M5 Max tops M3 Ultra in AI

This is less a product launch than a signal that Apple’s laptop-class silicon is becoming a serious local AI platform, not just a premium notebook flex.

• –The article claims M5 Max beats M3 Ultra in most MLX-style AI tests that matter for local inference, especially where prefill and GPU throughput dominate
• –Apple’s split CPU/GPU tile approach is the most interesting takeaway because it suggests future Apple silicon gains may come as much from packaging and thermals as raw node shrinks
• –The biggest caveat is software maturity: the author explicitly notes cross-platform inference stacks and low-bit support are still uneven, so some comparisons are directional rather than final
• –For AI developers, the relevance is practical: better sustained thermals, sub-2W idle behavior, and strong unified-memory performance make high-end MacBooks more viable for serious local model work
• –If these results hold up beyond one analyst teardown, M5 Max strengthens Apple’s case against workstation-class desktops for a slice of local inference workloads