Apple pushed out three press releases on Tuesday with no event, no keynote, and no warning: a new Mac mini, a new Mac Studio, and the silicon underneath both. All of it is up for pre-order now and ships September 22, roughly two weeks before the iPhone lineup takes over the news cycle.
The theme running through every line of it is local AI. Not Apple Intelligence in the cloud sense. Models running on the desk, in the box, with the memory to hold them.
M6 Is Apple’s First 2nm Chip
M6 moves Apple silicon onto TSMC’s 2-nanometer process for the first time. It carries a 12-core CPU (2 super cores, 4 performance cores, 6 efficiency cores), a 12-core GPU with a Neural Accelerator in every core, and a Dual 16-core Neural Engine, which is new to the M-series. Apple claims the world’s fastest single-threaded CPU performance, up to 1.2x multithreaded over M5, and roughly 30 percent more peak GPU compute for AI.
Memory tops out at 32GB with 170GB/s of bandwidth. That is a 10 percent bandwidth bump over M5, which is modest next to the AI numbers and worth noticing given how tight DRAM supply has been all year.
The Dual Neural Engine is the interesting architectural bit. System frameworks can drive both engines at once, so applications get faster model execution without doing anything special.
M5 Ultra Goes Quad-Die
M5 Ultra is the headline chip. Apple used next-generation UltraFusion to bond two dual-die M5 Max chips into a four-die package, a first for Apple silicon. Inter-die bandwidth is over 4.4TB/s with more than 6x the connection density, and the four dies present themselves to software as one processor.
The result is up to a 36-core CPU (12 super cores, 24 performance cores), up to an 80-core GPU, and 512GB of unified memory at 1.2TB/s. That memory bandwidth is 50 percent higher than M3 Ultra. Peak AI compute is up to 4.3x M3 Ultra and 9.8x M1 Ultra.
Neural Accelerators reach the Ultra tier for the first time here. There is also a 32-core Neural Engine and a Media Engine with twice the encode and decode blocks of M5 Max, enough to play 33 simultaneous streams of 8K ProRes 422 at 30fps.
Mac Mini: Small Box, Agentic Framing
The new Mac mini ships with either M6 or the M5 Pro released earlier this year. Against the M4 model, Apple claims up to 4x faster AI performance, 2x faster graphics, 2x faster storage, and 40 percent faster CPU.
The M5 Pro configuration goes to an 18-core CPU, a 20-core GPU, and 64GB of unified memory at 307GB/s.
Connectivity gets a real upgrade on both. Wi-Fi 7, Bluetooth 6, and 2.5Gb Ethernet as standard, with a 10Gb option. Two front USB-C ports and a headphone jack with high-impedance support. On the back, three Thunderbolt 4 ports on M6 and three Thunderbolt 5 ports on M5 Pro, plus HDMI and Ethernet. Genlock over USB-C is new, syncing a display to a camera including iPhone 17 Pro.
Thunderbolt 5 on the M5 Pro model also allows clustering multiple Mac minis to run larger models entirely on device. Apple is openly pitching the mini as an always-on deskside agent box for businesses, which is a notable shift in how it talks about its cheapest desktop.
Cheapest is doing some work in that sentence. The M6 Mac mini starts at $899, up from the $599 entry point of the M4 generation.
Mac Studio: A Local Inference Machine
Mac Studio gets M5 Max and M5 Ultra. The M5 Max build has an 18-core CPU, up to a 40-core GPU running 50 percent faster than the prior generation, up to 128GB of unified memory, and 614GB/s of bandwidth. The M5 Ultra build scales to 36 CPU cores, 80 GPU cores, and 512GB of memory.
Storage moves to a PCIe Gen 6 SSD architecture at up to twice the previous read and write speeds, which matters mostly for how fast a large model loads. Thunderbolt 5 runs at 120Gb/s across up to six ports. The Apple-designed N1 chip brings Wi-Fi 7 and Bluetooth 6 to Mac Studio for the first time. Display support goes to eight screens, or four Studio Display XDR panels at 5K and 120Hz.
The clustering story is more serious here than on the mini. Thunderbolt 5 plus RDMA creates a shared memory pool across machines, and Apple says four clustered Mac Studios deliver up to 3x the inference performance of one. Combined with the new Core AI framework and MLX, Apple is making a direct pitch at teams currently renting cloud GPUs by the token.
Apple’s own framing is blunt about the economics: run frontier-class open-weight models locally, privately, without watching token counts or cloud bills climb.
Pricing and Availability
- Mac mini with M6: from $899 ($799 education)
- Mac mini with M5 Pro: from $1,699 ($1,599 education)
- Mac Studio with M5 Max: from $2,499 ($2,299 education)
- Mac Studio with M5 Ultra: from $5,499 ($5,099 education)
Pre-orders opened today in 30 countries. Everything ships September 22, except the 512GB Mac Studio configuration, which slips to late October. Both machines are built around macOS 27 Golden Gate, arriving as a free update this fall with Siri AI in beta.
What the Timing Says
Apple did not hold an event for a chip generation transition and its most powerful Mac ever. It dropped three releases on a Tuesday morning and moved on. The September iPhone keynote is close enough that the Mac would have been a footnote there, and Apple clearly decided the desktop story deserved its own daylight.
The other read is supply. A 512GB memory configuration arriving two months after everything else, and a $300 jump on the entry Mac mini, both point back to the DRAM and NAND shortage that has been squeezing the Mac line since spring. The silicon was ready. The memory to fill it, apparently, is still catching up.
Leave a Reply