Apple Mac Studio M5 Ultra hits $5,499 as memory shortage delays top configs to October
Apple has launched the new Mac Studio powered by the M5 Max and an all-new M5 Ultra chip, targeting professionals who need serious muscle for AI workloads, 3D rendering, and high-end video editing. The M5 Ultra starts at $5,499 — $200 more than the M3 Ultra it replaces — while the M5 Max entry model opens at $2,499. Preorders are open now, with shipping starting September 22, though the 512GB M5 Ultra configuration won't arrive until late October.
The chips
The M5 Max packs an 18-core CPU, up to a 40-core GPU, and up to 128GB of unified memory. Apple Newsroom claims AI performance up to 3.9× faster than the M4 Max — a meaningful jump for tasks like local image generation or on-device transcription.
The new Mac Studio with M5 Max and M5 Ultra chips.
The M5 Ultra doubles that up: a 36-core CPU, up to an 80-core GPU, and up to 512GB of unified memory with 1.2TB/s of memory bandwidth — 50% more than the previous generation. Graphics performance is up 80% versus the M3 Ultra, and the chip can handle 33 simultaneous streams of 8K ProRes 422 video. That memory ceiling is the headline for anyone running large language models (LLMs) locally without sending data to a cloud server.
Ports, connectivity, and the price climb
Both models get Thunderbolt 5 ports capable of up to 120GB/s, PCIe Gen 6 SSD storage, Wi-Fi 7, and Bluetooth 6. The machine supports four 5K 120Hz Studio Display outputs simultaneously.
Mac Studio features Thunderbolt 5, Wi-Fi 7, and support for four 5K displays.
The $200 price increase on the Ultra tier is tied to a global DRAM shortage, per 9to5Mac, which is also why the 512GB configuration ships late — a real problem for enterprises planning AI infrastructure deployments before year-end. Education pricing softens the blow slightly: $2,299 for M5 Max and $5,099 for M5 Ultra. Apple also offers a lease program at $48.99/month for the Max and $110.10/month for the Ultra.
Who actually needs this
The on-device AI case is straightforward: 512GB of fast, unified memory means you can run very large language models entirely on the machine, with no data leaving your desk. For studios, researchers, or any professional handling sensitive material, that's a practical privacy argument — not just a marketing one. The catch is that the model capable of doing it most effectively is the one you can't buy until October.
