have gained significant market share in AI usage, capturing about a third of global AI activity by late 2025 [7]. Lower pricing for Chinese models—often under $1 per million input tokens compared to 5-10 times higher costs from US providers—further encourages higher token usage in China amid intensifying competition [2].
Sources
- The rise of China’s hottest new commodity: AI tokens — FT
- Chinese AI Models Challenge Western Pricing — Daily Brew
- US sees AI race with China as strategic battle for dominance — IBT International
- China vs US AI Investment in Africa: The 2026 Scorecard — Voxilens
- China AI industry looks unstoppable, but is it? | CNN Business
- 6 Graphs That Show Who’s Really Winning the US–China AI Race — TIME
- Surge in Chinese AI Market Share — Axios AI+
- Artificial intelligence: investment patterns and economic implications in leading countries — teme.junis.ni.ac.rs
- China vs. US in the AI Race: How China Is Closing the Gap - Medium
- US-China AI Gap: 2025 Analysis of Model Performance, Investment, and Innovation — Recorded Future
- Is China Beating the U.S. to AI Supremacy? | The Belfer Center for Science and International Affairs — The Belfer Center for Science and International Affairs
- The Complicated Stakes of the AI Race Between the U.S. and China — TIME
Related questions
- →What does AI sovereignty mean for Europe? Can you give examples of sovereign moves by different countries?
- →what is happening with Antrhopic and the pentagon?
- →describe the debate on AI sovereignty? What is it? What elements of AI would be subject to sovereignty?
- →What does meaningful AI development look like for countries without access to frontier compute — and what alternatives exist?