Who's Afraid of Chinese Models?
stratechery.com
July 20, 2026
19 min read
🔥🔥🔥🔥🔥
75/100
Summary
Chinese AI models are gaining attention for their rapid development and deployment in various applications. Concerns are rising over their potential impact on global AI competitiveness and ethical standards.
Key Takeaways
- Open weights models like Kimi K3 from China are approaching state-of-the-art capabilities, highlighting a shift in the AI industry towards marginal costs becoming significant again.
- The cost of goods sold (COGS) for AI models is directly correlated to revenue, making operational costs a critical factor in business models.
- Running inference on AI models incurs real costs, with Kimi K3 priced at $3 per million input tokens and $15 per million output tokens.
- Nvidia's approach to AI focuses on generating tokens efficiently, with metrics such as tokens-per-second and token cost becoming essential for decision-making.
Coverage & discussions
Hacker News · stratechery.com
Community discussions
Community Sentiment
Positives
- The idea that distillation could be considered fair use is gaining traction, as it levels the playing field for all AI developers — a win for innovation!
- Chinese models are pushing the boundaries by offering excellent open models for free, which could democratize access to powerful AI tools.
- The sticky nature of AI harnesses is being challenged, with many users finding it easy to switch between models, indicating a healthy competitive landscape.
Concerns
- Valuations of frontier labs like Anthropic and OpenAI are being questioned as Chinese models threaten to undercut their pricing strategies, raising concerns about their long-term viability.
- Critics argue that distillation can be seen as merely value extraction, which complicates the ethical landscape of AI model training.
- There’s skepticism about the stickiness of agent harnesses; some believe the real magic lies in the models themselves, suggesting that reliance on harnesses might be overhyped.
Related Articles
The arguments against open source AI are bad
Jul 23, 2026
China's open-weights AI strategy is winning
Jul 20, 2026
It's not a fear of "AI communism"; it's a fear of competitive market capitalism
Aug 4, 2026
Our position on open-weights models
Jul 27, 2026
The US is winning the AI race where it matters most: commercialization
May 13, 2026