Chinese Labs Release 3 Cheaper AI Models, Threatening $1 Trillion US Lab Profit Hopes
Updated
Updated · Vox.com · Jul 30
Chinese Labs Release 3 Cheaper AI Models, Threatening $1 Trillion US Lab Profit Hopes
3 articles · Updated · Vox.com · Jul 30
Summary
Three Chinese models released over two months — from Z.ai, Moonshot and Alibaba — are described as nearing or matching top US systems while costing far less, challenging assumptions that OpenAI and Anthropic can dominate frontier AI.
Kimi K3 and Qwen3.8 Max sharpen that threat because they are slated to be open source, letting companies run near-frontier models on their own hardware instead of paying US labs for access.
A key reason is distillation: Chinese firms are suspected of training on outputs from leading US models, a shortcut that can replicate advanced capabilities without the full post-training expense; Kimi K3 has at times identified itself as Claude.
That combination could turn frontier AI into a lower-margin business even if demand stays strong, shifting more value toward chipmakers and cloud providers than to the labs building the best models.
The broader tradeoff is that cheaper, widely distributed AI could spread benefits across businesses and users, while also making powerful systems harder to regulate and potentially easier to misuse.