Apple's new Mac Studio and Mac mini target local AI inference

Apple announced new Mac mini and Mac Studio desktops on Monday, alongside two new chips: the M6, the company's first 2nm chip for Mac, and a new Ultra-tier chip for the high end of the Mac Studio lineup, Ars Technica reported.

Neither machine gets major new features. This is a specs bump. But the way Apple is presenting the refresh makes the target audience clear: developers and researchers running AI models locally, a use case that barely existed when earlier versions of these machines were designed.

That shift traces back to macOS 26.2, which shipped last December. Apple's release notes described the update as enabling "low-latency communication between Thunderbolt 5 hosts for use cases including distributed AI inference using MLX," MLX being Apple's open source array framework built to help machine learning workloads take advantage of the M-series chips' unified memory. Since then, developers and researchers have been daisy-chaining multiple Mac minis or Mac Studios together to run inference on local large language models far bigger than any single mass-market device could handle on its own, an alternative to buying specialized Nvidia GPU hardware.

The base M6 packs a 12-core CPU split across two "super cores," four performance cores, and six efficiency cores, the first Apple chip to combine all three core types. Apple has not released verified benchmarks, but claims up to 40% faster multi-threaded CPU performance than the M4, two generations back. The GPU gets 12 cores, two more than its predecessor, and unified memory bandwidth rises to 160GB/s. Memory capacity on the base chip tops out at 32GB.

The Mac Studio's top-tier Ultra chip goes considerably further: 512GB of maximum memory, 36 CPU cores split between 12 super cores and 24 performance cores, 80 GPU cores, and up to 1.2TB/s of unified memory bandwidth. Apple builds it by joining two Max-tier chips on a single package.

The appeal comes down to cost and control. Developers have leaned harder on AI coding agents like Claude Code and Codex, but running frontier cloud models racks up token costs that some teams now question as a long-term budget line. Open-weight models such as recent Qwen and DeepSeek releases can handle many of the same coding tasks without per-token billing, since they run on hardware the developer already owns. The catch is that a standard MacBook Pro still falls well short of the memory needed to run the largest of those models, which is exactly the gap Apple is now selling hardware to close, an approach our earlier coverage of MacPaw's local AI stack for Mac showed other companies in the ecosystem are also building toward.

Both machines add Apple's N1 chip for Wi-Fi 7 and Bluetooth 6 support, and storage speeds up to 15GB/s, twice the previous generation. The Mac mini now ships with 2.5Gb Ethernet standard, with a 10Gb upgrade available.

• Mac mini with M6: starts at $899 with 16GB memory; M5 Pro configurations start at $1,699
• Mac Studio: starts at $2,499 with M5 Max; Ultra-tier configurations start at $5,499
• Top Ultra chip: 512GB max memory, 36 CPU cores, 80 GPU cores, up to 1.2TB/s bandwidth
• Base M6: 12-core CPU, 12-core GPU, up to 160GB/s bandwidth, 32GB memory cap
• Availability: preorders open today, shipping begins Sept. 22

Preorders open today, with shipping starting Sept. 22, though the 512GB memory configuration won't reach buyers until late October. Both machines ship running macOS 27 "Golden Gate."

This piece is informational, not a recommendation to buy, sell, or hold any asset.