Also M-series Macs have an insane price/performance/electricity consumption ratio in LLM use-cases.

Any M-series Mac Mini can run a pretty good local model with usable speed. The high-end models easily compete with dedicated GPUs.