Kimi K3, Qwen 3.8, and Anthropic's (Potential) Unravelling
Summary
The article surveys frontier AI labs and open-weight models (Kimi K3 and Qwen 3.8) and analyzes the economics of foundation-model development. It highlights cost drivers such as training vs inference, electricity, and data-center ownership, arguing that owning the infrastructure can turn variable costs into fixed costs and widen margins. It also assesses Anthropic's precarious position, OpenAI's potential moats, and the risk of an unbundled market as multiple labs compete.