Apple unveils M6 and M5 Ultra: 2nm chips and 512GB for AI

Apple introduced new Mac mini and Mac Studio models without prior notice on August 25, 2026 — and with them two chip debuts: the M6 is Apple's first processor built on a 2-nanometer process, and the M5 Ultra is the first M-series chip with a quad-die architecture. Both clearly target local AI workloads as the central selling point.

What's inside the M6? A redesigned 12-core CPU with two "super cores", a 12-core GPU with Neural Accelerators, and a dual 16-core Neural Engine. Apple quotes up to 1.2x faster multithreading than the M5, roughly 30 percent more GPU AI compute, and 170GB/s of memory bandwidth with up to 32GB unified memory. Silicon executive Sri Santhanam: "M6 combines a new CPU complex, two additional cores, a Dual 16-core Neural Engine, and more unified memory bandwidth."

The M5 Ultra in the new Mac Studio scales much further: up to 36 CPU cores, an up-to-80-core GPU, 1.2TB/s of memory bandwidth and up to 512GB unified memory. Apple puts peak GPU compute for AI at up to 4.5 times the M3 Ultra — enough to run even large open language models entirely on-device.

Where does pricing land? According to US reports, the Mac mini starts at $899 with the M6 and $1,699 in the M5 Pro configuration; the Mac Studio starts at $2,499. That moves serious local AI inference — until now the domain of expensive workstations — into compact desktop territory.

The context: Apple is increasingly positioning Macs as machines for privacy-friendly on-device AI while competitors bet on cloud inference. With 512GB of unified memory, the M5 Ultra explicitly targets developers and companies that want to run models in-house — a counterpoint to subscription AI services.