Edge silicon runs roughly 30‑50× below datacenter memory bandwidth, and device‑RAM ceilings cap what models fit, so frontier inference stays in metro datacenters; the research’s verified framing is that latency, not cost, sets the split – real‑time inference needs low‑round‑trip placement, which means a datacenter closer to users, not a chip in your pocket. I keep a quarantine list of myths that testing has refuted, and industry reporting confirms the AI‑PC/AI‑phone supercycle is dead on arrival, serving only narrow niches.