The number I keep staring at is 512GB of unified memory at 1.2TB/s. That’s the line where running frontier-class open-weight models locally stops being a hobbyist stunt and becomes a boring infrastructure decision. Apple says it directly: run massive models on device, without counting tokens or worrying about rising cloud costs. That pitch is aimed squarely at anyone who has watched an inference bill climb.
What I’d do: nothing yet. Pre-orders open now, machines land September 22, and the honest test is whether an on-device coding agent feels usable, not whether it benchmarks 3.9x faster than M4 Max. The Thunderbolt 5 clustering with RDMA is the tell — Apple expects one box won’t be enough.
The story — Apple announced the new Mac Studio with M5 Max and the new M5 Ultra on August 25, 2026. M5 Max offers an 18-core CPU, up to 40-core GPU, and 128GB unified memory; M5 Ultra scales to 36 cores, an 80-core GPU, and 512GB at 1.2TB/s. Neural Accelerators in every GPU core drive up to 4.3x faster AI compute. Thunderbolt 5, Wi-Fi 7, Bluetooth 6. Pre-orders today, available September 22. (Source)