🔮🇨🇳 Inside the Chinese AI labs where America’s AI controls created its toughest competiton

2026-05-13 · Source: Exponential View · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation, International Business & Trade · Depth: Intermediate, medium

Summary

A recent week-long visit to 14 AI and robotics labs across Beijing, Hangzhou, and Shanghai, including DeepSeek, MoonshotAI, and ByteDance, revealed that China's AI development is significantly impacted by US chip export controls. While Chinese AI compute stock lags the US by two to three years, and capital raised by Chinese AI startups in 2025 was $12.4 billion compared to $285 billion in the US, Chinese open-source models are only six to eight months behind US frontier models in benchmark performance. This efficiency is attributed to Chinese labs extracting 4-7x more intelligence per unit of compute than naive scaling predictions suggest, forcing them to be ruthlessly efficient. Despite the compute gap, Chinese models like DeepSeek V4 Pro and Kimi K2.6 offer significantly lower inference costs, with DeepSeek charging $0.43 per million input tokens and $0.87 per million output tokens, compared to Claude Opus 4.6's 11x and 28x higher costs, respectively.

Key takeaway

For AI Engineers and product leaders evaluating model deployment strategies, recognize that Chinese AI labs, despite compute limitations, are demonstrating superior efficiency in model training and inference. You should investigate integrating highly efficient, lower-cost Chinese open-source models, such as DeepSeek V4 Pro or Kimi K2.6, into your applications to potentially reduce operational expenses and expand accessibility, especially for consumer-facing services where cost-per-token is critical.

Key insights

US chip export controls have inadvertently spurred Chinese AI labs to achieve remarkable compute efficiency, narrowing the performance gap.

Principles

• Constraints can drive efficiency gains.
• Compute efficiency is a critical competitive advantage.

In practice

• Explore Chinese open-source models like DeepSeek R1 7B for local inference.
• Evaluate cost-effective Chinese models for inference tasks.

Topics

• US AI Export Controls
• Chinese AI Labs
• AI Compute Gap
• Model Training Efficiency
• Large Language Models

Best for: AI Engineer, Investor, Entrepreneur, AI Scientist, Director of AI/ML, Policy Maker

Related on AIssential

• What’s next for Chinese open-source AI — Chinese open-source AI models offer near-frontier capabilities at a fraction of the cost, rapidly reshaping global AI innovation.
• ChinAI #349: Tokens Made in China? — Chinese AI models are gaining traction among U.S. developers due to their compelling price-to-performance ratio.
• How are Chinese models so strong with so little investment? — Chinese AI models achieve high performance through distillation from Western models and a robust national AI ecosystem.
• China’s DeepSeek previews new AI model a year after jolting US rivals — DeepSeek's V4 model aims to challenge top US AI systems, emphasizing coding prowess and domestic chip compatibility.
• AI in China and the United States — China is outpacing the US in AI development across talent, technology, and infrastructure due to strategic policy differences.
• The Sequence Radar #783: Softbank, DeepSeek, MiniMax and The Sequence 2026 — AI's future is bifurcated between capital-intensive infrastructure consolidation and efficiency-driven algorithmic innovation.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Exponential View.