Overview
Cache-hit price drops to 0.006 dollars per million tokens. (Source: Tencent News, published 2026-09-14)
Context & Read-through
AI and compute are moving from the lab to the production line; the leap in model capability is triggering a chain reaction across hardware, chips and applications.
As training enters the 100k-GPU era, the focus shifts from ‘can we build it’ to ‘how low is the cost, how high is the efficiency’. The real dividing line for industry is whether model capability can be distilled into reusable business loops.
This column is an aggregation of industry information and technology viewpoints, and does not constitute any investment advice.