Apple’s M6 Mac mini and M5 Ultra Mac Studio target local AI

Apple has opened preorders ahead of its usual autumn Mac schedule. Apple has opened preorders for new Mac mini and Mac Studio computers, led by the M6 and M5 Ultra chips, with most configurations scheduled to ship on September 22, 2026. The move follows higher demand for machines running local AI agents such as OpenClaw, according to reporting from Cupertino.

The Mac mini changes little on the outside. Inside, the M6 brings a 12-core CPU, a 12-core GPU and Apple’s first chip built on a 2-nanometer process. Apple says the M6 Mac mini is up to 40 percent faster than the M4 in CPU performance, with graphics up to twice as fast and AI workloads up to four times faster. Its dual 16-core Neural Engine is dedicated to on-device AI tasks, and unified memory — shared by the processor, graphics and AI hardware — reaches 32GB with 170 GB/s of bandwidth.

The M5 Pro Mac mini is the more expandable option. It can be configured with up to 18 CPU cores, 20 GPU cores and 64GB of unified memory. Its three Thunderbolt 5 ports can link multiple Mac minis into a cluster, using remote direct memory access — a way for machines to exchange memory data with low overhead — for larger local AI workloads. Both models also add 2.5-gigabit Ethernet, Wi-Fi 7 and Bluetooth 6.

The Mac Studio is aimed higher. The M5 Max version starts at $2,499, while the M5 Ultra starts at $5,499. The Ultra chip uses Apple’s UltraFusion interconnect to combine two dual-die M5 Max chips and offers up to 36 CPU cores, 80 GPU cores and 512GB of unified memory. Apple reports up to 1.2 TB/s of memory bandwidth and up to 33 simultaneous 8K ProRes 422 streams at 30 frames per second.

So what changes in practice? Developers, creators and businesses can run larger models, image and video workloads, or persistent AI agents on the desktop rather than sending every task to a remote service. Several Mac Studios can also work together for models that exceed one machine’s memory. The limits are clear: these are Apple’s performance claims, and real results will depend on the model, precision, memory configuration and network speed; the highest-memory Mac Studio is not due until late October, with pricing still to be announced.