Apple Updates the Mac mini and Mac Studio for Local AI

Apple has refreshed two of its desktop lines — the Mac mini and the Mac Studio — and along with them introduced two new chips: the M6, the first 2-nanometer processor in the M-series for the Mac, and the M5 Ultra, now the most powerful chip in the family, especially for AI workloads. There are no revolutionary features here — this is essentially a spec bump. But the way Apple is pitching the new machines makes clear where the company is placing its bet: on local AI, something no one had in mind when the first versions of these machines were designed.

Betting on Local Inference

The popularity of desktop Macs for inference shot up after macOS 26.2 shipped this past December. The update added “low-latency communication between Thunderbolt 5 hosts” for distributed AI inference via MLX — the open framework that helps machine learning take advantage of the unified memory in M-series chips. Since then, enthusiasts and professional developers have literally been chaining Mac minis and Mac Studios together to run local models that wouldn't fit on a single mainstream computer. It has become an alternative to expensive specialized GPU hardware from Nvidia.

The Chips

The real story here is the processors themselves. The M6 gets a 12-core CPU that, for the first time, combines three types of cores at once: two “super cores,” four performance cores and six efficiency cores. Apple claims up to a 40% gain in multithreaded performance over the M4. The chip has a 12-core GPU and memory bandwidth of up to 160 GB/s, though capacity is capped at 32 GB.

This is where the M5 Ultra steps in for the Mac Studio: up to 512 GB of memory and up to 1.2 TB/s of bandwidth. It's essentially two M6 Max chips on a single die — 36 CPU cores and 80 GPU cores. Most people don't need anything like it, and that's the best confirmation of what the whole thing was about.

Why Developers Care

Many programmers are reworking their workflows around AI agents like Claude Code or Codex, but running cloud models is expensive, and the practicality of that approach is increasingly being questioned. Open models — the recent Qwen and DeepSeek, for example — can handle many of those same tasks locally, with no per-token fees. Granted, an ordinary laptop still isn't enough for large models — hence the interest in chains of desktop Macs.

The Mac mini with the M6 starts at $899, and the Mac Studio with the M5 Ultra at $5,499. Preorders are open, with shipping starting on September 22 (the 512 GB configuration comes in late October). The machines ship with macOS 27 Golden Gate.

Source: Ars Technica