A deep dive into Cerebras's thesis that inference latency and memory bandwidth will become increasingly important as AI moves toward reasoning models and agents. Covers the WSE-3 and CS-4 architecture, wafer-scale computing, their AWS/AMD/OpenAI partnerships, cloud transition, roadmap, economics, and the risks of scaling the business.