Why Your Mac Is Slow at Local AI (And It's Not the Chip)

Local AI on a Mac feels slow for a reason the tokens-per-second counter never shows: your local LLM re-reads the whole prompt every turn. An M5 Max MacBook Pro with 128GB of unified memory spent 8 hours making 15 edits with a coding agent, and the chip was never the problem. Every "why is local AI slow on my Mac" video lands on memory bandwidth, and that part is real. A Mac mini M6 moves 170 GB/s, an M5 Max moves 614 GB/s, and the M5's neural accelerators made prompt reading about 3.6x faster than the M4 Max on llama.cpp's own benchmarks. But bandwidth only sets how fast the model writes. Bef